317 69 5MB
English Pages 414 [416] Year 2008
This page intentionally left blank
NEXT GENERATION MOBILE ACCESS TECHNOLOGIES Implementing TDD
Over the past 20 years, mobile communication systems have grown from analog voice to packet data centric 3G, with each new phase of development placing different requirements on the air-interface. The arrangement of two-way communication in future wireless systems is primarily influenced by the asymmetric nature of data traffic, coverage issues related to high data rate transmissions, and limited available spectrum below 10 GHz (paired spectrum in particular). Time division duplex (TDD) has a number of characteristics that best support the requirements of future packet data traffic with very high peak data rates. Provided some largely interference-related issues in TDD can be solved, TDD can considerably enhance system performance, regardless of the chosen multiple access technology. This book covers all aspects of TDD systems from cellular TDD-CDMA to multi-hop relaying and TDD-OFDM systems, and discusses in detail the challenges and opportunities of this technology in the context of current and future wireless networks. Starting with a basic introduction to wireless telecommunications using CDMA and TDD techniques, the book progress to more specialized topics including interference and capacity analyses, the centralized and de-centralized DCA algorithms, and concludes with limitations and future directions. The novel concept of time-multiplexed busy burst transmission applied to OFDM for interference aware subchannel allocation and medium access is discussed. The authors also describe how TDD technology can be used to overcome the physical limitations in conventional cellular systems that typically prevent the systems operator obtaining very high data rates at any possible location. This is a valuable resource for engineers involved in the design and implementation of wireless systems, as well as graduate students and researchers working in the area of wireless communications. Further information on this title is available online at www.cambridge.org/9780521826228. H a r a l d H a a s is a professor of Electronic Engineering at the International University of Bremen Germany, where his research interests are in wireless communications. S t e p h e n M c L a u g h l i n is Professor of Electronic Communication Systems for the Institute of Digital Communications at the University of Edinburgh, Scotland. His current research is principally on the applications and development of novel adaptive (linear and nonlinear) signal processing techniques.
NEXT GENERATION MOBILE ACCESS TECHNOLOGIES Implementing TDD HARALD HAAS International University Bremen
STEPHEN MCLAUGHLIN The University of Edinburgh
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo Cambridge University Press The Edinburgh Building, Cambridge CB2 8RU, UK Published in the United States of America by Cambridge University Press, New York www.cambridge.org Information on this title: www.cambridge.org/9780521826228 © Cambridge University Press 2007 This publication is in copyright. Subject to statutory exception and to the provision of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published in print format 2008
ISBN-13
978-0-511-50803-5
eBook (NetLibrary)
ISBN-13
978-0-521-82622-8
hardback
Cambridge University Press has no responsibility for the persistence or accuracy of urls for external or third-party internet websites referred to in this publication, and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.
Contents
List of abbreviations page viii Acknowledgements xi 1 Introduction 1 1.1 Introduction 1 1.2 The multi-user access 4 1.3 The cellular concept 6 1.4 Modes of channel operation 8 1.5 Objectives of the book 10 1.6 Structure of the book 12 2 Wireless telecommunications using CDMA and TDD tech15 niques 2.1 Introduction 15 2.2 Multiple access methods in wireless communications 17 2.3 TDD inherent properties 30 2.4 The TDD based air interface in UMTS 45 2.5 Radio resource allocation techniques 46 2.6 Summary 52 3 Interference and capacity analyses 54 3.1 Introduction 54 3.2 Capacity definition 55 3.3 Adjacent-channel interference in a CDMA-TDD system 60 3.4 Co-channel interference in a CDMA-TDD system 89 3.5 Conclusions 93 4 Centralised DCA algorithm using the TS-opposing idea 95 4.1 Introduction 95 4.2 TS-opposing technique applied to a single cell 95 4.3 TS-opposing technique in a multiple-cell environment 104 4.4 Conclusions 133 v
vi
5 5.1 5.2 5.3 5.4 5.5 5.6 5.7 6
Contents
Distributed DCA algorithm utilising the TS-opposing idea135 Introduction 135 Problem formulation 136 TS assignment plan 138 TS-opposing algorithm 143 System model 145 Results 150 Conclusions 155 UTRA-TDD Opportunity-Driven Multiple Access (ODMA) 157 6.1 Introduction 158 6.2 UTRA-TDD ODMA Background 160 6.3 Path loss investigation 168 6.4 Capacity Coverage Analysis 175 6.5 Conclusions 184 7 Routing strategies in multi-hop CDMA networks 186 7.1 Introduction 186 7.2 Multi-hop network architectures 187 7.3 Using path loss and interference-based metrics to route 190 7.4 Interference-based admission control 194 7.5 Congestion-based routing 196 7.6 Signalling overheads and latency 199 7.7 Results 200 7.8 Conclusions 212 8 Multi-hop DCA 214 8.1 DCA techniques 214 8.2 Combined routing and DCA algorithm 214 8.3 Results 223 8.4 Conclusions 225 9 Radio resource metric estimation 228 9.1 Radio resource metric estimation applied to radio resource allocation 229 9.2 Radio resource metric mapping function 231 9.3 Multirate transmission operations for a TDD-CDMA system 236 9.4 Radio resource metric region 255 9.5 Conclusions 267 10 Interference-based cancellation techniques for TDD 271 10.1 Introduction 271 10.2 Motivation 271 10.3 Performance analysis of linear precoding 273
Contents
10.4 Precoding techniques in classification 10.5 Power scaling factor 10.6 Joint transmission 10.7 Transmitter precoding 10.8 Decorrelating prefilters jointly optimised sequences 10.9 Pre-RAKE diversity 10.10 Complexity 10.11 Other techniques 10.12 Simulation results 11 Smart Antennas for TDD-CDMA Systems 11.1 Introduction 11.2 Channel modelling issues 11.3 Channel capacity issues: information-theoretical background of smart antennas 11.4 Uplink processing algorithms 11.5 Downlink processing algorithms 11.6 Future TDD wireless systems 11.7 Discussion and conclusions 12 Cellular OFDMA-TDD 12.1 Motivation and problems 12.2 Interference analysis 12.3 The busy-tone approach 12.4 Delay–Throughput Performance 12.5 Numerical Results 12.6 Busy-tone approach applied to cellular OFDMA/TDD 12.7 Conclusions 1 Derivation of T: Unconstrained Optimisation Bibliography Index
vii
275 279 281 283 286 289 294 296 297 300 300 302 308 311 319 325 333 336 336 337 345 350 358 363 375 377 378 401
List of abbreviations
3G 3GPP 4G AAD ACI ACIR ACLR ACS AWGN BER BLER BS CB CCI cdf CDMA COI COST CT DCA DCL DCS DECT DL DS ETSI
Third generation 3rd Generation Partnership Project Fourth generation Angle-of-arrival difference Adjacent-channel interference Adjacent-channel interference ratio Adjacent-channel leakage ratio Adjacent-channel selectivity Additive white Gaussian noise Bit-error rate Block error rate Base station Citizen band Co-channel interference Cumulative density function Code-division multiple access Cell of interest European Co-operation in the field of Science and Technical research Cordless telephony Dynamic channel assignment Degree of confidence level Dynamic channel selection Digitally enhanced cordless telecommunications Downlink Direct sequence European telecommunications standards institute viii
List of abbreviations
FCA FDD FDMA FH FM FRAMES GOS GSM HIPERLAN HSPA HSDPA HSUPA IEEE IMT-2000 IS-95 ITU kbps LAN MAI MAN MC-CDMA (MC)2 -CDMA Mcps MCL MoU MS MP MUD OCL ODMA OFDM OFDMA ORACH OVSF PCL pdf PHS PN
Fixed channel assignment Frequency-division duplex Frequency-division multiple access Frequency hopping Frequency modulation Future radio wideband multiple access system Grade of service Global System for Mobile Communications High performance radio local area network High speed packet access High speed downlink packet access High speed uplink packet access Institute of Electronic and Electrical Engineers International mobile telephony (third-generation networks are referred as IMT-2000 within ITU) Interim standard-95 International Telecommunications Union kilobit per second Local area network Multiple-access interference Metropolitan access network Multi-carrier CDMA Multi-code MC-CDMA Mega chips per second Minimum coupling loss Memorandum of understanding Mobile station Maximum packing Multiuser detection Optimal capacity limit Opportunity-driven multiple access Orthogonal frequency division multiplexing Orthogonal frequency division multiple access ODMA random access channel Orthogonal variable spreading factor Pessimistic capacity limit Probability density function Personal handyphone system Pseudo-noise
ix
x
List of abbreviations
PSTN QoS RCA RF RNC RRM RU RV Rx SDMA SF SDP TD-CDMA TDD TDMA TS TV Tx UMTS UTRA UTRA UDD UL VCE W-CDMA WiMAX WINNER
Public switched telephone network Quality of service Random channel assignment Radio frequency Radio network controller Radio resource management Resource unit Random variable Reception Space-division multiple access Spreading factor Semi-definite program Time-division CDMA (hybrid TDMA–CDMA interface) Time-division duplex Time-division multiple access Time slot Television Transmission Universal Mobile Telecommunications System UMTS terrestrial radio access (ETSI) Universal terrestrial radio access (3GPP) Unconstrained delay data Uplink Virtual centre of excellence in mobile and personal communications Wideband CDMA Worldwide interoperability for microwave access Wireless World Initiative New Radio
Acknowledgements
This book has been rather long in the gestation and many people deserve our thanks. Phil Meyler, our editor at Cambridge University Press, encouraged us to prepare the book and, along with his assistants Emily Yossarian, Anna Littlewood, Dawn Preston and Sabine Koch has shown considerable patience in dealing with two harassed academics. Our colleagues have looked on in amusement as we toiled in the writing and have provided advice and support as required in particular Professor Peter Grant, Professor Bernie Mulgrew, Dr Dave Laurenson and Professor Lajos Hanzo should all be mentioned in dispatches. We also would like to acknowledge the graduate and undergraduate students of the Cellular and Wireless Communications group at Jacobs University Bremen (Germany) who helped in revising and formatting the manuscript. The authors would also like to thank the Royal Academy of Engineering, who provided a Vodafone Fellowship to Harald Haas which enabled a series of visits to Edinburgh that aided in the completion of this book. We would like to thank our friends and collaborators for their contributions specifically the former and current researchers at the University of Edinburgh; Tom Rouse on chapters 6, 7 and 8; Yeonwoo Lee on Chapter 9; Stamatis Georgoulis and David Cruickshank on chapter 10; John Thompson and Ali Dakdouki on Chapter 11 as well as Gunther Auer, Project Manager at DoCoMo Euro-Labs Munich (Germany), and Peter Omiyi, postdoctoral fellow at Jacobs University Bremen (Germany), on Chapter 12. Finally this book is dedicated to our families who have suffered too much from our long absences in pursuit of our research goals.
xi
1 Introduction Harald Haas Stephen McLaughlin
1.1 Introduction Over the last 20 years personal computers have developed as a mass consumer product which, coupled with the growth in widespread broadband access, offers users a multiplicity of services. The Internet is an enormous source of information (both valid and invalid) that has evolved into a virtual store, a library, a chat-room and a playground, and has enabled novel means of interacting. The simple email service is an example of how material streams (transport of letters, documents, etc.) have been replaced by information streams. These issues are leading to fundamental changes in how we do things. A further significant step in the ‘digital revolution’ has been the demand for mobility, the so called anyone, anywhere, anytime mentality. Over the world digital mobile telephony has been a huge success. In particular, in Europe, the foundation was laid when 13 countries agreed to adopt a single digital standard for cellular mobile communications (The Memorandum of Understanding (MoU) signed on 7 September 1987 in Copenhagen). The development of the Global System for Mobile communications (GSM) followed, and the historic milestone of 1 billion mobile subscribers was reached in 2000 (Eylert, 2000). It is predicted that by the year 2010 there will be more than 1.7 billion terrestrial mobile subscribers worldwide. In addition, we have seen the wide deployment of WLAN (wireless local area network) technology, the deployment of W-CDMA (wideband code division multiple access) systems worldwide and their enhancement to HSPA (high speed packet access) offering increasing data rates, with peak downlink data rates of approximately 14 Mbps likely to be available in 2007. Other new systems that promise high data rates and large coverage have appeared such 1
2
Next Generation Mobile Access Technologies
as WiMAX (worldwide interoperability for microwave access) (Ghosh et al., 2005). It is widely recognised that the future challenge is to merge the dataoriented services prevalent on the Internet with wireless communication to fulfil the vision of anyone, anytime and anywhere (Mohr, 1998). Indeed this is already happening with mobile operators experimenting with offering TV services over their network, broadcasters offering communication and data services, and traditional fixed operators moving into the wireless access space. It could be argued that the first step into the area of wireless data applications was taken by the standardisation of the Universal Mobile Telecommunication System (UMTS)† in Europe and the equivalent WCDMA system in Japan. These systems aim to provide 144 kbps in vehicular environments, 384 kbps in outdoor to indoor environments and 2 Mbps in indoor or picocell environments (low mobility populations) (Mohr, 1998). The extension of these services to HSDPA (high speed downlink packet access), (initial deployments in 2006), and planned extensions to HSUPA (high speed uplink packet access) and the long term evaluation of 3GPP (3rd generation partnership project) will ultimately push peak data rates to 100 Mbps in the downlink. The pace is continuing with many researchers in academia and industry discussing requirements with respect to fourth generation wireless communication systems, some have already appeared in the literature, (see, e.g. (Mohr, 2002; Yamao et al., 2000)). Yet today the predominant service over existing digital wireless systems is speech, with its unique property that it requires a symmetric full duplex channel, and instant messaging, rather than true data services. The radio-frequency spectrum has become an expensive commodity. In the UK, for example, a sum equivalent to US$154 per citizen was paid for 2 × 15 MHz radio-frequency spectrum for UMTS (Sidenbladh, 2000). In order to use the radio-frequency spectrum efficiently and at the same time meet the requirements of future wireless communication systems, a high degree of flexibility is required (Luediger and Zeisberg, 2000). As a result, a single radio-channel access interface which is tailored to the needs of a specific service will not fulfil the extended requirements efficiently. Therefore, new radio interface concepts have to be investigated which, for example, is something that is done as part of the European WINNER (Wireless World Initiative New Radio) project (Acx et al., 2006; Klang, 2006). When it comes to achieving two way communication, two basic methods can be distinguished: frequency-division duplex (FDD) and time division† Sometimes also referred to as universal mobile telecommunications services (Holma and Toskala, 2000).
Introduction
3
duplex (TDD). The TDD technique has primarily been used for cordless, noncellular and hence short-range communication. Due to the unique properties of TDD, as, for example, the simplicity to arrange channel asymmetry, it was considered as a potential candidate for wireless packet data services (Povey et al., 1997). On the 29 January 1998 in Paris it was decided that the TDD technique in combination with TDMA (time division multiple access) and CDMA (code division multiple access) will be used in UMTS as part of a multi-mode air-interface. Little attention has been paid to this system during the initial deployment phase of UMTS, but it enjoys increasing interest for services such as mobile broadband and mobile TV (television). With an ever-increasing demand for packet-data services, such as Web browsing and media streaming, data traffic on the air interface will soon exceed the amount of voice traffic. While voice traffic causes a symmetric load on uplink and downlink, this clearly does not hold true for packet data traffic. In general, the latter requires variable channel asymmetry support with respect to uplink and downlink traffic, at least on an instantaneous basis. Methods such as HSDPA used in UMTS (UTRA-FDD) only partially solve this problem as this method, for example, does not support fast file uploads. Hence further system amendments such as HSUPA are currently being considered. In addition to its flexible support of traffic asymmetry, only the TDD mode offers an efficient and flexible support for ad-hoc or multi-hop communication. This is significant because a hybrid form of cellular and ad-hoc mode operation offers the most promising solution to cater for the high-data rates and the highly imbalanced traffic of future wireless systems. This stems from the fact that in cellular systems with high data-rate transmission, the cell range shrinks significantly, which, in turn, would require a large number of base stations to avoid coverage ‘dead zones’. This problem already exists in UMTS, and it will become more important with the envisaged high data rates of next-generation wireless systems. As a conclusion, an air interface for the next generation wireless systems is envisaged to employ the TDD mode. In fact, WLAN systems have successfully demonstrated the deployment of TDD in a hot-spot environment for packet data transmission. However, if the TDD mode is used in a cellular context some TDD-specific issues have to be addressed. Due to the fact that uplink and downlink transmission is at the same frequency band additional interference might occur, namely mobile-to-mobile and base-to-base interference, the extent of which being dependent on cell specific channel asymmetry and synchronisation. These additional interference sources may
4
Next Generation Mobile Access Technologies
cause severe capacity losses. But it has been shown that simple dynamic channel assignment (DCA) algorithms, (Haas et al., 2000a, 2002; Haas and McLaughlin, 2001a,b) can be found which avoid capacity losses - in fact, it has been demonstrated that in some cases capacity in a cellular TDD system can even be higher than in an equivalent FDD system
1.2 The multi-user access
Wireless Channel Unused radio resources user
user 3
1
user
2
user
4
Interference Fig. 1.1. The principles of multi user access.
The basic mechanism of the communication system that will be considered in much of this book is that a set of entities (users) access a common medium which, in this case, is the radio channel. This concept is depicted schematically in Figure 1.1. The frequency spectrum or bandwidth that is allocated to a certain system is a limited resource, which is indicated by the rectangular frame. In general, there are many coexisting wireless systems. In order to avoid interference to and from other systems a certain level of protection is required. This is indicated by the dark, shaded frame. The aim is to accommodate as many simultaneously active users as possible (ca-
Introduction
5
pacity) within the limited total radio resource where each user transmits the highest possible number of bits per unit time and per unit bandwidth. In the example, four users are considered each of whom requires an equivalent fraction of the total radio resource (illustrated by a circle), i.e. the bandwidth is the same for each user. In the picture, the size of the circles and hence the required radio capacity are constant. In real systems the size may be time variant (which may be visualised with breathing circles in Figure 1.1). Consider, for example, a speech service and periods when a speaker is silent. There is then no requirement to transmit data and thus the size of the circle would shrink to merely a single point in the space. An ideal multiple-access technique supports the time-variant radio capacity because this means that, at any given time, only those resources which are allocated are those which are actually required. Consequently, situations are avoided where more capacity is allocated than would actually be required. If the system is not designed carefully, users or mobile stations (MSs) will interfere with each other (the intersection of circles). Therefore, either the interference is eliminated via signal processing algorithms, if possible, or each user gets some extra protection which, in the illustration, is equivalent to moving users apart. This measure, however, results in unused radio resources which, considering the immense costs for the radio frequency spectrum, is inefficient. Therefore, the aim is to accommodate as many users as possible (minimising the area not covered with circles) while keeping the interference at a tolerable level. Interference in a cellular system can be categorised as: • Multiple access interference (MAI) is the interference that results from simultaneous transmissions in a multipoint-to-point or point-to-multipoint topology when orthogonality does not exist. • Adjacent channel interference (ACI) is the interference that results from simultaneous transmissions in adjacent frequency bands or channels that either belong to the same operator, or a different operator. • Intersymbol interference (ISI) is the interference between successively transmitted symbols due to multipath propagation. • Interchannel interference (ICI) is the interference on the same link between frequency channels that are simultaneously used to transmit parallel data streams on spatially separated multiple antennas • Co-channel interference (CCI) is the interference that results from the reuse of the same radio-frequency channels at spatially separated locations In general, different domains exist to achieve orthogonality, or quasi-
6
Next Generation Mobile Access Technologies
orthogonality, to allow multi-user access. In practice the following dimensions are used: • • • •
frequency ⇒ frequency-division multiple access (FDMA) time ⇒ time-division multiple access (TDMA) space ⇒ space-division multiple access (SDMA) code ⇒ code-division multiple access (CDMA)
The particular multi-user access technology results in different levels of interference in the different interference categories mentioned before, depending on numerous factors such carrier frequency, propagation channel, mobility and deployment scenario. It is non-trivial to find the optimum multiuser access technique for a given scenario. A further aspect in the design of a communication system is the spectral efficiency usually measured in bits per second, per Hertz, per square metre, [bits/second/Hz/m2 ]. Regardless of the actual multiple access technology, the same radio-frequency resource has to be reused as often as possible in order to increase the spectral efficiency, but that inherently increases CCI which, in turn, reduces the system capacity, and the classical capacity tradeoff in a cellular system is identified. The cellular concept first introduced by (MacDonald, 1979) resulted in a systematic approach to control CCI. Due to its importance it will be explained in more detail in the following section.
1.3 The cellular concept One of the key goals of mobile communication systems is to make services available anywhere and anytime which poses great challenges on the design of such systems. In particular, this requirement enforces the reuse of the limited radio-frequency spectrum. The reuse need to be organised such that CCI does not degrade the system performance below the guaranteed grade of service (GoS). In conventional wireless systems a mobile entity is linked to a base station (BS). Base stations are connected to a radio network controller which uses additional interfaces that enable access to the public switched telephone network (PSTN). The principle structure of a cellular wireless system is shown in Figure 1.2. The transmitted signals experience a distance-dependent attenuation due to the wireless channel. Since the transmit powers are limited, the coverage area of a BS is also limited. The area covered by a BS is referred to as a cell. When modelling cellular systems, cells are approximated
Introduction
7
Public switched telephone network (PSTN)
Radio network controller (RNC)
Fig. 1.2. A cellular wireless system.
by hexagons as they can be used to cover a plane without overlap (tessellation) and represent a good approximation of circles. Since the total radio resource available is limited, the spatial dimension is used to allow wide area coverage. This is achieved by splitting the radio resource into groups. These groups are then assigned to different contiguous cells. This pattern is repeated as often as necessary until the entire area is covered. A single pattern is equivalent to a cluster. Therefore, a radio resource that is split into i groups directly corresponds to a cell cluster of size i. In this way it is ensured that the same radio resource is used only in cells that are separated by a defined minimum distance. This mechanism is depicted in Figure 1.3 (different groups of radio resource units is indicated by different shades of grey.). As a consequence the separation distance grows if the cluster size increases. Hence, increasing the cluster size acts in favour of low interference. However, an increased cluster size means that the same radio resource is used less often within a given area. As a result fewer users per unit area can be served. Therefore, there is a tradeoff between cluster size and capacity. One problem with the cellular concept is that it allows very little flexibility, and it is, for example, not straightforward to simply add new BSs. Due to this inflexibility the system is very often designed for worst-case interference scenarios. In most of the cases, however, this results in an inefficient use of the radio-frequency spectrum especially in the case of packet data services where there is no continuous transmission or where the transmission gaps might be exploited by other users. Hence, in an ideal scenario the total available radio resource would be used in every cell whilst the interference was kept at a tolerable level. Herein lies a potential advantage of CDMA over all other multiple access modes since the same frequency
8
Next Generation Mobile Access Technologies
4 cell cluster
Fig. 1.3. The cellular concept.
carrier can be reused in every cell (Viterbi, 1995a) due to CDMA’s inherent interference resistance. However, it has been shown that interference avoidance and interference mitigation techniques in combination with dynamic channel allocation and scheduling allow the full frequency reuse for other multiple access technologies, too. The cell capacity, finally, depends on many system functions such as power control, handover, scheduling, link adaptation etc. In the interest of high spectral efficiencies, the overall aim for next generation cellular systems is to avoid a fixed frequency reuse.
1.4 Modes of channel operation There are three basic modes for operating a communication channel: simplex, half duplex and full duplex. The basic mechanisms are depicted in Figure 1.4. In the case of a simplex communication the information is User A
RX
User A
RX TX
User A
RX/TX
User B
TX
User B
TX RX
User B
TX/RX
Simplex
Half duplex
Fig. 1.4. Modes of channel operation.
Full duplex
Introduction
9
passed from one entity to another without permitting any acknowledgement (one-way communication). Notable examples are television and radio broadcasting. A half-duplex channel can send and receive, but not at the same time. This means one entity transmits at a time while the other entity listens, and vice versa. User A indicates when he/she wishes to terminate transmission giving the counterpart, user B in this case, the chance to talk. This leads to a ‘pingpong’ type of communication. This technique is used in talkback radio and CB (citizen band) radio where only one person can talk at a time. Note that access to the Internet merely requires a half duplex channel: consider user A sending a download request in principle, no further information needs to be transmitted and, thus, user A can go into the receive mode until all the required information is downloaded†. In a full-duplex channel, information travels in both directions simultaneously. Two entities can receive and transmit at the same time. Telephony is an eminent example from this category. In wireless communication systems two methods are used to achieve a full-duplex channel: time division duplex (TDD) and frequency division duplex (FDD). If the receive and transmit slots of a half-duplex channel are repeated periodically at short intervals, a full-duplex channel can be emulated by a half-duplex channel. This is exactly the mechanism used in TDD. In contrast, an FDD system separates both directions in the frequency domain so as to eliminate crosstalk. This means that the full-duplex channel is accomplished by two independent simplex channels. The basic mechanism of TDD and FDD are shown in Figure 1.5. In cellular communication the direction from the BS to the MS is referred to as the downlink‡. Similarly, the direction from the MS to the BS is the uplink. The advantage of FDD is that it represents a true full-duplex channel which does not need any coordination between uplink and downlink transmission. The disadvantage is that two separated channels have to be maintained. Given that many new services do not require a full-duplex channel (predominately data applications as illustrated by an Internet session), FDD offers more features than would be required. In the case of a file download, for instance, the uplink channel is underused or even unused, which results in the waste of expensive radio resources. In comparison, the TDD technique does not represent a true full-duplex channel. It requires coordination † In reality the protocols involved are more complex, but the basic principle remains the same. ‡ The terms ‘downlink’ and ‘forward link’ are synonymous. The terms ‘forward link’ is used primarily in the American literature. A similar dualism can be observed between ‘uplink’ and ‘reverse link’ and ‘handover’ and ‘handoff’, respectively.
10
Next Generation Mobile Access Technologies frequency
f3
downlink
FDD f2
uplink
f1
uplink
downlink
TDD time
Fig. 1.5. The principles of TDD as compared with FDD.
(synchronisation), but due to its nature, it ideally supports services that basically only require an asymmetric half duplex channel. Given that future wireless communication is evolving towards the wireless Internet, the significance of TDD will grow.
1.5 Objectives of the book One key objective of this book is to present an in-depth treatment of cellular TDD systems with a primary focus on TDD-CDMA systems, although consideration is also given to TDD-OFDM based systems. In this context, the book tries to answer the question whether the TDD technology can efficiently be used in cellular systems, and, if yes, under what conditions this can be achieved. It suggests algorithms that help to mitigate performance limitations due to TDD’s inherent properties, and it shows techniques and algorithms that boost the advantages of TDD in such deployment scenarios. The particular properties of TDD are considered to carry out interference and capacity analyses of cellular TDD-based systems. In this context, special emphasis is placed on the calculation of CCI and ACI for different cell layouts, power control algorithms, handover schemes and time slot (TS) synchronisation. As TS synchronisation is generally assumed to be critical to interference, particularly in a TDD system, the investigation of its impact on ACI and CCI is of utmost importance.
Introduction
11
TDD enables a high degree of flexibility. For instance, an FDD system can be considered as a system that is composed of two separate TDD systems each having no switching point and supporting only one link direction. This flexibility can be exploited for numerous hybrid duplexing solutions. One of these solutions, referred to as TDD underlay, is covered in this book. It will be demonstrated that dynamic channel allocation (DCA) algorithms are an essential component when using the TDD technology. These algorithms, for example, are necessary in order to enable a cell independent adjustment of channel asymmetry - one of the key requirements of next generation cellular systems. A number of such algorithms which make use of TDD inherent properties are proposed and discussed. Several of these algorithms are based on the concept of TS opposing, or TS hopping. The growing importance of meshed wireless networks, ad hoc and multihop communication for the support of high transmission rates in future mobile communication systems has a significant impact on TDD technology. For instance, ad hoc and multihop communication effectively is only feasible with TDD. In this book this enabling functionality of TDD is discussed and explained using ODMA (opportunity-driven multiple access) which was first proposed in the context of 3G systems. The requirements, limitations and advantages of ODMA are demonstrated. The special properties of TDD with respect to interference place a particular importance on the cross-layer design of such systems. It is aimed at utilising information that resides at the physical layer for resource allocation at higher layers. Therefore, resource metric estimation techniques tailored for a TDD-based system are treated in this book. The channel reciprocity in TDD also enables a new class of receiver architectures that use channel knowledge at the transmitter in order to pre-distort transmit data. Transmitter and receiver architectures as well as signal processing algorithms that exploit this effect are discussed in this book. Due to the growing demand for higher data rates and increased capacity in wireless systems consideration has been given to the deployment of smart antenna systems. The book considers the use of smart antenna approaches in TDD-CDMA systems; how the reciprocal nature of the TDD channel can be exploited to implement multi-user detection schemes where the complexity is predominantly in the transmitter; and how to manage the resources in a TDD-CDMA system where multiple users with several traffic types exist. All of the above are focused on cellular systems; consideration is also given as to how peer-to-peer systems would interact and operate within a cellular system by considering multi-hop relaying. TDD in combination with OFDMA (orthogonal frequency-division mul-
12
Next Generation Mobile Access Technologies
tiple access) for cellular systems is considered. The issues of medium access control (MAC) protocols and dynamic subchannel allocation algorithms for full frequency reuse cellular systems are covered. A joint MAC protocol and dynamic subchannel allocation algorithm based on time-multiplexed busy tones is presented. The busy tones allow the implementation of both interference avoidance and link adaptation and this makes them a powerful tool for dynamic subchannel allocation in OFDMA-TDD cellular systems. The reciprocal nature of the channel in TDD is a prerequisite to be able to use this tool.
1.6 Structure of the book This chapter has been a brief introduction into the principles of wireless communication. Chapter 2 discusses a CDMA system in more detail by comparing it with alternative multiple access methods. It also shows the problem that arise when using the total available bandwidth in every cell. The impact of two basic interference sources (inter-cell and intra-cell) on capacity are discussed. In addition, the properties of the TDD mode are presented where the focus is on the interference mechanisms in a cellular TDD system. The final part of Chapter 2 categorises DCA algorithms. The reason for this is that DCA algorithms are an appropriate means to avoid interference and thereby increase the capacity. Chapter 3 presents an in-depth interference analysis of a CDMA-TDD system. The analysis is carried out at the system level. Monte Carlo techniques, verified by an analytical approach, are used to calculate interference. This investigation reveals the interesting property that opposed transmission and reception may not necessarily yield capacity reductions in TDD systems. This property enables the use of TDD in achieving cell independent asymmetry in a cellular network without a significant capacity reduction. The resulting ‘TS-opposing technique’ is exploited extensively in the subsequent chapters, which address DCA algorithms in a cellular TD-CDMA/TDD network. Chapter 4 is dedicated to investigating a novel centralised DCA algorithm using the TS-opposing technique. The results corroborate the fact that TDD can be used to achieve cell independent asymmetry without a significant capacity loss. Chapter 5 applies the TS-opposing technique to a decentralised DCA algorithm using a novel time slot assignment plan. The results reveal that the capacity of a cellular TDD system can be greater than an equivalent FDD system.
Introduction
13
Chapters 6, 7 and 8 focus on the issue of multi-hop relaying in a TDDCDMA system. Chapter 6 examines UTRA-TDD ODMA. Initially an interference analysis is performed using a system-level simulation. This shows that in a single cell the greatest interference occurs close to the centre. This suggests that, if possible, it may be desirable to route around this area. A further investigation shows that the coverage capacity trade-off of a relaying system is more flexible than a conventional one. Chapter 7 develops several new routing algorithms. The algorithms follow from the findings of chapter 6 and from a desire to minimise the overall transmitted power and maximise system capacity. The algorithms reflect different requirements including complexity, speed to achieve routing and required overheads. It is shown that relaying requires a lower overall transmitted power than a conventional system, and when in-cell calls are involved relaying can increase system capacity. Chapter 8 extends the most successful routing algorithm, congestion based routing, into the time domain as a form of DCA. The results show that this integrated approach can further increase capacity gains over a non-relaying system. Chapter 9 introduces the notion of managing the resources in a cellular TDD-CDMA system with diverse traffic types. New methods (i.e. resource metric estimation) are introduced that (a) base the decision as to which radio resource to use, not only on interference, but also, for example, on the statistics of the channel state information, and (b) make the decision as to which radio resource to use at the physical layer. For (a) it is shown that the TDD mode is ideally suited due to the reciprocity of the channel. The performance of the new methods presented is assessed in terms of throughput and delays. This also includes reports on the interaction between new resource metrics and other link-adaptation techniques such as adaptive modulation/coding and power control. The objective of Chapter 10 is to consider techniques that remove the multi-user detection algorithms from the mobile terminal while maintaining satisfactory performance. The motivation for this is the chance for significantly reduced complexity of the receiver required in the mobile terminal. The chapter will initially discuss the motivation for this, the special conditions that make this idea realistic in a WCDMA-TDD system Chapter 11 provides an introduction to the topic of antenna array processing for use in the BSs of TDD-CDMA systems. Advantages and drawbacks are explained, and the essential and important channel modelling concept of smart antennas is introduced. In Chapter 12 TDD is combined with OFDMA. This technology is ap-
14
Next Generation Mobile Access Technologies
plied to a cellular and multi-hop system with full frequency reuse. As a consequence CCI is found to cause a significant problem which is mitigated by dynamic and adaptive dynamic subchannel allocation algorithms (DSA). These algorithms are designed so that they make use of the frequency selectivity in broadband OFDM transmission. Again, for this the channel reciprocity in TDD is exploited. The presented algorithms are based on the concepts of minislots and time-multiplexed busy tone transmission, which are also introduced in this chapter.
2 Wireless telecommunications using CDMA and TDD techniques Harald Haas
2.1 Introduction With the advent of third generation mobile communication systems TDD technology was applied for the first time to cellular systems. These systems primarily make use of CDMA (code-division multiple access) for managing multiuser access. Therefore, the scope of this chapter is to study the combination of CDMA and TDD technologies applied to cellular systems. The capacity of cellular systems that use CDMA in combination with frequency-division duplex (FDD) have been intensively studied (Chebaro and Godlewski, 1992; Wu et al., 1997; Corazza et al., 1998; Veeravalli and Sendonaris, 1999; Takeo et al., 2000) following the pioneering papers by (Gilhousen et al., 1991) and (Viterbi and Viterbi, 1994). Gilhousen and Viterbi applied spread-spectrum techniques, which, until then, had primarily been used in military applications, to commercial wireless communication systems — the foundation for which was laid by Shannon’s communication theory (Shannon, 1948). Due to its flexibility in supporting various transmission rates, CDMA had been widely adopted for third-generation mobile communication systems. During the European FRAMES (future radio wideband multiple access system) project in the mid nineties, different air interface proposals were investigated including a hybrid TD-CDMA/FDD system (delta concept), results of which are reported in (Ojanper¨ a et al., 1997), (Pehkonen et al., 1997), and (Klein et al., 1997). The time-division duplex (TDD) technique has been applied successfully to frequency-division multiple access (FDMA) systems and time divisionmultiple access (TDMA) systems (Esmailzadeh et al., 1997; Horikawa et al., 1997). Notable examples are the second-generation cordless telephony (CT) system CT2 (Tuttlebee, 1992), the digitally enhanced cordless telecommunications (DECT) system and the personal handyphone system (PHS) (Zeng 15
16
Next Generation Mobile Access Technologies
et al., 1999). Further notable examples of hot-spot systems that use TDD are the family of wireless local area network (LAN) standards that are based on IEEE 802.11. These short-range wireless communications standards are primarily designed for high data rate applications (up to 54 Mbps), but require low mobility populations. Another TDD representative which falls into the same category, but with even lower transmission ranges, is ‘Bluetooth’, which is used for cordless data communication between electronic devices (Garg et al., 2000). Moreover, in January 1998 ETSI (European Telecommunications Standards Institute) decided to specify two air interfaces for the European candidate of the third-generation mobile telecommunications systems, UMTS (universal mobile telecommunications system). The UMTS terrestrial radio access (UTRA) is divided into FDD and TDD modes (Ojanper¨ a and Prasad, 1998). The UTRA-TDD mode employs a hybrid form of CDMA and TDMA (Miya et al., 1998) which is referred to as TD-CDMA (Haardt et al., 2000). The UTRA-TDD interface is closely related to the ’delta concept’, but uses TDD instead of FDD. In China, the TD-SCDMA (time division - synchronous code division multiple access) which is based on a hybrid solution of CDMA and SDMA (space division multiple access) in combination with TDD is considered a very promising candidate for third-generation cellular networks. In addition, during the last five years metropolitan area network (MAN) systems based on the IEEE 802.16 standard have evolved and have gained considerable momentum with variants that even support some level of user mobility. The most popular representative of these systems is WiMAX (Worldwide Interoperability for Microwave Access). The use of TDD in all these systems has a significant impact on the system capacity, which has not, as yet, been investigated widely. This chapter should shed some light on these issues. UMTS belongs to the IMT 2000† family and is designed to be capable of offering new services, including multimedia and mobile Internet. A common property of these packet-data oriented services is that they often result in an unbalanced load in the uplink and downlink directions. This requires the support of an asymmetrical use of a communication channel. The TDD mode can easily arrange channel asymmetry, but is subject to more interference mechanisms if it is used in a cellular environment (Povey et al., 1997). A further unique property of TDD is that it enables the use of the opportunity-driven multiple access (ODMA) technique, which is based on intelligent relaying (Harold and Nix, 2000b). ODMA is based on localised and † International mobile telephony (IMT), third-generation networks are referred to as IMT 2000 within ITU (International Telecommunications Union).
Wireless telecommunications using CDMA and TDD techniques
17
self-organising network structures which are envisaged to form a key component of future mobile communication systems (Michail and Ephremides, 2000; Prehofer and Bettstetter, 2005). ODMA will be treated more fully in chapters 6, 7 and 8. Some general pioneering work on the capacity of ad hoc networks is done by (Gupta and Kumar, 2000). This chapter sets out the background of TDD and CDMA techniques. It contains a brief description of different multiple access modes in section 2.2 where a particular emphasis is put on CDMA techniques. An overview of the properties inherent to the TDD method can be found in section 2.3. In section 2.4 the basic radio interface properties of UTRA-TDD are described. In section 2.5 channel assignment issues are discussed. The chapter is summarised in section 2.6. 2.2 Multiple access methods in wireless communications The basic problem in connection with multiple access in wireless communications is to divide a finite radio resource in such a way that portions of this resource can be assigned to a number of independent users without creating significant mutual interference. This requires a set of orthonormal functions (Jung et al., 1993). Orthonormality can be arranged in four basic dimensions: (a) frequency dimension, (b) time dimension, (c) space dimension and (d) code (power) dimension. These correspond to (a) FDMA, (b) TDMA, (c) space-division multiple access (SDMA) and (d) CDMA. In the following, these multiple access methods are discussed in brief. 2.2.1 Cellular FDMA systems In FDMA systems the radio frequency spectrum is divided into several frequency bands separated by a certain guard band. Each frequency band can be used simultaneously. The guard band is required to reduce interference resulting from adjacent channel power leakage due to receiver and transmitter imperfections (crosstalk or adjacent channel interference (ACI)). Each frequency band is regarded as a physical channel assigned to a single user. When ACI is neglected and only a single cell is considered, all users are separated in an orthogonal fashion. In the case of a cellular FDMA system, however, the same frequency is reused in some other cell separated by a certain minimum distance D†. This mechanism is illustrated in Figure 2.1. The spatial separation of cells that use the same frequency (f1 in the ex† Since in FDMA systems a single frequency represents a channel, in this context D is described as the frequency reuse distance. This concept can also be applied to other multiple access techniques, which is why D can be generalised as channel reuse distance
18
Next Generation Mobile Access Technologies
D
f1
R
f1
Fig. 2.1. Frequency reuse distance in cellular FDMA and TDMA systems.
ample) reduces co-channel interference (CCI) but diminishes the spectral efficiency (Prasad and Kegel, 1991). The factor qs = D/R is defined as the CCI reduction factor (Lee, 1989b), where R represents the cell radius. The relationship between the number of cells in a frequency reuse pattern (cluster size) K and qs is found in (Lee, 1997, page 516) and yields: √ (2.1) qs = 3K. Moreover, in (Lee, 1989b) the relationship between the required carrier to interference ratio γ u at the BS (indicated by the superscript ‘u’ for uplink†) and qs is found for the scenario of six interfering cells at distance D where the MSs of the interfering cells are located at the centre whereas the MS in the desired cell is located at the cell boundary. This results in qs =
D = 6 (γ u )ϕ , R
(2.2)
where ϕ is the path loss exponent. The number of channels per cell is: M=
Bt , Bc K
(2.3)
where Bt is the available bandwidth and Bc is the channel bandwidth. Substituting (2.1) and (2.2) into (2.3) and assuming that the propagation power † Note that symbols which are followed by the superscript ‘u’ are associated with the uplink channel; symbols which are followed by the superscript ‘d’ are associated with the downlink channel. This applies throughout this book.
Wireless telecommunications using CDMA and TDD techniques
19
loss increases according to the fourth power of the distance, i.e. ϕ =4, then the system capacity can be denoted as follows (Lee, 1989b): M= Bc
B t 2 3
number of channels/cell .
(2.4)
(γ u )
Given that the theoretical maximum number of channels in a single cell environment is M = Bt /Bc , it can be seen from (2.4) that the capacity of a cellular system is reduced by a factor of (2/3) (γ u ) as a consequence of the frequency reuse. Note that the capacity reduction decreases as γ u diminishes. 2.2.2 Cellular TDMA systems In TDMA systems the entire bandwidth is used by each MS. The orthogonality between users is achieved in the time domain by dividing the time scale into timeslots (TSs) which are periodically allocated to each MS for the duration of a call. Guard times between TSs are required in order to prevent symbol collisions. These collisions can occur due to signal propagation time differences. These guard times, as in guard bands, result in wasted radio resources. A sequence of TSs forms a communication channel. The transmitter is silent between the consecutive TSs of a certain sequence. This results in a bursty transmission of data and a time compression of the information which is to be sent. The TDMA technique requires precise synchronisation between the communicating entities. Therefore it is more complex than the FDMA technique. It is found in (Lee, 1989b) that the capacity of cellular TDMA and FDMA systems is the same. Therefore, (2.4) can also be applied to TDMA systems. The basic difference is that the transmitted powers in a TDMA system are greater than in an FDMA system. If a TDMA interface consists of n channels, then the transmitted powers are 10 log n times higher than in an FDMA system (Lee, 1989b). The capacity of a TDMA systems is investigated in (Jovanovi´c and Gazzola, 1997). 2.2.3 Cellular SDMA systems In SDMA systems the multiple access is achieved by a spatial separation of the transmitted signals. The key principle of SDMA is the use of adaptive antenna arrays as a mechanism to form multiple independent beams per resource. Hence, two or more users in the same cell can share a common resource simultaneously (Galvan-Tejada and Gardiner, 2001). Techniques
20
Next Generation Mobile Access Technologies
such as antenna sectorisation, fixed beams or antenna beamforming can be applied (Thompson et al., 1996; Paulraj and Papadias, 1997; Paulraj and Ng, 1998; Winters, 1998). These techniques enable the transmitter to deliver the required signal power to the desired users and at the same time reduce the interference from other links. As a counterpart to the beamforming algorithm at the transmitter, a spatial filtering algorithm is required at the receiver in order to eliminate residual interference from other users. In particular, the spatial filtering requires knowledge of the spatial covariance matrix which contains essential spatial channel parameters. The following parameters, for example, form an essential input to the spatial filtering: (a) number of dominant propagation paths, (b) the direction of arrival of all dominant propagation paths and (c) the attenuation of each path. The receiver, in turn, uses the information contained in the spatial covariance matrix for its beamforming. The spatial covariance matrix can only be used directly if the channel guarantees a certain grade of reciprocity, however, which normally is the case when using the TDD method. Therefore, TDD is the preferred duplex method in combination with SDMA†. The respective models, architectures, techniques and algorithms are discussed in detail in Chapter 11.
2.2.4 Cellular CDMA systems CDMA differs from TDMA and FDMA in that it permits multiple access at the same frequency carrier and at the same time. The user separation is achieved in the power or code domain because all coexisting users appear as noise by utilising spread-spectrum techniques. These techniques establish the foundation for CDMA. Therefore, a brief summary of spread spectrum communication is presented in the following. For over half a century spread-spectrum techniques had been used in military communication systems before they were considered for commercial wireless applications (Cooper and McGillem, 1986; Pickholtz et al., 1991; Omura and Yang, 1992; Kchao and St¨ uber, 1993; Goiser, 1998). In a spreadspectrum system, the frequency bandwidth is greater than the minimum bandwidth required to transmit the desired information. There are different methods as to how the spreading of the spectrum can be accomplished: Direct-sequence (DS) spread-spectrum: A signal with a certain information bit rate is modulated on a frequency carrier with a much higher bandwidth than would be required to transmit the informa† The FDD mode would require a frequency transformation of the spatial covariance matrix.
Wireless telecommunications using CDMA and TDD techniques
21
tion signal. Each user is assigned a unique code sequence† which has the property that the individual users information can be retrieved after despreading. Frequency-hopping (FH) spread-spectrum: The available channel bandwidth is subdivided into a large number of contiguous frequency slots. The transmitted signal occupies one or more of the available frequency slots which are chosen according to a pseudo-random sequence. Time-hopping spread-spectrum: A time interval which is much larger than the reciprocal of the information bit rate is subdivided into a large number of TSs. The information symbols are transmitted in a pseudo-randomly selected TS. Chirp or pulse-FM modulation system: The frequency carrier is swept over a wide band during a given pulse interval. It is common in all spread-spectrum techniques that the available bandwidth, B, is much greater than the bandwidth required to transmit a signal with an information data rate, W . The ratio B/W is the bandwidth spreading factor or processing gain, pg. The processing gain results in interference suppression, which makes spread-spectrum systems highly resistant to interference or jamming. This property in particular makes spread spectrum techniques interesting for application to wireless multiple access communication where a large number of uncoordinated users in the same geographical area access a radio-frequency resource of limited bandwidth. Using the spread-spectrum technique, the number of simultaneously active users permitted is proportional to the processing gain (TIA/EIA/IS–95, 1995). Since the early 1980s, this has led to the development of the CDMA technology which primarily utilises the pseudo-noise (PN) DS spread-spectrum technique (Cooper and Nettleton, 1978; Baier and Koch, 1991; Schilling, 1991). Apart from the PN direct sequencing, a second category of CDMA techniques exists. This is described as orthogonal DS-CDMA (Verdu, 1998, Chapter 1). The most significant differences between orthogonal DS-CDMA and PN DS-CDMA are highlighted below: Orthogonal DS-CDMA systems: Each data symbol is spread by an orthogonal code sequence. Notable examples of orthogonal spreading codes are Walsh codes and OVSF (orthogonal variable spreading factor) codes (Adachi et al., 1997). The number of users is upper bounded by the time-bandwidth product rather than by multiple † In some literature code sequences are also referred to as ’signature waveforms’.
22
Next Generation Mobile Access Technologies
access interference. This technique belongs to the same class of orthogonal systems as TDMA and FDMA systems. The disadvantage of the aforementioned orthogonal codes is that they do not fulfil the pseudo-noise properties. Therefore, the performance of such a system is not robust against non-synchronous data transmissions and multipath propagation. In addition, synchronisation is often difficult to achieve because the size of the off-peak value relative to the peak value of the autocorrelation function is high. PN DS-CDMA systems: These systems use spreading codes which are not orthogonal, but have (almost) ideal properties with respect to the autocorrelation and the mutual cross-correlation functions (Cooper and McGillem, 1986; Viterbi, 1995a; Goiser, 1998). The properties of the PN code sequences significantly determine the performance of the spread-spectrum system. Therefore, several code families have been developed in recent years, the most important of which are Gold codes, Kasami codes and random codes. The main advantages over orthogonal codes are as follows: • The users can be asynchronous. This means that the bit transmissions need not be aligned as, for example, in the uplink direction. Despite the asynchronous bit-overlaps the spread-spectrum signals are still ‘quasi-orthogonal’. • The number of users is no longer constrained by the time-bandwidth product of the code sequences (soft-capacity), but is primarily interference limited. • The channel resources are shared dynamically. The reliability depends on the number of simultaneously active users rather than on the usually much larger number of potential users. This means that it is possible to swap the quality of service (QoS) for an increased capacity. As a consequence, the calculation of the system capacity becomes more complex. In this book, PN DS-CDMA systems are considered because orthogonal CDMA systems would require an ideal channel. In addition, CDMA based standards use, at least, a PN code for the final scrambling of the transmitted data. The wireless communication standards that utilise CDMA techniques, for example IS–95 and UMTS, use a combination of orthogonal codes and PN codes (Rappaport, 1996; Holma and Toskala, 2000), but this is merely aimed at increasing the robustness of the system. In recent papers, however, the concept of channel overloading is introduced. By using the channel
Wireless telecommunications using CDMA and TDD techniques
23
overloading technique the overall system capacity can be increased (Sari et al., 2000a,b). In one application, orthogonal DS-CDMA is overloaded with PN DS-CDMA techniques. It is reported in (Sari et al., 2000a) that this method can increase the single cell capacity by about 40%. Since in this book PN DS-CDMA techniques are considered, henceforth the expression CDMA will be used to describe this particular multiple access method. As mentioned above, the capacity calculation of a CDMA system is more complex since it is interference limited. Each user contributes to the common noise floor, which is usually assumed to be Gaussian (Gilhousen et al., 1991). Thus, interference is a most important parameter in a CDMA system and capacity analyses focus on calculating interference quantities (Gilhousen et al., 1991). Since interference is dependent on many factors, for example power control, adjacent channel leakage and handover strategies to name only a few, the capacity figures can vary significantly (soft-capacity). The capacity in the uplink direction of CDMA systems has been investigated by many researchers (Gilhousen et al., 1991; Liberti and Rappaport, 1994; Hashem and Sousa, 1997; Kim et al., 1997; Tam and Lau, 1997; Wu et al., 1997; Jeon et al., 1998). CDMA is used in the second-generation mobile communication standard IS-95 (TIA/EIA/IS–95, 1995) which gained special interest after it was claimed that CDMA can achieve a greater spectral efficiency than conventional FDMA and TDMA methods (Gilhousen et al., 1991). In (Viterbi, 1991), for example, Viterbi claims that the capacity of a CDMA system can be: Capacity (CDMA) ≈ 1 bit/sector/Hz/cell , assuming the voice activity of each user to be 50% and the sectorisation gain to be 4–6 dB. This figure was compared to the capacity of GSM (Global System for Mobile Communications) Capacity (GSM) ≈ 1/10 bit/sector/Hz/cell , where a frequency reuse factor of 1/4, i.e. qs =3.46, was assumed. Theoretically, when considering a single cell and an AWGN (additive white Gaussian noise) channel the multiple access schemes CDMA, FDMA and TDMA are equivalent with respect to spectral efficiency (Jung et al., 1993; Proakis and Salehi, 1994). Therefore, the greater spectral efficiency of CDMA systems primarily results from three basic principles: (i) the same channel is used in every cell (a channel reuse factor of 1) (Viterbi, 1995a);
24
Next Generation Mobile Access Technologies
(ii) interruptions in transmission are exploited, e.g. quiet periods of a speaker, when assuming a voice service (Gilhousen et al., 1991); (iii) antenna sectorisation. Apart from the above methods there are further techniques such as macro diversity (Jung et al., 1993) and soft handovers (Viterbi et al., 1994; Wong and Lim, 1997) which are exploited in CDMA systems to enhance the spectral efficiency. However, it was demonstrated that the advantages of CDMA systems were slightly overestimated (Chebaro and Godlewski, 1992, 1996; Corazza et al., 1998) due to two basic hypotheses which usually cannot be fulfilled in a realistic environment: (i) perfect power control; (ii) all MSs are allocated to the most favourable BS, i.e. the BS offering the lowest path loss. In further analyses, it became obvious that the requirements on power control in CDMA systems is a critical issue (Kudoh, 1993; Prasad et al., 1993; Jalali and Mermelstein, 1994; Mohr and Kottkamp, 1996; Kikuchi et al., 2000) as otherwise the capacity can suffer significantly. Furthermore, in (Chebaro and Godlewski, 1992) it was demonstrated that the allocation of a MS to the nearest BS rather than the BS offering the lowest path loss can increase interference by a factor of 4. This requires special handover techniques as otherwise capacity can suffer significantly. 2.2.4.1 The uplink in a CDMA system Ideally, in a CDMA system all coexisting users appear as Gaussian noise. Therefore, the required carrier to interference ratio, γ u , when assuming a multiple cell environment can be denoted as follows: γu =
Puui M i=j
ρj Puuj
own-cell interference
+
(2.5) Iou
other-cell interference
+
N
noise power
where Puui is the signal power of the desired user in the uplink; Puuj is the interference power received from a MS using the same channel in the same cell; ρj is the voice/data activity factor (usually modelled as independent Bernoulli random variables (Veeravalli and Sendonaris, 1999)) of the jth user; Iou is the interference power from other cells; N is the thermal noise
Wireless telecommunications using CDMA and TDD techniques
25
power and M is the number of MSs per cell. It can be seen that the interference is composed of three parts: own-cell interference, other-cell interference and noise power.
Own-cell interference This is equivalent to multiple access interference (MAI) due to the cross correlation of the spread-spectrum signals in a CDMA system. Intensive research on receiver structures is carried out to eliminate or reduce MAI by multi-user (H¨ am¨al¨ ainen et al., 1996; Moshavi, 1996; Verdu, 1998; Hassell-Sweatman et al., 2000; Verdu, 2000) or joint– detectors (Klein et al., 1995; Jung and Blanz, 1995). The problem of these receivers is that the complexity increases with the length of the spreading codes.
Other-cell interference Other-cell interference can be divided into CCI and ACI conveyed by neighbouring or coexisting cells in a cellular environment. In (Viterbi and Viterbi, 1994) the power ratio of co-channel interference to the desired signal power is calculated. The results of this paper show that co-channel interference can be 6.23M times higher than the desired signal power if a MS is assigned to the closest BS. This figure varies considerably if the MS is allocated to a BS which offers the lowest path loss. If, for example, the MS can choose the best out of three closest BSs the inference ratio decreases to 0.74M . Since own-cell interference is also proportional to M , the ratio of other-cell interference to own-cell interference is independent of the number of users and varies between 6.23 and 0.74 for the observed case. Therefore, perfect own-cell interference cancellation may, in the worst case, only reduce the total interference by ≤ 1/ (6.23 + 1) = 13.8% (equality applies if the thermal noise power is zero). For comparison the interference reduction due to perfect own–cell interference cancellation in the case of assigning the MS to the best BS (best out of three BSs) yields ≤ 1/ (0.75 + 1) = 57.1%. Given that the frequency reuse of a cellular CDMA system is generally considered to be one (Viterbi, 1995a), it is obvious that other cell interference can significantly reduce the advantages obtained by multiuser detectors. This mechanism was put into a more general context using the Shannon capacity equation (Viterbi, 1995b; Lee, 1997). (Viterbi, 1995b) compared the capacity of a multi-cell system without interference cancellation with the capacity of a multiple-cell system assuming perfect own-cell interference cancellation. In the latter case, the total carrier-to-
26
Next Generation Mobile Access Technologies
noise ratio is: M
Ci C = , N0 N0
(2.6)
i=1
where N0 is the thermal noise density power and Ci is the carrier power of the ith user. The Shannon’s channel capacity for the AWGN (additive white Gaussian noise) channel is:
C (2.7) W < B log2 1 + N0 B
where B is the total channel bandwidth and W is the total bit-rate calculated as W = M i=1 Wi with Wi being the bit-rate of a single user. If the bit-rate of each user is the same it holds that: ˜ Wi W =M
(2.8)
˜ is the number of simultaneously active users when assuming ideal where M interference cancellation. Substituting (2.8) into (2.7) and rearranging yields:
˜ M C (2.9) < log2 1 + pg N0 B where pg is the processing gain defined as pg = B/Wi . It is shown in (Viterbi and Viterbi, 1994) that in the case when all cells are equally loaded, and all BSs employ power control on their populations of MSs, the other cell interference is proportional to C. Furthermore, the assumption that the interference from other cells is Gaussian is justified by the central limit theorem. Thus, the interference power from the neighbouring cells can be written as follows: Iou = N0 B = f C
(2.10)
where f is the proportionality factor described above. The relationship in (2.10) is substituted into (2.9) which yields:
˜ M 1 (2.11) < log2 1 + pg f In order to assess the impact of interference cancellation for a cellular CDMA system in the following the performance of ideally coded users without interference cancellation is considered. Perfect power control is assumed, i.e. the power received from each MS within a cell is the same (Puu = Puu1 = Puu2 =, · · · , PuuM ). Hence, the own-cell interference can be expressed as follows: Iown = (M − 1)Puu ≈ M Puu = C
(2.12)
Wireless telecommunications using CDMA and TDD techniques
27
where M is the number of simultaneously active MSs in the multi-cell environment without interference cancellation. Using (2.10) and (2.12), and assuming N to be negligible due to the interference limited nature of the system, (2.5) becomes: γu =
Puu M (Iown +
Iou )
=
1 Eb /N0 = (1 + f ) pg
(2.13)
The bit-energy to interference ratio is expressed by Eb /N0 . From (2.13) the number of users in the multi cell environment can be found as follows: pg (2.14) M= (Eb /N0 )(1 + f ) In a AWGN channel Eb /N0 has a lower bound (Shannon, 1948) as follows: Eb /N0 > ln 2
(2.15)
This bound equals Shannon capacity for a channel with infinitely wide bandwidth. Applying (2.15) to (2.14), the number of users in a multi-cell environment divided by the processing gain has an upper bound denoted by: M 1 < pg (1 + f ) ln 2
(2.16)
Let G be the ratio of the capacity in the case of perfect interference cancellation, eqn. (2.11), to the capacity in the case in which no interference cancellation is used, eqn. (2.16), then
˜ M 1 (2.17) = (1 + f ) ln 1 + G= M f The relative number of users with ideal interference cancellation, eqn. (2.11) and without interference cancellation, eqn. (2.16), and the respective capacity gain, eqn. (2.17), are depicted in Figure 2.2. From the results in Figure 2.2 two important conclusions of a DS-CDMA system can be deduced: (i) The capacity of a cellular CDMA system can be greater than the processing gain. In contrast, the capacity of a FDMA or TDMA system is always less than or equal to the processing gain which can directly be seen from the system capacity equation of those systems, eqn. (2.4). (ii) As the other cell interference increases the total capacity diminishes and the gain due to multiuser detection decreases significantly. With a ratio of other-cell interference to own-cell interference of f = 0.6 the ideal interference cancellation merely increases the total capacity
28
Next Generation Mobile Access Technologies
Fig. 2.2. Theoretical capacity limits of a DS-CDMA multiple cell system with and without interference cancellation.
by less than 60% — if f = 4 the capacity gain has decreased considerably to only about 13%. Therefore, in order to achieve a high cellular capacity, the aim is to minimise other-cell interference as only this enables the efficient use of techniques such as interference cancellation. For this reason, it becomes a main goal of this book to reduce other cell interference in a cellular TD-CDMA/TDD network.
2.2.4.2 The downlink in a CDMA system The main differences between the uplink and downlink are: (a) synchronous transmission can be applied in the downlink, whereas in the uplink asynchronous transmission must be assumed and (b) each BS may transmit user specific signals and a common pilot signal for coherent demodulation (applied, for example, in IS-95). A consequence of (a) is that if orthogonal codes are used (for example Walsh codes) to distinguish individual users, the orthogonality in the downlink can be maintained (no own-cell interference) assuming that multipath propagation does not violate the orthogonality at the mobile receiver (H¨am¨al¨ ainen et al., 1997; Takeo et al., 2000; St¨ uber and Yiin, 1991). Therefore, an orthogonality factor τ is defined (H¨ am¨al¨ ainen
Wireless telecommunications using CDMA and TDD techniques
29
et al., 1997), Eb τ= I0
Eb N0
−1
,
(2.18)
where Eb /I0 is the bit-energy to interference ratio when the orthogonality is not maintained and, thus, the signal is corrupted by own-cell interference. The ratio Eb /N0 is the bit-energy to interference ratio for the case that orthogonality is entirely maintained. From the definition of τ it can be seen that the higher its value the more the signals are corrupted by multipath propagation. It is reported that τ may vary between 0.3 and 0.8 (H¨ am¨al¨ ainen et al., 1997) with the greater value obtained in environments that are subject to severe multipath propagation. When an additional pilot signal is used, the total carrier power yields (symbols with the superscript d are associated with the downlink): P cd = Ppilot +
M
P˜cdi ,
(2.19)
i=1
P users
d where Ppilot is the pilot signal power and P˜ci is the code power for the ith user. A factor ψ is used in (Gilhousen et al., 1991) to model the user-specific fraction of the total carrier power,
ψ=
Pusers . P cd
(2.20)
In (Gilhousen et al., 1991), ψ = 0.8 is used. With the approximation of M −1 ˜ d M ˜ d P ci ≈ i P ci , the carrier-to-interference ratio at a particular MS, i d γi , can be modelled as follows: γid
=
d P˜ci /ai
⎞ ⎛ M τ ⎝ ˜ d ⎠ + N + Iod P cj ψ ai i=j noise power other-cell interference
(2.21)
own-cell interference
where ai is the path loss between the desired user i and the respective BS. In a severe multipath environment (τ = 0.8) the advantages due to the synchronous transmission may be cancelled by a greater transmitted carrier power as a consequence of the pilot signal (ψ = 0.8). For this scenario when
30
Next Generation Mobile Access Technologies
τ = ψ, (2.21) yields: γid =
⎞ ⎛ M 1 ⎝ d P˜cj ⎠ ai i=j
own-cell interference
d P˜ci /ai
+
Iod
(2.22) +
other-cell interference
N
noise power
2.3 TDD inherent properties This section addresses the properties inherent to TDD. The basic mechanism, the advantages and disadvantages of TDD are discussed. In TDD, both the uplink and downlink transmissions are carried out at the same radio-frequency carrier. In contrast, the FDD technique requires a separate frequency carrier for each transmission direction (paired radio-frequency spectrum needed). For new global wireless communications systems or in the case of spectrum re-farming it turns out to be difficult to identify a paired radio-frequency spectrum. This is even more complicated when the allocated system bandwidth increases in order to support high-rate transmission, as this would require a proportional increase in the guard band. The TDD mode, in contrast, only needs a single carrier to achieve duplex communication. In general, FDD can be viewed as two separate TDD systems where each of those systems only supports one link direction, i.e., maximum exploitation of link asymmetry in favour of uplink and downlink respectively. This realization forms the basis for various hybrid duplexing techniques. One representative is the TDD underlay, which is discussed in section 2.3.6.
2.3.1 Channel reciprocity Since the uplink and the downlink use the same radio frequency channel, both directions experience the same propagation conditions (channel reciprocity), provided that the TDD frame length is shorter than the coherence time of the channel. This assumption holds for slowly moving user populations. The reciprocity of the channel is a very important property which can be exploited in numerous ways. In this context, it is very important to point out that the exploitation of the channel reciprocity is clearly not limited to the physical layer. It will be shown that it can be used for efficient medium access control, dynamic channel allocation, scheduling, etc. Some
Wireless telecommunications using CDMA and TDD techniques
31
basic features of TDD and their exploitation for an improved system design are discussed in the following.
Pre-RAKE concept The reciprocity of the radio channel can be exploited in the design of receiver structures. A notable example is the Pre-RAKE architecture (Esmailzadeh and Nakagawa, 1993a; Esmailzadeh et al., 1995; Elkamy, 1997; Kadous et al., 1997; Povey et al., 1997; Choi et al., 2000) depicted in Figure 2.3† By using the Pre-RAKE technique it is possible to
Data
PN code correlator
PN code spreading Tx
RAKE filter
data decision
Rx
channel estimate data decision
Tx
Rx PN code correlator
pre-RAKE filter
PN code spreading
data
channel mobile
base station
c Fig. 2.3. The Pre-RAKE concept. 1994 IEEE
retain the advantages of the RAKE combining and at the same time use a simple single-path receiver at the mobile unit. The channel will only be estimated once, at the BS. With the knowledge of the impulse response of the channel, the BS performs the RAKE combining function before transmission (hence Pre-RAKE). Thus, the output signal of the Pre-RAKE combiner is the result of the convolution of the spread-spectrum signal and the timereversed channel impulse response. Since it is assumed that the channel does not vary between the reception and transmission period, a simple receiver structure can be applied at the mobile which ideally only consists of a matched filter. Therefore, the processing effort in the mobile is significantly reduced and, as a consequence, the power consumption can be kept low. This principle can also be applied to multiuser detection, and in particular joint detection in the downlink. The channel reciprocity effectively enables the joint the detection operation to be carried out at the transmitter yielding the concept of joint transmission (Meurer et al., 2000). † Figure 2.3 is reproduced with permission from: (Povey, 1994)
32
Next Generation Mobile Access Technologies
Space diversity Space diversity is also known as antenna diversity. In terrestrial wireless mobile communications, objects such as buildings, trees and mountains cause reflections of the transmitted signal. As a consequence, the signal may arrive via different paths. Due to the accompanied phase shifts the signals may overlap constructively or destructively causing the so-called ‘fading’ effect. This effect is usually modelled by a Rayleigh distribution (Rappaport, 1996). Ideally, the fading statistics can be considered to be uncorrelated between antennas that are spatially separated by more than half the wavelength (Jakes, 1974, p. 311). In fact, the required spacing strongly depends on the disposition of the scatterers and the resulting angular spread of the signal. The antenna separation needs to be greater than half the wavelength in the case of a small angular spread. A small angular spread can be found at macro-cell BSs with an antenna height of several metres above the ground. In these cases the antenna separation needs to be about 10–20 times the wavelength. The small correlation of signals on spatially separated antennas can be exploited by antenna arrays that collect the (ideally) uncorrelated signals and combine them so as to improve the bit-error performance. The reciprocity of the channel in TDD may be used to avoid antenna arrays at the MS, but still able to exploit space diversity on both links. This may be accomplished by an antenna array at the BS. The BS determines the antenna elements that receive the strongest signals and, in turn, transmit the highest power on these antennas (transmit diversity), while the reciprocal channel ensures that the MS receives a strong signal at a single antenna. It was demonstrated in (Hayashi et al., 1995) and (Steiner, 2000) that transmit diversity in TDD can significantly improve the bit error performance. Open-loop power control In FDD the frequency separation of the uplink and the downlink causes both links to experience different fading conditions. This means that the power control unit cannot determine the required transmit power directly from the instantaneously received power level. In the downlink, for example, the received power level of a MS needs to be reported to the BS in order to enable the BS to determine the necessary power it has to transmit (closed-loop power control). This procedure can result in large delays and, in addition, requires extra capacity for signalling. In contrast, under the assumption of a reciprocal channel, a TDD system experiences correlated fading on both links. This means, for example, given that the path loss is known a priori, that an MS can directly determine the power it needs to transmit from the received power level (open-loop power control). This mechanism does not involve extra overhead for signalling.
Wireless telecommunications using CDMA and TDD techniques
33
The benefits of open-loop power control are diminished if large delays occur between a transmit and receive TS at the same entity. Such delays may be induced when the TDD mode and the TDMA technique are combined. Several studies have been undertaken to examine the advantages of open-loop power control in TDD (Esmailzadeh and Nakagawa, 1993b; Sanada et al., 1995; Yongsheng and Daoben, 1997). 2.3.2 Ad hoc and multihop communication Future mobile communication systems are characterised by transmission rates of up to 100 Mbps. This has significant consequences for the network design since the transmit power is proportional to the data rate for a given Tx-Rx separation distance. The received power per symbol of some user k can be described as follows: Pr , (2.23) Sk = Rs where Pr is the received power and Rs is the symbol rate. In general, the received power Pr can be calculated as: Pr =
Pt , LP (d, σ)
(2.24)
where Pt is the transmitted power and LP is the pathloss. The long-term statistics of the channel depend on the Tx–Rx separation distance d and the standard deviation σ of the overlaid lognormal process as a consequence of shadowing effects. Substituting (2.24) into (2.23) yields: Sk = Pt − Lp (d, σ) − Rs
[dBW].
(2.25)
In order to be able to decode the symbol, a certain minimum symbol power Skmin is required which primarily depends on the modulation alphabet as well as the the channel-coding scheme. Under this assumption, (2.25) can be solved for Pt : Pt = Skmin + Lp (d, σ) + Rs
[dBW].
(2.26)
From (2.26) it can be seen that if the transmission rate increases from 100 kbps to 100 Mpbs, for the same propagation environment and transmission system, the transmit power has to be increased by a factor of 1000, or 30dB. It can easily be seen that this is impossible for a number of reasons, such as battery lifetime, health issues, etc. The only option to reduce the transmission power is to reduce the transmission distance d, which eventually results in splitting one large hop into multiple shorter hops. Such a scenario
34
Next Generation Mobile Access Technologies
is depicted in Figure 2.4. An FDD air interface with higher bandwidth in f FDD DL
Tx
Rx
Tx
Rx
FDD UL
Rx
Tx
Rx
Tx
t0
t1
BS
t0
t1
MS1
t0
TDD airinterface1
TDD airinterface2
t1
MS2
Fig. 2.4. Multihop communication assuming an FDD systems with larger bandwidth in the downlink. Since mobile stations need to be able to transmit and receive at the same frequency band, the two frequency bands for uplink and downlink, respectively, are both effectively operated in the TDD mode.
the downlink to cater for channel asymmetries in favour of the downlink is shown in the figure, and a simple two-hop scenario is assumed. The BS transmits a large packet to MS1 and receives a smaller packet from MS1 in the conventional FDD manner. While MS1 is receiving, it cannot transmit. Therefore, it has to wait until the transmission from BS is finished, whereupon it can start forwarding the packet to MS2 at the same frequency band. Note that this automatically results in a slotted system. Furthermore, MS1 need to be able to transmit in the FDD downlink band. In a cellular system that employs FDD, a mobile station is only transmitting in the assigned uplink band. A similar deviation from the conventional FDD approach can be observed for the uplink in the example in Figure 2.4. As a consequence, the original FDD assumption effectively results in two stand-alone TDD systems. This is corroborated by the facts: for multihop communication a slotted frame structure is needed, and the mobile stations need to be able to transmit and receive at the same frequency band. In conclusion, in order to support ad hoc and/or multihop communication the only possible duplexing technique is TDD. Note that the uplink and downlink transmission in the example in Figure 2.4 could have been realized by only a single frequency band and TDD with four timeslots. This mechanism is utilised in ODMA
Wireless telecommunications using CDMA and TDD techniques
35
(opportunity-driven multiple access), a technology that is discussed in detail in Chapters 5, 7 and 8.
2.3.3 Busy-tone concept The key principle of the busy-tone concept is that the receiver sends out a busy-signal on a time-multiplexed channel when it has successfully received a data packet (Omiyi and Haas, 2004; Omiyi et al., 2007). Note that this is different from the dual busy-tone concept in (Haas and Deng, 2002) where a busy-signal is transmitted on a different radio frequency channel, which means that the algorithm in (Haas and Deng, 2002) does not make use of the channel reciprocity. Channel reciprocity, however, is a key element of the busy-tone concept (see Figure 2.5).
Fig. 2.5. Channel-sensing TDMA TDD using busy-tone.
Assume that BS2 has successfully transmitted a data packet to MS2. MS2 (the receiver) transmits a busy-signal in an adjacent mini-slot. A potential transmitter in the neighbourhood, e.g. MS1, listens to that busy-channel prior to transmission and refrains from transmission if the received signal power exceeds a certain threshold. The channel reciprocity is important because, if the received busy-tone signal power at MS1 in the example is above a given threshold, a transmission from MS1 (assuming the same transmit power as used for the busy-tone) would, in turn, result in an interference level at the victim receiver, MS2, which is greater than the known threshold. This means that the potential new transmitter knows that it would cause
36
Next Generation Mobile Access Technologies
an interference level greater than, or equal to, the threshold if it transmits with the same power as the known busy-tone power. In other words, the new transmitter knows exactly the level of interference it would cause to an already coexisting transmission. This enables the new transmitter to determine whether it can transmit or not. In the example, the region in which the busy-tone power can be heard above the threshold is the circular region around MS2. As MS1 is within that region, it does not transmit the scheduled packet and waits until the busy-tone power is below the threshold. In the example in Figure 2.5, MS3 is outside the hearability region of the busy-tone transmitted from MS2. As a consequence, it can transmit using the same channel as BS2. The fact that the busy-tone signal is below the threshold means that the attenuation on the link between MS2 and MS3 is high enough in order to avoid detrimental interference at MS2. It can easily be seen that this protocol works in an ad hoc as well as a cellular deployment. Furthermore, it also works with any arbitrary rate of asymmetry in the network because the decision whether to transmit or not on a particular channel depends solely on interference. It does not matter what the actual source and sink of interference is. Furthermore, the busy-tone serves another very important purpose. It is clear that the own transmitter must be within the hearability region of the busy-tone. Otherwise it would not be able to deliver packets to the intended receiver with the respective quality of service (QoS). As a consequence, when the busy-tone is heard by the own transmitter, it is an indication that the packet has been received successfully, i.e. the busy-tone serves as an acknowledgement at the own transmitter. If the intended receiver gets the packet in outage it would not transmit the busy-tone and would (a) release the channel for other potential transmitters and (b) notify its associated transmitter that it has to re-send the packet once a timeslot is identified in which the busy-tone power is below the threshold. Note that the minislot duration occupies only a very short fraction (about 10%) of the entire timeslot. The busy-tone concept applied to OFDMA (orthogonal frequency division multiple access) is discussed in detail in Chapter 12.
2.3.4 Round-trip delays Since the radio signals propagate with finite speed (the speed of light), the transmitted signal arrives with a certain delay (called the access delay) at the target entity (the BS or MS). Due to the TDD principle the same entity can only then start transmitting its information which arrives at its destination with the same access delay. Hence, during a time period of twice
Wireless telecommunications using CDMA and TDD techniques
37
the access delay, which is equivalent to the round-trip delay, no information can be sent. This mechanism is shown in Figure 2.6. The round-trip delays Entity
Transmission Reception Round trip delay
Base station
Mobile station
guard time
guard time
guard time
time slot (TS) duration
time
Fig. 2.6. The round trip delay in TDD.
require a certain guard time. It is easy to see that guard times result in a reduction of the spectral efficiency. As round-trip delays increase with increasing transmitter–receiver separation, they represent a major problem in TDD systems since they are primarily used for short-range wireless communication (DECT, WLAN (IEEE 802.11), Bluetooth, to name only a few). One possible measure to decrease the ratio between guard time duration and the TS duration is to simply increase the TS duration. This however can violate the channel reciprocity. Clearly, these conflicting requirements result in a trade-off between channel reciprocity and spectral efficiency, which was investigated in (Povey, 1994). Recently, relaying and ad hoc systems have attracted the interest of researchers (Gupta and Kumar, 2000). The relaying technique implicitly reduces the problems resulting from round-trip delays since there are now multiple hops from BS to MS, and vice versa. This increases the probability of shorter hops. 2.3.5 Synchronisation and channel asymmetry A powerful advantage of TDD is its support of channel asymmetry (Povey et al., 1997; Jeon and Jeong, 2000) without the need to resort to on-thefly filter reconfigurations (as would be required in FDD). This property becomes increasingly important as it is predicted that wireless data applications, e.g. wireless Internet, will demand up to five times more capacity in
38
Next Generation Mobile Access Technologies
the downlink than in the uplink. However, if the TDD interface is used in a multiple cell environment severe interference problems can occur. These problems arise primarily because TDD is exposed to additional interference mechanisms (in comparison with FDD). These mechanisms are illustrated with the aid of Figure 2.7. This figure shows a simple cellular structure
BS1
BS2
MS1 MS2
BS1
TX TX
BS2 MS1
RX
RX
RX TX TX : Transmit Slot
MS2
RX
TX
RX: Receive Slot time
Fig. 2.7. Interference scenarios in cellular TDD systems with cell-independent channel asymmetry.
consisting of two adjacent cells. In each cell a different rate of asymmetry is used.† This causes asynchronous TS overlaps (shaded areas) which has a significant impact on the total interference. In this context, the BSs interfere not only with the neighbouring MSs, and vice versa (FDD-equivalent interference scenario), but also with other BSs. Similarly, MSs interfere with adjacent MSs. The latter interference scenarios are depicted in Figure 2.7 (BS→MS and MS→BS interference is left out for reasons of clarity). In particular, the interference between MSs can be severe given that the distance between two MSs at the cell boundary can be very low. However, in (Holma et al., 1999) BS↔BS interference was reported to be more significant than † Note that symmetric traffic is only a special case of the set of achievable rates of asymmetry.
Wireless telecommunications using CDMA and TDD techniques
39
MS↔MS interference. It is important to note that downlink power control was not assumed in that investigation. In order to avoid BS↔BS and MS↔MS interference the frames must be synchronised and both cells need to adopt the same rate of asymmetry. Clearly, this results in a significant limitation which may greatly affect the spectral efficiency and flexibility. If the TDD mode is used in TDMA an additional degree of freedom exists (in the time domain) which can be used to resolve additional interference by using dynamic channel assignment (DCA) algorithms (Punt et al., 1998; Chen et al., 1998). The TDD mode of UMTS (UTRA-TDD) consists of a TD-CDMA interface which adds another dimension (code domain) that can be utilised to establish a connection (Haardt et al., 2000). The combination of a code, TS and frequency in UTRA-TDD is defined as a resource unit (RU) (Mihailescu et al., 1999). Chapter 3 is dedicated to characterising other cell interference in a hybrid TD-CDMA/TDD system, assuming different rates of asymmetry in neighbouring cells. The results of the interference study will be used to develop DCA strategies, and it can be shown that asynchronous TS overlaps may be used constructively to permit cell-independent asymmetries (see Chapter 4). It can also be shown that asynchronous TS overlaps can be utilised to enhance cell capacity (see Chapter 5).
2.3.6 The TDD underlay concept Due to increasing asymmetric traffic on the air interface, one communication direction of an FDD interface will be underused provided that the carrier spacing is kept constant. The main idea behind the TDD underlay is to exploit this underused radio spectrum of a cellular CDMA-FDD system. A coexisting TDD interface, which only needs an unpaired frequency spectrum, utilises the underused FDD frequency band for additional connections. Technically a TDD interface can be used within either the FDD uplink or downlink frequency band. A hierarchical system architecture (Chih-Lin et al., 1993; Lee, 1994) consisting of TDD pico cells which exclusively use the FDD resources of a macro cellular overlay is proposed (Ma et al., 1997; Sunay et al., 1997). The DCA decides whether to use the FDD uplink or downlink frequency band. The TDD underlay concept is depicted in Figure 2.8. The dot-dashed lines with hollow arrowheads indicate the interference paths when the TDD pico–cell uses the FDD uplink frequency band. It can
40
Next Generation Mobile Access Technologies
Fig. 2.8. The TDD underlay concept.
be seen that the macro–cell MS interferes with the pico–cell BS and MS. In turn, the pico–cell entities interfere with the macro–cell BS. The proposed CDMA-TDD underlay in (Sunay et al., 1997) uses a PN code sequence with a unique cell-phase offset relative to the CDMA-FDD macro cellular overlay. The authors reported substantial capacity gains without a significant deterioration of the quality of service (QoS). The picocells are assumed to consist of a single indoor BS and a single indoor MS and are randomly distributed. Note, however, that severe interference can be anticipated if macro and pico-cells are in close proximity (Takeo, 1996). Further results on the TDD underlay concept are reported in (Haas and Povey, 1998, 1999). In these studies the TDD system is only considered to be operated in the FDD uplink band due to an anticipated channel asymmetry in favour of the downlink. In another investigation, the feasibility of the TDD underlay was confirmed (Wong and Sousa, 1998, 1999) provided that the BS separation distance is properly chosen. The frequency usage when operating the TDD underlay in a system with two air interfaces (FDD and TDD) and harmonised frame structures is shown in Figure 2.9†. The components of the TDD pico–cell or indoor cell are followed by the letter ‘i’ (BSi, MSi respectively) whilst the FDD macro-cell or outdoor cell entities are marked with an additional ‘o’. † Figures 2.9, 2.10 and 2.11 are reproduced with permission from (Haas and Povey, 1999).
Wireless telecommunications using CDMA and TDD techniques
41
FDD Downlink
Frequency
BSo
Rx / MSi Tx / BSi
Tx / MSi Rx / BSi
FDD Uplink
Paired Band
MSo
Unpaired Band
BSi
Rx / MSi Tx / BSi
Tx / MSi Rx / BSi
MSi
Power Time Fig. 2.9. Frequency use in a dual interface system in which the TDD underlay is c applied in order to achieve greater flexibility. 1999 IEEE
Note, that the TDD underlay may be used to accomplish cell-independent channel asymmetry in the TDD subsystem without asynchronous TS overlaps. Furthermore, the flexibility on the air interface is increased as, for example, uplink radio resources can be converted into downlink capacity. In (Haas and Povey, 1999) the system in Figure 2.9 is analysed under the following conditions: u = – 8.5 dB; • carrier-to-interference threshold: γpico u • carrier-to-interference threshold: γmacro = – 19.0 dB; • 10 dB wall attenuation around the indoor pico-cell;
42
• • • • • •
Next Generation Mobile Access Technologies
symmetrical speech service of 8 kbps; spatially uniform user distribution; tolerable outage of 5% ; pico-cell radius of 50 m; macro-cell radius of 1000 m; path loss exponent of 3.8.
Note that, although voice service is used in this example above, the basic concept is equally applicable to packet data traffic. The capacity (number of simultaneously active MSs) is calculated with the standard deviation of lognormal shadowing (σ) and the BS separation distance (d0 ) as a parameter. The results are depicted in Figure 2.10. From Figure 2.10 it can be seen that as d0 decreases there is a corresponding increase in the FDD uplink capacity required for an additional TDD link, which is due to the high interference from the TDD underlay entities. Conversely, as d0 increases there is a corresponding decrease in the likelihood that spare capacity in the FDD uplink can be exploited by the TDD underlay. The reasons are that the macro-cell mobiles cause high interference since the distance between a MSo and the BSi can be very small, and that the transmission powers of the macro-cell mobile, MSo, are highest at the cell edge. In Figure 2.10 the solid curves show the remaining capacity in the FDD uplink as a function of the number of MSi’s. In contrast, the dashed curves show the additional TDD capacity dependent on the number of MSo’s. In both cases the arrows indicate the functional relationship (x → f (x)). It can be seen that the higher the BS separation the more the basic FDD uplink capacity is preserved. On the other hand, the higher the BS separation the less the capacity that can be found in additional pico-cells due to the greater transmission powers of the macro-cell mobiles at the outer cell regions. This trade-off leads to an optimum for the BS separation. The general aim is to convert unused TDD uplink capacity into additional communications channels. The issue is that the additional capacity cannot be guaranteed because it depends on the instantaneous interference scenario, i.e., the interference that is generated from close-by macro-cell mobiles. For the time being it is not assumed that a macro-cell mobile is handed over to the pico-cell. The mutual dependencies are explained more precisely with the aid of Figure 2.10(c) and with the following example. Initially 10 macrocell mobiles are assumed, which leads to approximately 17 additional links in the pico-cell. In turn, 17 pico-cell users will accommodate up to about 72 macro-cell mobiles. Here the effects of mutual interference become apparent because 72 macro-cell mobiles would generate too much interference to per-
Wireless telecommunications using CDMA and TDD techniques a) rb=200m
b) rb=300m 80
UTRA-FDD uplink capacity Additional UTRA-TDD capacity
70
Additional pico cell mobiles - MSi
Additional pico cell mobiles - MSi
80
σ=6 σ=8
60 50
σ=6
σ=11
40 30 20 10 0
σ=8 0
10
D
20 30 40 50 60 FDD uplink users - MSo
70
50
σ=6 σ=8
40 30 20
UTRA-FDD uplink capacity Additional UTRA-TDD capacity
σ=6
50
σ=8 σ=6
40
σ=8
30 20 10
D 0
σ=8
D
10 0
10 20 30 40 50 60 70 80 90 100 FDD uplink users - MSo
d) rb=500m 60
σ=11
10 20 30 40 50 60 70 80 90 100 FDD uplink users - MSo
Additional pico cell mobiles - MSi
Additional pico cell mobiles - MSi
σ=6
60
c) rb=400m
60
0
UTRA-FDD uplink capacity Additional UTRA-TDD capacity
70
0
80
80 70
43
D
UTRA-FDD uplink capacity Additional UTRA-TDD capacity
50
σ=11
σ=8
σ=6
40 30
σ=6
20
σ=8 10 0
0
10 20 30 40 50 60 70 80 90 100 FDD uplink users - MSo
Fig. 2.10. The additional numbers of MSs in the pico-cell are indicated by the dashed lines. The number of supported mobiles in the FDD uplink is shown by the solid lines. Different lognormal shadowing scenarios are depicted in each plot. Moreover the effects of different BS separations are shown in plots (a)–(d): (a) c d0 = 200 m, (b) d0 = 300 m, (c) d0 = 400 m and d) d0 = 500 m. 1999 IEEE
mit an additional TDD link. Hence, the flexible exchange of radio resources between the FDD and the TDD mode is limited. It is desirable to define a measure for the flexibility of the pico-cellular underlay. This is derived with the aid of Figure 2.11. The set E is defined as the operational area of the additional pico-cell capacity and F is defined as the operational area of the FDD uplink capacity. Let D = E ∩F then it can be stated that the flexibility increases with D. This can be applied to the results in Figure 2.10(a)–(d). Set D for σ = 8 dB is highlighted for d0 = 200 m to d0 = 500 m. It can be seen that the maximum is reached for d0 = 300 m whereas for d0 = 500 m it shrinks to just one point. Furthermore, for small BS separations D is de-
Next Generation Mobile Access Technologies
cap. A-TDD
R Add. UT
number of pico cell mobiles MSi
44
E
D =E ∩F
UT
RA
D
-FD
Du
plin
F
kc
ap.
number of macro cell mobiles MSo
Fig. 2.11. The area defined by D serves as a measure for the flexibility of the TDD Underlay. Within the area specified by D the FDD uplink radio resources can be c exchanged between the macro and pico cells. 1999 IEEE
termined by the intensive use of FDD uplink capacity. In contrast, for high BS separations only the reduced additional TDD capacity limits D. The approximate maximum capacity of one CDMA-TDD carrier is 64 (8 users per slot and 8 slots per frame in total) assuming equal data rates, symmetrical services and no multi-user detection. This maximum for the CDMA-FDD uplink is about 81 users with an overall spreading factor (SF) of 256. These figures can be compared with the cumulative maximum in Figure 2.10(b) with σ = 8 dB, which is about 70 users (e.g. 40 pico-cell users and 30 macro-cell mobiles simultaneously). Thus, there is no significant increase in spectral efficiency. Lognormal shadowing with σ = 11 dB results in no advantage to the TDD underlay, i.e. D = ∅ for all d0 . Furthermore, lognormal shadowing with σ = 6 dB yields the best results for d0 = 300 m. In this case even the spectral efficiency can be significantly increased because in total more than 81 users can be accommodated within the UTRA-FDD uplink (e.g. with 70 MSo’s about 50 MSi’s can be served in the FDD uplink, resulting in an increased spectral efficiency of about 50%). Slow fading (shadowing), usually modelled as a lognormal random variable with zero mean, has a significant impact on the performance of the TDD underlay concept. The reason for this strong dependency is that the interference path and the desired path cannot be assumed to be strongly
Wireless telecommunications using CDMA and TDD techniques
45
correlated since co-location is not feasible. As a consequence the interference signal can be significantly higher than the received signal at the desired BS. Furthermore, the extra capacity in the FDD uplink cannot be guaranteed at any time; it is scenario dependent. However, channel asymmetry is primarily demanded by packet-oriented data services which may not require continuous access to the channel (relaxed real-time requirements). Therefore, the conditions under which additional capacity can be gained from the TDD underlay and the properties of packet-data services complement each other.
2.4 The TDD based air interface in UMTS The air interface of UMTS is known as UMTS terrestrial radio access (UTRA) where the TDD and FDD modes are referred to as UTRA-TDD, and UTRAFDD respectively. The UTRA-TDD mode is composed of a TDMA and a CDMA component. This hybrid multiple access technique is described as TD-CDMA. The basic radio interface parameters such as chip rate, bandwidth and modulation are harmonised between the UTRA-TDD and UTRAFDD systems. A good overview of the UTRA-TDD interface can be found in (Haardt et al., 2000). The channel spacing is 5 MHz with the channel raster being 200 kHz (3GPP, TSG, RAN, WG4, 2000), which means that the carrier frequency must be a multiple of 200 kHz. The frame duration is 10 ms. One 10 ms frame corresponds to one power control period. A frame is divided into 15 timeslots (TS) and one TS consists of 2560 chips resulting in a chip rate of 3.84 Mcps. In addition, UTRA-TDD also supports a low chip rate option (1.28 Mcps) in order to facilitate future system extensions. Each TS can be allocated to either the uplink or downlink. This provides the basic mechanism to easily enable the asymmetric use of the channel. In any configuration, at least one TS has to be allocated to the downlink and at least one TS has to be allocated to the uplink. OVSF codes with a maximum SF of 16 are used for channelisation. The short spreading factor permits the use of joint detection (Klein et al., 1995). In the downlink either an SF of 16 or 1 is applied. Parallel physical channels can be used to support higher data rates above the basic rate of 16 kbps. In the uplink the SF is variable and ranges from 1 to 16 (3GPP, TSG, RAN, 1999a). The spreading and scrambling procedures are more detailed described in (3GPP, TSG, RAN, 2003) and are summarised here for convenience. The spreading of data by a real-valued channelisation code of length Q is followed
46
Next Generation Mobile Access Technologies
by a cell-specific complex scrambling sequence, v = (v 1 , v 2 , . . . , v 16 ), where the index specifies the chip-number. The elements of the complex valued scrambling codes shall be taken from the complex set: V = (1, j, −1, −j)
(2.27)
In (2.27) the letter j denotes the imaginary unit. A complex scrambling code is generated from the binary scrambling codes, vk = (v1 , v2 , . . . , v16 ), of length 16. The relation between the elements is given by: π (2.28) v k = exp j(k + ℓ π) k = 1, . . . , Q ℓ = {0, 1} , 2 where k is the chip number and ℓ the actual value of the binary chip. In addition, the relationship between ℓ and the binary scrambling codes, vk , is: 0 for vk = 1 (2.29) ℓ= 1 for vk = −1. Due to the TDMA component an effective interference avoidance mechanism exists, because interference between neighbouring MSs can be eliminated by using different TSs provided that the frames are synchronised. However, this requires the use of specially designed channel allocation strategies, especially if different channel asymmetries between neighbouring cells are to be supported. 2.5 Radio resource allocation techniques In a cellular network certain radio resources allocation methods are required to mitigate the detrimental impact of interference (CCI and ACI). Three basic concepts of radio resource allocation can be distinguished (Cox and Reudink, 1974; Ahlin and Zander, 1998): • static or fixed channel assignment (FCA) techniques; • dynamic channel assignment (DCA) techniques; • random channel assignment (RCA) techniques. The principles of these methods are described in the following section. 2.5.1 Fixed channel assignment techniques A FCA method allocates a fixed fraction of all available channels to an individual cell of a cellular environment. The same group of channels is only used in cells that are separated by a minimum distance D. The channel reuse distance D ensures that CCI does not deteriorate the system performance
Wireless telecommunications using CDMA and TDD techniques
47
greatly. The cluster size basically determines the system capacity since it specifies the maximum number of simultaneously active connections that can be supported at any given time. The group size that equals the number of channels per cell, M , can be found from (2.3). It can be seen that M is increasing with a decreasing cluster size, K, but this also means that the interference is higher which, in turn, reduces the capacity or QoS (quality of service). The impact of the cluster size K on capacity can be studied with the following model: let the number of users who request a channel, Mc , be Poisson distributed with mean E(Mc ) = λ, then the assignment failure probability can be defined as follows: ∞ λi−1 E [max(0, Mc − M )] (i − M ) p(λ) = = exp(−λ) E (Mc ) i!
(2.30)
i=M
The relative traffic load can be expressed as:
λ (2.31) M Substituting (2.31) into (2.30) yields the final assignment failure probability: ω=
p(ω) =
∞
i=M
(i − M )
(M ω)i−1 exp[−(M ω)] i!
(2.32)
Note, that (2.32) is merely based on traffic theory and assignment failures due to high CCI are therefore not considered. The results of (2.32) for a different number of available channels per cell is plotted in Figure 2.12. It is obvious that the assignment failure rate increases as the relative load, ω, grows. The interesting result is that for the same relative load the failure rate increases with a decreasing number of channels per cell, M . This means that a system with a greater number of channels per cell is more efficient than a system with only a few channels. This effect is known as the trunking gain. Consequently, fixed channel assignment techniques result in poor spectral efficiency. Given that CCI varies with the cell load, there might be traffic scenarios where a lower channel reuse distance can be tolerated in favour of a temporarily higher number of channels available in a single cell (or cluster of cells). This would require methods that dynamically monitor interference and load situations throughout the network and which carry out channel reconfigurations accordingly. In contrast to DCA strategies, FCA techniques are not designed to achieve this flexibility. CDMA systems such as the UTRA-FDD interface of UMTS reuse the same channel in every cell, which, in theory, makes FCA or DCA techniques
48
Next Generation Mobile Access Technologies
Fig. 2.12. Probability of channel assignment failures for varying reuse cluster sizes and a constant number of totally available channels.
superfluous (Holma et al., 1999), but requires special handover techniques (soft-handover). In contrast, the TDD mode of UMTS, UTRA-TDD, requires certain methods of intelligent channel assignment (Haardt et al., 2000) due to the hybrid TD-CDMA interface and the additional interference mechanisms induced by TDD (see section 2.3.5).
2.5.2 Dynamic channel assignment techniques DCA techniques enable a cellular system to adapt flexibly to different load situations thereby increasing the throughput and decreasing the call blocking. In addition, the efforts for frequency planning can be reduced or eliminated. An important issue, however, is to ensure that the DCA algorithm does not lead to the instability of the system (Everitt and Manfield, 1989). In this context, it is reported that under high load conditions DCA algorithms can perform worse than FCA techniques due to continuous channel reassignments (Beck and Panzer, 1989). Consequently, systems are investigated where FCA and DCA techniques are combined (Tan et al., 1998; Kunz, 1999). As a result, for example, a subset of channels is assigned in a fixed way and the remaining channels are allocated to a common pool. A
Wireless telecommunications using CDMA and TDD techniques
49
DCA algorithm uses the common pool from which it ‘borrows’ channels in order to allocate them to cells where heavy traffic occurs. In this way, the number of channels of a cell can be increased dynamically, but there is still a fixed number of channels which ensure a certain QoS in situations where the entire network is heavily loaded. In addition, DCA and handover techniques are investigated for narrowband TDMA systems in a multi-layer (micro, pico and macro-cell layer) environment (Nanda and Goodmann, 1992; West and St¨ uber, 1994; Chu and Rappaport, 1995; Lo et al., 1997). A classification of different DCA approaches was made in (Beck and Panzer, 1989) and (Ahlin and Zander, 1998), and is repeated here for convenience (Figure 2.13†). In the figure, different DCA strategies are represented
Optimum DCA
C re han -u ne sa l bi lit y
Adaptability to interference
FCA
Adaptability to traffic
Maximum Packing (MP)
c Fig. 2.13. Classification of dynamic channel assignment algorithms. 1989 IEEE
by points in the space – except the origin, which represents the FCA scheme. The three axis represent different optimisation criteria. One criterion is the knowledge about the load in every cell of the network. A DCA algorithm can be designed to optimise the number of active MSs in every cell (adaptability to traffic) – resulting in maximum packing (MP). A second criterion is the adaptability to interference variations, which is particularly important for interference-limited systems such as CDMA. This requires the DCA to have † Figure 2.13 is reproduced with permission from: (Beck and Panzer, 1989).
50
Next Generation Mobile Access Technologies
information of the instantaneous interference situation. A third criterion is to optimise the channel reuse distance, which would eventually mean that each channel can be used in every cell. An ideal DCA algorithm tries to optimise each of the three parameters which would yield the optimum solution at the far upper corner of the cube. The ideal DCA algorithm would require information beyond the scope of a single cell. From this requirement it inherently follows that the respective DCA algorithm would ideally be operated at a central site. Consequently, two basic DCA schemes can be distinguished (Jorguseski et al., 1999): • centralised DCA schemes; • decentralised DCA schemes. A centralised DCA algorithm collects the required information for channel assignment decisions from the associated BSs and MSs. This type of DCA algorithm is located at a higher hierarchical level of the mobile network architecture. A centralised DCA algorithm can, for example, be located at the radio network controller (RNC) (Mihailescu et al., 1999) which connects several BSs. The basic disadvantage is that a great amount of signalling is necessary to supply the vital information about the load, interference and channel status. In a decentralised DCA algorithm the channel assignment decision is made by a local instance (Cimini et al., 1992, 1994; Das et al., 1997b; Prakash et al., 1999). Thus, only local information is available. Hence, the complexity is reduced considerably when this type of DCA algorithm is used. The DECT standard, for example, uses a decentralised DCA algorithm (Punt et al., 1998). Given that a decentralised DCA only has a limited knowledge about the system state, a global optimum is very difficult to achieve. As a result of the DCA characterisation in Figure 2.12, three types of strategies can be formulated: Traffic-adaptive channel allocation: Instead of splitting the available radio resource into sets of channels that are then assigned to cells, the traffic-adaptive channel allocation techniques try to dynamically assign the required number of channels to cells (Haas et al., 1997). In order to avoid the use of the same channel in the neighbouring cell, compatibility matrices are established. Since the interference level from neighbouring cells can vary significantly, the compatibility matrices have to be designed in a way that also ensures a reliable connection under severe interference conditions. This results in capacity losses when static compatibility matrices are assumed, since
Wireless telecommunications using CDMA and TDD techniques
51
in low interference situations more channels per unit can be accommodated. The complexity of traffic-adaptive channel allocation algorithms increases exponentially with the number of cells, (M K)L , where L is the total number of cells. Due to this complexity, graph theory is often used to solve these issues (Haas et al., 1997). The optimum traffic-adaptive DCA algorithms result in the MP solution. This requires intra-cell handovers (channel reassignments). As a consequence, users may be reshuffled (although the QoS is still fulfilled) in order to optimise the total number of users. Hence, the complexity of MP strategies is further increased, which makes this type of DCA algorithm difficult to analyse. In (Yeung and Yum, 1995) the cell group decoupling method is proposed in order to calculate upper bounds on blocking performance. However, the MP problem is easier to solve if the cells are placed along a line. In this scenario, the complexity only increases linearly with L. The optimum solution is found using the Greedy algorithm (Ahlin and Zander, 1998). In (Das et al., 1997a,b), (Chang et al., 1998), and (Ortigoza-Guerrero and Lara-Rodriguez, 1996) a different approach was taken to achieve the dynamic adaptation to traffic variations: channels were ‘borrowed’ from a common pool and assigned to cells which experience heavy traffic. This technique resembles the previously described combination of DCA and FCA methods. Reuse partitioning: The entire set of channels is divided into subsets, similar to the FCA strategy. But, in the case of reuse partitioning, every group of frequencies is associated with a different reuse distance, D (Zander and Frodigh, 1992). If the reuse distance is small it is more likely that severe interference is the reason why a channel cannot be assigned. Due to the user mobility, however, the interference level may be in favour of allocating a channel with a small reuse distance, thereby increasing the total capacity. Hence, the respective DCA algorithm always tries to use a channel with the lowest reuse distance. It is reported that reuse partition algorithms can double the capacity compared to FCA schemes (Lucatti et al., 1997). Interference-based DCA schemes: Channels are assigned based purely on the interference power observed (no use of compatibility matrices). If the signal-to-noise ratio drops below a certain threshold a new channel is acquired. This requires steady and reliable measurements of the interference power and power control to minimise interference. Due to its simplicity (easy to implement as distributed DCA algorithm) this type of DCA is widely used. A notable example
52
Next Generation Mobile Access Technologies
is the DCA algorithm of the DECT standard. Furthermore, since CDMA are interference-limited systems, interference-based DCA algorithms play an important role in such systems (Shin et al., 1999). In (Argyropoulos et al., 1999), a C/I-based DCA algorithm is compared with an FCA algorithm assuming non-uniform traffic. It was found that the DCA techniques increase the throughput by up to 16% at the cost of higher average transmitted powers. Due to the sensitivity of CDMA to interference, the combination of DCA techniques and power control techniques also represents an interesting approach (Lozano and Cox, 1999).
2.5.3 Random channel assignment techniques In the case of narrowband FDMA systems, the RCA methods are closely related to slow FH techniques. The basic idea is to change the channels randomly in order to mitigate poor channel conditions (deep fade) of a static channel. Thereby the interference condition in each hop are considered to be independent. It is anticipated that by changing the channels continuously the average signal-to-noise ratio is sufficiently high in order to enable errorcorrecting codes and interleaving techniques to achieve the required QoS.
2.6 Summary Compared with other techniques discussed in this chapter, CDMA offers a high degree of flexibility since it is merely interference limited. The cost associated with the flexibility is that CDMA is very sensitive to system functions that have a vital impact on interference (handover, power control, channel assignment, etc.). An important consequence is that the actual system capacity significantly depends on these functions (soft-capacity). The capacity enhancement of CDMA over FDMA and TDMA systems is primarily due to the fact that the same frequency can be used in every cell. However, this requires a careful system design. In particular, power control and inter-cell handover are important functions as otherwise interference can rise considerably, which can cause a significant capacity reduction. In the past, CDMA systems were almost exclusively considered in combination with FDD, it is only recently that CDMA has been associated with TDD. A notable example is the UTRA-TDD interface of UMTS, which is a hybrid TD-CDMA/TDD system. There are distinct advantages of using the TDD mode which are primarily due to the reciprocity of the channel and the property that channel asym-
Wireless telecommunications using CDMA and TDD techniques
53
metry can easily be adopted. In addition, a CDMA-TDD system can be used to increase the flexibility in a coexisting CDMA-FDD interface (TDD underlay concept). Furthermore, TDD enables the efficient use of enhanced technologies such as SDMA and ODMA. The main disadvantage of TDD is that additional interference scenarios can exist. These additional interference mechanisms occur if the network is not synchronised or the neighbouring cells apply different channel asymmetries. Furthermore, handover techniques such as soft-handover are of limited use in TDD systems such as UTRA-TDD. Since CDMA is sensitive to interference, the capacity of a CDMA-TDD system can suffer significantly due to the additional interference mechanisms of TDD. If this problem can be resolved or eased, the CDMA-TDD system architecture represents a very flexible air interface tailored to the needs of future applications such as wireless Internet. Therefore, new mechanisms have to be developed to exploit TDD’s inherent advantages whilst minimising its disadvantages. DCA techniques seem to be an appropriate means of achieving this requirement. In particular, interference-based DCA algorithms are the most promising candidates as they inherently support the requirement of low interference. In order to find the best DCA strategy it is critical that the interference in such a system is first characterised quantitatively. Therefore, the next chapter is dedicated to interference and capacity issues of a cellular CDMA TDD system. The results of these studies provide a valuable input to the investigation of interference-based DCA algorithms.
3 Interference and capacity analyses Harald Haas
3.1 Introduction A special property of the TDD mode is that not only the mobile stations (MSs) can interfere with the base stations (BSs), but also the MSs may interfere with each other. The same holds for BSs, which can interfere with MSs as well as other BSs. This property of a TDD interface creates complex interference situations. The resulting interference has a significant impact on capacity. In this chapter, other-cell interference including adjacent-channel interference (ACI) and co-channel interference (CCI) are examined. In this context, multi-operator scenarios are considered since in UMTS (Universal Mobile Telecommunications System), for instance, any one operator may only obtain a single TDD carrier. In order to assess the impact of interference on capacity (the number of users that can be served simultaneously), in section 3.2 a new equation is derived to calculate the system capacity relative to the non-interfered state (a single, isolated cell). In addition, the pole capacity (the theoretical upper bound of capacity) is calculated for the case of ideal power control and non-ideal power control. In section 3.3 ACI is studied in single-cell and multiple-cell environments. The probability density function (pdf) of interference in the single-cell environment is derived analytically in order to validate the simulation platform. In the multiple-cell environment the effects of handover and power control on interference and capacity, respectively, are examined. In section 3.4 CCI is analysed with a special focus on distribution of interference with respect to the basic interference sources: MSs and BSs. The interference characterisation provides valuable insight into critical modes of operation. This information is to be utilised for the development of dynamic channel assignment techniques in subsequent chapters. 54
Interference and capacity analyses
55
3.2 Capacity definition In communication systems where many MSs are connected to a single BS, the characteristics of the uplink (UL) and downlink (DL) are different. Since in CDMA systems all users share a common radio resource, the different characteristics of the UL and the DL have a significant impact on these systems. This is because the UL is a multi point-to-single point transmission (i.e. the signals arrive asynchronously) whereas the DL is a single pointto-multi point transmission (i.e. the signals arrive synchronously). Since the signals of all MSs but the desired MS appear as noise, a close-by MS transmitting at high power can affect the quality of all other connections. This is the well-known near-far effect, which requires tight power control in the UL. This means, ideally, that the received signal powers of all MSs must be the same. Due to the near-far effect, the UL is often considered as the capacity limiting direction (Viterbi and Viterbi, 1993; Corazza et al., 1998). In the next section capacity is analysed assuming ideal power control in the UL whilst in section 3.2.2 non-perfect power control is considered.
3.2.1 Capacity assuming ideal power control In order to demonstrate the sensitivity of capacity on interference, a mathematical model for the relative capacity is established in this section. This means that the capacity relative to the state without any other cell interference (single, isolated cell) is calculated. In addition, it is assumed that dynamic power control in such a system is used. This requires the system to modify the power level of the desired signal in order to maintain the target carrier to interference ratio in the presence of other cell interference. The maximum desired signal power level is limited, however, as otherwise the cell coverage would be affected. Therefore, some form of admission control is necessary to prevent the reduction of the cell coverage area, or to prevent the deterioration of the quality of service of the existing connections. In general, the model follows closely the definitions used in papers by Gilhousen and Viterbi (Gilhousen et al., 1991; Viterbi and Viterbi, 1994). For reasons of simplicity, these papers assume ideal power control in the UL, i.e. the received signal power of each MS is the same. As a consequence, the effects of the near-far effect are eliminated. With the assumption of ideal power control and the same bit-rate for each user, the bit-energy to interference ratio at the BS, as given in (Gilhousen et al., 1991), can be calculated as
56
Next Generation Mobile Access Technologies
follows: εu =
Puu pg , (M − 1) + Iou + N
Puu
(3.1)
Iown
where Puu is the received, power-controlled signal power from the desired user at the BS, and M is the number of simultaneously active users of equal bitrate. The interference power is represented by Iou , N is the thermal noise power and pg is the processing gain. The term [Puu (M − 1)] represents own-cell interference. Note that symbols are followed by the superscript u are associated with the uplink channel; symbols which are followed by the superscript d are associated with the downlink channel. The target is to calculate the capacity relative to the single cell capacity. To start with, (3.1) is solved for Puu : Puu =
Iou + N . (pg/εu ) + 1 − M
(3.2)
Since the denominator in (3.2) must be positive and greater than zero, an upper bound for the cell load can be determined which yields Mmax . This upper limit is also referred to as the pole capacity (Holma and Toskala, 2000, Chapter 8): pg (3.3) Mmax = u + 1 . ε The admission control of a CDMA system ensures that the number of users is below this threshold. Hence, in this book a cell load factor is defined as follows: χ=
M0 Mmax
0≤χ≤1,
(3.4)
where M0 is the maximum number of simultaneously active users that can be permitted, assuming no other cell interference. The maximum cell load factor is assumed to be a fixed parameter set in the admission control. The actual value depends on the cell radius. For smaller cells a higher value of χ can be tolerated than for larger cells because the target receive power Puu can be greater due to lower path losses. However, the number of users, M0 , must always be less than the pole capacity to ensure that the desired signal Puu does not approach infinity (singularity in (3.2)) or becomes negative. Since a certain cell load also corresponds to a certain power Puu , χ can be interpreted so as to define the upper bound for the dynamic power control.
Interference and capacity analyses
57
Thus, substituting (3.3) into (3.2) yields: Puu =
Iou + N Iou + N ≤ Mmax − M M0 χ1 − 1
where M ≤ M0
(3.5)
A single cell is modelled by setting Iou = 0. For this case and by setting M = M0 , (3.5) becomes u = Pref
N . Mmax − M0
(3.6)
Hence, (3.6) is the desired maximum receive power level for the ideal case (single-cell case) where no other cell interference is present. This serves as reference power level and is fixed in order to allow for a fair comparison (it is recognised that in an interference-limited system transmit power and capacity are directly related). A few constraints have to considered. The reference level must always be higher than the receiver sensitivity of the system, which depends on the ambient temperature, T , and the channel bandwidth, B, Psens = ι T B,
(3.7)
where ι is the Boltzmann constant. Moreover, the maximum reference level depends on the available transmit power of the mobile and the maximum cell radius as well as the propagation channel. The higher the reference level, the more the system is operated close to the pole capacity, which would imply small cell radii. Therefore, it is important to fix the receive power value as this will ensure that coverage is maintained, and that coverage is not traded-off against a higher capacity. Now, (3.5) is solved for M , which yields: Iu + N (3.8) M = Mmax − o u . Pu Since it is aimed to obtain the capacity relative to the non-interfered state, (3.6) is substituted into (3.8). Hence, M represents the capacity assuming that the received signal power is the same as for the single-cell case, which also ensures that the cell radius is maintained: M = Mmax −
(Iou + N ) (Mmax − M0 ) . N
(3.9)
Simplifying (3.9) yields: M = M0 −
Iou (Mmax − M0 ) . N
(3.10)
58
Next Generation Mobile Access Technologies
If other cell interference equals zero, it can be seen that M = M0 , as expected. Finally, the relative remaining capacity as a consequence of Iou can be found by dividing (3.10) by M0 :
M Iou 1 −1 . (3.11) =1− M0 N χ The relative capacity in (3.11) is interpreted as follows. First, it is a measure of the capacity reduction due to other-cell interference assuming ideal power control (the received signal of each user is the same). Second, it is inherently assumed that the desired receive power is constant regardless of the cell load. The actual value of the target receive power is derived from a singlecell scenario with an associated number of supported users of M0 , and thus determined from the maximum cell load factor χ. This ensures that the cell coverage is maintained. From (3.11) it can be seen that a higher value of χ results in less vulnerability to interference. In the limit when χ = 1 the system is insensitive to interference because it is operated at the pole capacity, which would mean a target receive power of infinity in which case the cell radius is shrunk to a single point. From this can be drawn the important conclusion that smaller cells are more resistant to interference, for a given MS power budget. For reasons of simplicity, ideal power control was assumed in this section. In order to consider a more realistic scenario, in the following section the pole capacity is calculated assuming non-ideal power control.
3.2.2 Capacity assuming non-ideal power control In the remainder of this chapter the transformation: = 10 log (·) (·) 10
(3.12)
symbol describes the corresponding variable is used frequently. The hat (·) in the logarithmic scale. In order to account for the near-far effect it is required that the useful signal power from each user arrives at the same level at the BS. However, this would require ideal power control. In real systems power control inaccuracy cannot be avoided. This and excessive multi-path conditions are the reasons for a varying the bit-energy to interference ratio, εu , at the BS receiver. In order to maintain a certain frame error rate the resulting εu varies and the statistics can be approximated by a lognormal probability density function (pdf). Therefore, for an arbitrary user i, the bit-energy to interference
Interference and capacity analyses
59
ratio can be denoted as follows: εui = M
pg Puui
j:j=i
Puuj + Iou + N
,
(3.13)
where Puui is the received signal power from the desired user i. Analogously, Puuj is the signal power received from the other user j. The interest is on a service independent capacity. Therefore, it is part of the assumptions leading to (3.13) that the entire set of users (defined as the capacity) are permanently active. In an interference limited system such as a cellular CDMA network, the power level of the useful signal Puui at the receiver can be utilised to optimise capacity. This can be seen from (3.13). Since capacity is dependent on Puui the target power level at the BS can be used to enhance capacity provided that the increased power does not result in a significant increase of interference in the other cells. It has been demonstrated by (Veeravalli and Sendonaris, 1999) that Puui can be derived from (3.13) and becomes, Puui =
pg εui
+1
N + Iou ⎛
⎞ .
(3.14)
⎟ ⎜ M ⎟ ⎜ εuj ⎟ ⎜1 − ⎜ pg + εuj ⎟ ⎠ ⎝ j=1 U
From (3.14) it can be seen that a feasible solution for Puui exists only if U=
M j=1
εuj < 1. pg + εuj
(3.15)
This equation can be used to deduce the theoretical capacity maximum Mmax . It was demonstrated previously that εui can be approximated by a random variable (RV) which follows a lognormal distribution, which means that εui is Gaussian distributed. The mean of εui (μuε ) is primarily dependent on the receiver architecture and the forward error correcting (FEC) coding scheme. The standard deviation of εui , σεu , is dependent on the power control performance and the severity of the multi path fading. In (Viterbi and Viterbi, 1993) it is reported that σεu for the American IS-95 (Interim Standard-95) CDMA system is nominally 2.5 dB. Without loss of generality the M RVs, εu1,...,M , can be assumed to be independent and identically disεu
tributed (i.i.d.). It is straightforward to calculate the pdf of pg +i εu , which i was done in (Veeravalli and Sendonaris, 1999), but the sum of these RVs, U , as given in (3.15) involves numerical convolution.
60
Next Generation Mobile Access Technologies 30
σ =0.1 ε σε=0.5 σε=1.0 σε=1.5 σε=2.0 σε=2.5 σε=3.0
28 26 24 22 20
Outage [%]
18 16 14 12 10 8 6 4 2 0
3
4
5
6
7
8
Number of users
Fig. 3.1. Probability of outage as a function of simultaneously active MSs with: pg = 16, μuε = 3.5 dB and σεu as a parameter.
Outage is defined as the probability P r(U ≥ 1). The results are shown in Figure 3.1. In addition, the maximum number of users for less than 5% outage are summarised in Table 3.1. As expected, power control imperfections increase outage considerably. For instance, when increasing σεu from 0.1 dB to 3.0 dB and initially assuming six simultaneously active MSs the outage rises from approximately 0% to 15.5%. In contrast, in the case of only four active MSs outage increases from about 0% to 1% for the same variations of σεu . When assuming non-ideal power control, the pole capacity is not deterministic as it was in the case of ideal power control (see (3.3)). However, it is interesting to note that for both cases the pole capacity, Mmax , is independent of thermal noise and interference.
3.3 Adjacent-channel interference in a CDMA-TDD system In this section the effects of ACI on system capacity are investigated. A general description of ACI considering TDD properties is carried out in section 3.3.1. ACI is then investigated for two TDD deployment scenarios. In section 3.3.2 ACI is calculated assuming a single interfering cell. This
Interference and capacity analyses
61
Table 3.1. The theoretical upper capacity limit Mmax (assuming less than 5% outage) using the following parameters: pg = 16, μuε = 3.5 dB. σεu [dB]
Mmax
0.1
8
0.5
8
1.0
7
1.5
6
2.0
6
2.5
5
3.0
5
model is chosen to account for the fact that the TDD mode is ideally suitable to cover traffic ‘hot spots’ (Holma and Toskala, 2000, Chapter 12). The location and load of the interfering cell, the frame synchronisation and the adjacent-channel protection factor are varied. Initially, a simple model for the correlation of the desired signal power and the interference signal power, taken from (Viterbi et al., 1994), is assumed. In section 3.3.3 a multiple-cell environment is applied. The interfering network consists of a cluster of seven hexagonal cells. With this scenario in particular the impact of power control and handover margins are investigated. A model for the correlation of signal paths reported by (Klingenbrunn and Mogensen, 1999) is applied.
3.3.1 Characterisation of adjacent-channel interference In Figure 3.2†, a possible multi-operator interference scenario is depicted. A scenario of two adjacent carriers belonging to different operators is considered. Since network planning between independent operators can not be assumed cells can overlap randomly. Therefore, a transmitting entity and a receiving entity can be located in close proximity. The interference protection merely depends on the adjacent-channel protection factor. In the case of CCI, in contrast to ACI, neighbouring cells are ideally separated such that cells do not overlap. Hence, interference protection in the case of CCI is achieved by a spatial separation of the transmitter and the victim receiver. In addition, handover techniques can be used to circumvent high † Figures 3.2 and 3.3 are reproduced with permission from: (Haas and Povey, 1999).
62
Next Generation Mobile Access Technologies
Fig. 3.2. Adjacent TDD carriers can belong to independent operators (operator A and operator B) with the consequence that cells can overlap randomly. The c potential interference links with respect to the UL direction are shown. 1999 IEEE
CCI. Since the mechanisms that create CCI and ACI are different it seems obvious that the quantitative characteristics of both interference types are also different. Clearly, the focus in this section is to investigate adjacent channel interference applying to a CDMA/TDD interface. The mutual interference mechanisms in an FDD system are that MSs interfere with the neighbouring BSs, and vice versa. In this system the UL and the DL are separated in the frequency domain. Therefore, crosstalk between UL and DL is negligible. However, when considering the TDD system the complexity in terms of interference is increased since both the UL and DL are time multiplexed on the same carrier frequency. If the frames and time slots (TSs) of two cells are not synchronised additional interference scenarios occur. As compared to an FDD mode, in the TDD mode both the MSs and BSs can interfere with each other. Either interference link (MS↔MS or BS↔BS) can be characterised as interference between the same types of entity (MS or BS). Therefore, these interference paths inherent to a TDD system are henceforth called ‘same-entity interference’. Analogously, the interference scenarios MS↔BS and BS↔MS are specified as ‘other-entity interference’. Frame synchronisation in the TDD system has an impact on the quantity of same-entity interference and other-entity interference as
Interference and capacity analyses
63
can be found with the aid of Figure 3.3. This figure shows a possible TS toff
tslot
tslot
BSa
Ibb
Ibb
BSb
Imb Ibm
Ibm
Imb
MSa
Imm
Imm
MSb
t TX RX
c Fig. 3.3. Interference in a TDD system dependent on frame synchronisation. 1999 IEEE
arrangement of the scenario in Figure 3.2. The model is composed of four entities, BSa, MSa, BSb and MSb, where BSa and MSa, and BSb and MSb, respectively, form a communication link. Therefore, if BSa transmits MSa receives and vice versa. Since each cell may belong to a different operator the frames or TSs are unlikely to be aligned in time. This is modelled by an arbitrary time offset, toff . This time offset is normalised by the time slot (TS) duration, tslot , yielding the synchronisation factor: α=
toff . tslot
(3.16)
Due to the frame misalignment, BSa and BSb interfere with each other and thereby create Ibb . In the same way as there is interference between the BSs, MSa and MSb interfere with each other and generate Imm . Both types of interference are previously categorised as same-entity interference. It can be found that same entity interference is proportional to α. Similar properties can be found for other-entity interference except that it is proportional to (1 − α). Hence, as the synchronisation factor α increases other-entity interference diminishes, but same-entity interference increases, and vice versa.
64
Next Generation Mobile Access Technologies
This leads to two special cases: (a) if α = 1 only same entity interference exists and (b) if α = 0 only other entity interference exists. Case (b) emulates an equivalent FDD interface. The consequence is that interference is present during the entire receive period. Since other-entity and same-entity interference can be considered as independent (due to different interference sources) the magnitude of each type of interference can vary greatly. Therefore, it is interesting to find out whether it is possible to exploit the fact that same-entity interference and other-entity interference are different to minimise interference by altering frame synchronisation. This question is explored in depth in the following. Using the synchronisation mechanism introduced above, ACI in a TDD system, at an arbitrary location specified by its x and y coordinates can be expressed as follows: L Mj P cdj P cui,j 1 + (1 − αj ) , αj Iad (x, y) = κI ai (x, y) aj (x, y)
(3.17)
j=1 i=1
where L is the number of neighbouring cells taken into consideration, Mj is the total number of active users in the neighbouring cell j, P cui,j is the transmitted carrier power of user i in cell j, P cdj describes the total carrier power transmitted by BS j and ai (x, y) represents the path loss between the interfering user i and the location of interest (x, y). Similarly, aj (x, y) is the path loss between the location of interest and the BS of cell j. The synchronisation between cell j and the point of interest (x, y) is expressed by αj . The adjacent channel protection factor, κI , or adjacent channel interference ratio (ACIR) (3GPP, TSG, RAN, WG4, 1999) is determined by two factors: (a) one related to the transmitter filter and referred to as the adjacent-channel leakage ratio (ACLR) and (b) one related to the receiver filter and described as adjacent-channel selectivity (ACS). The relationship between ACIR, ACLR and ACS was investigated in (3GPP, TSG, RAN, WG4, 1999) and found to be: κI =
1 , (1/κL ) + (1/κS )
(3.18)
where κL is the ACLR and κS is the ACS. The transmitted powers P cui,j and P cdj are random variables that are determined by several factors, the most important of which are the location of the MSs, the path loss, the severity of lognormal shadowing, the handover algorithm, the power control algorithm and the receiver architecture. The path losses between the interferer and victim receiver, ai (x, y) and aj (x, y),
Interference and capacity analyses
65
are also random variables. Moreover, in the case of ACI the sink of interference and the desired receiver may be in close proximity so that the path loss on the desired link and interference link cannot be assumed to be uncorrelated. Frame synchronisation between different operators may also vary randomly, i.e. 0 ≤ αj ≤ 1. Due to its complexity the ACI power as described in (3.17) is calculated using Monte Carlo techniques. However, in order to verify the Monte Carlo model the pdf of the interference power is calculated analytically for simplified scenarios and the results are compared with those obtained by the Monte Carlo approach. If the pdf of interference at the BS location is known, it is possible to analyse the impact of interference on capacity using (3.11). The TDD mode is generally considered to be used in low-mobility environments and to cover ‘hot spot’ traffic areas (Holma and Toskala, 2000, Chapter 12). Therefore, in the following, a two-cell scenario (one cell per operator) is used to study the effects of ACI. This is in contrast to a multi-cell environment where handover techniques can be used to reduce interference significantly (Viterbi and Viterbi, 1994; Viterbi et al., 1994) and MSs are assigned to a single BS regardless of the path loss. It will be shown that this is not necessarily the ‘best’ scenario with respect to interference. 3.3.2 Single interfering cell In the case of a single interfering cell, the cell topology shown in Figure 3.4 is applied. The MSs are distributed uniformly within the cell area of the interfering cell. The pdf of ACI is calculated using the equation given in (3.17). The approach is twofold: first, Monte Carlo simulations are carried out. Second, a closed-form solution of the pdf is derived in order to validate the simulation platform as well as the analysis. 3.3.2.1 Propagation model The path loss is modelled according to the indoor propagation environment with no wall or floor losses (ETSI 30.03, V3.2.0 (1998-04), 1998): a = 37 + ϕ 10 log (d) + ξ [dB] ,
(3.19)
where ϕ is the path-loss exponent and d the distance between transmitter and receiver. Lognormal shadowing is modelled by ξ with standard deviation σ and zero mean. The BSs of the two cells that use adjacent carrier frequencies can be located in close proximity. The possibility that this scenario occurs is particularly high if the cells belong to different operators. In such a case, the
66
Next Generation Mobile Access Technologies
d1
d2
R
d0
φ BS separation distance
cell of interest (COI)
Interference link
Mobile Station
Desired link
Base Station
Fig. 3.4. A single cell causing ACI at a cell located a distance d0 from the cell of interest (COI).
desired signal and the interference signals experience similar multipath conditions, and both signals, therefore, cannot be considered uncorrelated. This mechanism is illustrated in Figure 3.5. In (Viterbi and Viterbi, 1994) a simple model of the signal correlation was introduced. In Viterbi’s paper a joint Gaussian probability density for losses to two or more base stations is assumed. The random component as a consequence of shadowing is considered to be composed of two components: (a) a component that is common to all BSs, and (b) a component that is unique to the receiving BS. The common component described in (a) is schematically depicted in Figure 3.5 by the grey shaded areas on the signal path. Thus, the lognormal shadowing on the propagation path to the ith BS is given as follows:
ξi = r1 υ + r2 υi ,
where
def
r12 + r22 = 1 ,
(3.20)
and υ is the random component common to the desired and the interference path and υi is the uncorrelated random component on each path. Further-
Interference and capacity analyses
67
interference path
interference path
desired path
desired path
Fig. 3.5. The correlation of the desired and the interference signal is dependent on the location of the transmitter relative to the first and second receiver.
more it is defined that: E(ξi ) = E(υ) = E(υi ) = 0 ,
(3.21)
2
for all i ,
(3.22)
E(υ, υj ) = 0
for all i ,
(3.23)
for all i and j, i = j .
(3.24)
Var(ξi ) = Var(υ) = Var(υi ) = σ E(υi , υj ) = 0
Using the presuppositions (3.21)-(3.24), the correlation coefficient yields: def
= 0
def
= 0
E[(ξi − E[ξi ]) (ξj − E[ξj ])] E(ξi , ξj ) = , r= σ2 σ2 def
= 0
def
= 0
(3.25) def
= 0
r12 E(υ 2 ) + r1 r2 E(υυj ) +r1 r2 E(υυi ) +r22 E(υi , υj ) = , σ2
(3.26)
r12 Var(υ) = , σ2 = r12 = 1 − r22 .
(3.27)
σ2
(3.28)
The joint pdf between the random component on the desired link ξi and
68
Next Generation Mobile Access Technologies
the random component on the interference link, ξj , yields (Papoulis, 1991, Chapter 7): (ξi − r ξj )2 1 , (3.29) exp − p(ξi | ξj ) = 2 σ 2 (1 − r2 ) σ 2 π (1 − r)
In (Viterbi et al., 1994) a constant value of r1 = 0.25 is assumed. This is a reasonable assumption since in (Viterbi et al., 1994) only co-channel interference is investigated, and therefore the cell topology is fixed. In the analysis here, the BS separation distances can vary from co-location to several times the cell radius. The following relationship between r1 and the BS separation distance is introduced: ⎧ ⎨ 1 − d0 if d ≤ R , 0 R (3.30) r1 = ⎩0 otherwise .
where R is the cell radius and d0 the BS separation distance (see Figure 3.4). From substituting (3.30) into (3.28) it can be seen that the correlation coefficient equals one if both cells are co-sited. This is obvious because the desired path and interference path are exactly the same. The signals are considered to be uncorrelated if the BS separation distance equals the cell radius. It is recognised that there is little experimental verification of this model, but it is a logical extension of reports on the correlation of shadow fading (e.g. as given in (Gudmundson, 1991)). In section 3.3.3 a correlation model reported by (Klingenbrunn and Mogensen, 1999) is applied which also takes into account the angle of arrival difference of the signals. This model has been a subject of greater experimental verification. Monte Carlo simulations are carried out assuming a spatially uniform user distribution to obtain the pdf: p(Iad ) at the BS of the cell of interest (COI). The results of the Monte Carlo approach are verified against the analytical derivation of p(Iad ). The pdf, p(Iad ), can be used to study interference properties for various cell topologies and applications in which handover techniques are not applied. Therefore, this pdf can also be used to investigate the feasibility of the TDD underlay described in Chapter 2. 3.3.2.2 Power-control models In the UL, ideal power control is assumed. In the DL a similar model as d in (Holma et al., 1999) is used. First, the code power, P˜cj , of the MS j which experiences the greatest path loss is calculated. The same code power is then d d applied to each user within the same TS, P˜ci = P˜cj with i = 1, · · · , M . This ensures that the required bit-energy to interference ratio is fulfilled for
Interference and capacity analyses
69
d all M users as shown in the following: Let the maximum code power, P˜cj , be determined by the MS for which the maximum path loss, aj , applies. The bit-energy to interference ratio, εdj , at the respective MS results in:
εdj
=
d P˜cj pg d τ (M − 1) P˜cj + aj (Iad + N )
,
(3.31)
where τ is the orthogonality factor. Since: ai < aj
for all
i = j ,
(3.32)
εdi > εdj
for all
i = j .
(3.33)
it follows that:
Hence, it is ensured that for each MS the required bit-energy to interference ratio, εd , is greater than a minimum threshold. 3.3.2.3 Analytical derivation of the pdf of ACI In this subsection a closed-form expression for the pdf of ACI at the BS of the cell of interest is derived. Circular cells instead of hexagonal cells are considered in order to simplify the analysis. Also, the pdf is calculated assuming one uniformly distributed interfering user. The results obtained can be easily extended to multiple interfering users by appreciating that the location of users are independent from each other. In addition, the interference is calculated for ideal synchronisation of TSs.
Power
Pc
d2
d1 d0
BS1
Iad
Pc Pr a1
Pr φ
BS0
Pr Interference link Desired link
Pc κI a2
Iad
Location BS0
MS
BS1
Mobile Station (MS) Base Station (BS)
Fig. 3.6. Simulation model to derive interference pdf at the BS of a neighbouring c cell. 2004 IEEE
70
Next Generation Mobile Access Technologies
In Figure 3.6† the relationship between the transmitted code power (Pc ), received power (Pr ) and interference power (Iad ) is depicted. Using the standard channel model reported in (ETSI 30.03, V3.2.0 (1998-04), 1998), the interference power Iad can be denoted as follows: γ ξ D −ξ I −κ I d1 10 10 , (3.34) Iad = Pr d2 where ξD is the random component due to shadowing on the desired link and similarly, ξI is the shadowing component on the interference link, d1 is the distance between the MS and the BS to which it is connected, γ is the path-loss exponent, d2 is the distance between the MS and the victim BS and κI is the adjacent channel interference ratio (ACIR) in dB. The distance d2 can be expressed as a function of the polar coordinates of the MS, d1 and φ, and the BS separation distance d0 : d2 = d21 + d20 + 2 d1 d0 cos(φ) . (3.35) Substituting (3.35) into (3.34) and using the substitution β = ln(10)/10 yields: −0.5γ 2
d0 d0 × +2 cos(φ) Iad = Pr 1 + d1 d1 exp [ β (ξD − ξI − κI )] . (3.36) The distance d0 is a constant and d1 is a random variable. The following substitution is introduced: d0 . (3.37) v= d1 The aim is to obtain the pdf of Iad in logarithmic units. Therefore, (3.36) is transformed as follows: γ ln v 2 + 2v cos(φ) + 1 . (3.38) Iad = Pr + ξD − ξI −κI − 2β ξ
It can be found that (3.38) involves three independent random variables: v, ξ = ξD − ξI and φ. The pdf of these random variables is calculated in the following. Since it is assumed that the MS is uniformly distributed the pdf, p(d1 , φ), can be derived by calculating and dividing appropriate cell areas, p(d1 , φ) =
d1 , R2 π
(3.39)
† Figures 3.6 and 3.7 as well as Tables 3.2, 3.3 and 3.4 are reproduced with permission from: (Haas and McLaughlin, 2004).
Interference and capacity analyses
71
where R is the cell radius. As a consequence of the substitution used in (3.37), the pdf, p(v, φ), yields: p(v, φ) =
p(d1 , φ) d (v(d1 )) d( d1 )
=
d20 −3 v , π R2
d0 ≤ v ≤ ∞. R
(3.40)
d1 =f −1 (v)
The pdf of √ ξ = ξD − ξI is a normal pdf with zero mean and standard devi˜ , where σ ˜ is the standard deviation of lognormal shadowing ation, σ = 2 σ between transmitter and receiver:
ξ2 1 exp − 2 . (3.41) p(ξ) = √ 2σ 2 πσ In order to derive the final pdf of ACI, a random variable transformation system of third order has to be solved which is carried out by using auxiliary functions (Papoulis, 1991, Chapter 6): y = f (φ) = φ ,
(3.42)
w = f (ξ) = ξ .
(3.43)
Using (3.42), (3.43) and (3.38) the Jacobian required to solve the random variable transformation system can be established:
J(v, ξ, φ) =
∂ Ia d (v,ξ,φ) ∂v ∂y(v,ξ,φ) ∂v ∂w(v,ξ,φ) ∂v
∂ Ia d (v,ξ,φ) ∂ξ ∂y(v,ξ,φ) ∂ξ ∂w(v,ξ,φ) ∂ξ
∂ Ia d (v,ξ,φ) ∂φ ∂y(v,ξ,φ) ∂φ ∂w(v,ξ,φ) ∂φ
∂ Ia d (v,ξ,φ) ∂v
∂ Ia d (v,ξ,φ) ∂ξ
∂ Ia d (v,ξ,φ) ∂φ
,
(3.44)
,
(3.45)
−γ (v + cos(φ)) ∂ Iad (v, ξ, φ) . = ∂v β (v 2 + 2 v cos(φ) + 1)
(3.46)
which yields: J(v, ξ, φ) =
0 0
0 1
1 0
and thus, J(v, ξ, φ) = −
The pdf of Iad can be denoted as follows: p(Iad ) =
!
2π
0
!
∞
Q p(vi , φ) p(ξ)
−∞ i=1
|J(vi , ξ, φ)|
dξ dφ . v i =fi−1 (Ia d ,ξ,φ)
(3.47)
72
Next Generation Mobile Access Technologies
The inverse function of (3.38), vi = fi−1 (Iad , ξ, φ), results in i = 1, · · · , (Q = 2) solutions, as follows: ⎧ ⎪ ⎪ ⎨− cos(φ) + (−1)i √χ vi (Iad , ξ, φ) = u ⎪ ⎪ ⎩ 0
where
χ=
#
cos2 (φ)
if χ ≥ 0 and d0 i √χ ≤ ∞, R ≤ u + [−1]
otherwise,
$
% 2 β u Iad − ξ − Pu + κI + exp − −1 γ
(3.48)
(3.49)
In the following, (3.40), (3.41) and (3.46) are substituted into (3.47): $ % ! 2π ! ∞ β d20 ξ2 √ = ζ exp − 2 dξ dφ, (3.50) p(Iad ) 2σ γ R2 σ 2 π 3 0 −∞ where
ζ=
2 i=1
−
vi2 + 2 vi cos(φ) + 1 . vi3 [vi + cos(φ)]
(3.51)
The final pdf, p(Iad ), can be found by substituting the value of vi from (3.48) into (3.51). This pdf can be used to examine many different interference problems in cellular communications. For example, it has been used by the author to investigate the feasibility of a TDD underlay (Haas and Povey, 1999), as described in Chapter 2. The pdf (3.50) is calculated numerically and the results are used to verify the pdf calculated using the Monte Carlo approach. The parameters used for the verification are summarised in Table 3.2. 3.3.2.4 Comparison of analysis approaches The comparison is carried out for different parameters in order to obtain sufficient evidence as to whether both approaches lead to similar results. First, the pdf, p(Iad ), is calculated for two different BS separation distances, d0 . Second, the standard deviation of lognormal shadowing, σ, is varied for each deployment scenario. The results for the mean and the standard deviation of p(Iad ) for d0 = R are summarised in Table 3.3, and similarly, the results for d0 = 2R are shown in Table 3.4. It is found that the results of both methods are extremely similar. The mean differs by a maximum 0.1% and the standard deviation varies by about
Interference and capacity analyses
73
Table 3.2. Simulation parameters used in the verification of the analytically derived pdf of ACI with results obtained by Monte Carlo c simulations. 2004 IEEE Parameter
Value
BS separation distance, d0
100 m
Cell radius, R
50 m
Path loss exponent, ϕ
3.0
ACIR, κI
30 dB
Desired signal power, Puu
-111 dBm
Monte Carlo runs
10,000
Table 3.3. Comparison of mean and standard deviation of Iad for d0 = R. c 2004 IEEE Monte Carlo simulation Var Iad E Iad -147.47
9.96
Analytical calculation Var Iad E Iad -147.53
9.86
4
-147.45
10.60
-147.47
10.46
8
-147.32
12.72
-147.45
12.53
16
-147.47
18.47
-147.43
18.66
Standard deviation σ [dB] 2
1%. The variations tend to increase for a small σ, which is obvious as in this case the pdf is dominated by the cell geometry rather than the lognormal shadowing, and thus a difference is inevitable due to the use of hexagonal cells for the Monte Carlo simulations and a circular cell in the analytical derivation. The results also reveal some interesting properties. For example, the standard deviation for σ = 2 and σ = 4 does not differ greatly (between 5% and 10%). In this case the pdfs are primarily dominated by the cell geometry and the random user distribution. In contrast, for σ = 8 and σ = 16 the variance of p(Iad ) changes significantly (between 50% and 65%) from which it can be inferred that shadowing is the dominating random factor in this
74
Next Generation Mobile Access Technologies
Table 3.4. Comparison of mean and standard deviation of Iad for d0 = 2 R. c 2004 IEEE Monte Carlo simulation Var Iad E Iad
Analytical calculation E Iad Var Iad
2
-156.28
7.66
-156.44
7.33
4
-156.22
8.38
-156.44
8.11
8
-156.32
10.85
-156.40
10.63
16
-156.26
17.62
-156.44
17.46
Standard deviation σ [dB]
case. Furthermore, the expected value is constant when varying σ. This is anticipated since the mean of lognormal shadowing is zero and independent of the cell geometry and user distribution. It is interesting to note that the expected value increases by about 9 dB when the BSs are separated by twice the cell radius instead of only the cell radius itself. This means that ACI can increase about eight times within the observed interval of BS separation distances which points towards potential problems that may be caused by ACI since, as described previously, cell planning cannot be assumed. In addition to the results in Table 3.3 and Table 3.4 the respective pdfs and cumulative distribution functions (cdfs) of p(Iad ) are depicted in Figure 3.7. When comparing the pdfs in Figure 3.7(c) with the pdfs in Figure 3.7(d) it can be found that for d0 = R, i.e. when the BS of the interfering cell is located at the cell boundary of the victim cell, the pdfs are skewed towards greater values of Iad . This can be explained by the fact that the MS can be in close proximity to the victim BS. In this case, the high transmitted power (due to the location at the cell boundary) results in great ACI. In contrast, for d0 = 2 R the opposite behaviour can be observed, i.e. the tails of the pdfs for greater values of Iad converge to zero more rapidly than for lower values of Iad . The reason for this is that the distance between the interfering MS and the victim BS is always more than the cell radius. This has a significant impact on interference. In the case of d0 = R and σ = 8, the probability that the interfering signal power is greater than, for example, -130 dBm is still about 10%, whereas for d0 = 2R, the probability that Iad > -130 dBm is only about 1%.
Interference and capacity analyses Monte Carlo sim Analytical approach
0.9
Cumulative density function
0.5
0.4
σ=4 0.6
0.5
0.4
0.3
0.3
0.2
0.2
0.1
0.1
180
170 160 150 140 Adjacent channel interference power [dBm]
130
200
120
(a) Cdf of Ia d for d0 = R.
0.06
σ = 16
0.7
Cumulative density function
σ = 16
σ=4
190
σ=8
0.8
0.7
0.6
Monte Carlo sim Analytical approach
0.9
σ=8
0.8
200
75
1
1
0.08
Monte Carlo sim. Analytical approach
0.07
190
180
170 160 150 Adjacent channel interference power [dBm]
140
130
(b) Cdf of Ia d for d0 = 2 R. Monte Carlo sim. Analytical approach
0.05
Probability density function
Probability density function
0.06 0.04
σ=4 0.03
0.02
σ=8
0 240
220
200
σ=4 0.04
0.03
0.02
σ = 16
0.01
0.05
0.01
180 160 140 120 100 Adjacent channel interference power [dBm]
(c) Pdf of Ia d for d0 = R.
80
60
40
0 250
σ=8 σ = 16 200
150 100 Adjacent channel interference power [dBm]
50
(d) Pdf of Ia d for d0 = 2 R.
Fig. 3.7. A comparison of the cdfs and pdfs obtained by the analytical approach c in (3.50) with the results of Monte Carlo simulations. 2004 IEEE
3.3.2.5 Capacity results It has been shown for a simplified scenario that the results of the Monte Carlo approach and the results of the analytical model do not differ significantly. This gives sufficient confidence to extend the Monte Carlo model to investigate the impact of different frame synchronisations on ACI. The cdfs of ACI caused by the transmitted code powers of the MSs and BS are calculated using the parameters in Table 3.5. The 95th percentile of interference is chosen to be used in the further capacity analyses. This means that a capacity reliability of 95% is guaranteed, i.e. in 95% of all cases the capacity is higher than the one shown. The interference results for a user population of four MSs are depicted in Figure 3.8. The interference caused by the BS is shown by dotted lines, whereas the interference resulting from the MSs are indicated by solid lines.
76
Next Generation Mobile Access Technologies
Table 3.5. Simulation parameters for ACI analysis. Parameter
Value
Cell radius, R
50 m
Bit rate
16 kbps
Chip rate
3.84 Mcps
Standard deviation of lognormal shadowing, σ
10 dB
Receiver noise figure
5 dB
Max. MS TX power
15 dBm
Max. BS TX power
24 dBm
Bit-energy to interference ratio, εu
3.5 dB
Tolerable outage, Pout
5%
Note that when the interfering cells completely overlap, the MS→BS interference is lowest (due to power control and high cross-correlation of lognormal shadowing between the interference and desired path), but the BS→BS interference is highest. As the BSs are separated, the BS→BS interference is decreasing monotonically and at the same time the MS→BS interference is growing until the BS separation is approximately the cell radius. The reason for the MS→BS interference peak is due to the high transmission powers of MSs at the outer regions of the cell, which causes an increase in MS interference when the victim BS is moved towards the cell boundary. However, the MS→BS interference diminishes in a single cell scenario as the cells move further apart. A highly synchronised transmission, α = 0.01, compared to opposed transmission, α = 0.99, results in greater MS→BS interference but lower BS→BS interference. However, despite the almost ideal synchronisation of α = 0.01 (1% synchronisation error) the interference power, Ibb , for a BS separation distance of about 7 m is still -105 dBm – in comparison, the useful signal is -109 dBm. This means that the interference power from the close-by BS using the adjacent carrier is about 2.5 times greater than the useful signal power. Clearly, this renders the co-location of BSs difficult. The most relevant discovery in this chapter is shown in Figure 3.8(b) where the total ACI power is depicted. Notice that for certain conditions it is advantageous to apply opposed synchronisation of TSs (α = 0.99) rather than synchronous transmission and reception (α = 0.01). This is highlighted by the circle in Figure 3.8(b). It will be shown that the same effect can
Interference and capacity analyses
77
(a) ACI power from MSs (Im b ) and BS (Ib b ).
(b) Accumulated ACI power, Ia d = Im b + Ib b .
Fig. 3.8. ACI power assuming four active interfering users.
also be observed when considering a multiple cell model as will be done in section 3.3.3. Therefore, this fundamental discovery is further exploited by the development by dynamic channel allocation (DCA) algorithms presented in chapters 4 and 5. In Figure 3.9 the impact of ACI on capacity is shown. The capacity is calculated using (3.11) with other cell interference, Iou , being replaced by ACI, Iad . When interpreting the capacity results, note that the outage threshold is 5% which means that in 95% of all user distribution scenarios the capacity is better than or equal to the actual figures presented. The cell load factor in (3.11) determines the useful signal, Puu , at the BS in the COI. A linear increase of χ means that the useful signal Puu increases exponentially with a singularity at χ = 1, which can be found by using (3.5).
78
Next Generation Mobile Access Technologies
(a) χ = 0.5, κ I = 30 dB
(b) χ = 0.75, κ I = 30 dB
(c) χ = 0.5, κ I = 35 dB
(d) χ = 0.75, κ I = 35 dB
Fig. 3.9. Cell capacity with four interfering MSs. The relative capacity is shown for different cell load factors (χ) and different ACIR factors (κI ).
With the given parameters the pole capacity yields Mmax = (pg/εu ) + 1 = (16/100.35 ) + 1 ≈ 8. Therefore, the admission control of the system will restrict the maximum cell load to less than eight users in order to prevent outage of users that are not able to achieve the increased Puu target at the BS receiver (usually MSs at outer regions of the cell). In Figure 3.9 the cell load is assumed to be four MSs which is equivalent to χ = 0.5. If for the same actual user population, the cell load factor in (3.11) is increased, this has the equivalent effect of a greater number of active MSs, which inherently means an increased signal power Puu . Thus, by keeping the user population constant, but increasing χ, it is possible to emulate dynamic power control as a means of coping with ACI. If, for example, in the case of four active MSs the cell load factor, χ, is increased from 0.5 to 0.75, the useful signal level increases by a factor of three (4.77 dB).This can be found using (3.5). The results in Figure 3.9(b) and Figure 3.9(c) reveal that the increase of Puu
Interference and capacity analyses
79
by a factor of three has a similar effect as an increase of the ACIR, κI , by 5 dB. This is an expected result and thus helps to validate the entire model. As mentioned before, there is a potential disadvantage when increasing Puu which is that the coverage area can be affected. This yields the well known capacity-coverage trade-off in CDMA systems (Veeravalli and Sendonaris, 1999). An extended investigation was carried out in (Haas et al., 2000c), assuming UTRA parameters, but the results are omitted in this book as the focus is merely on the relationship between interference and capacity. Another interesting finding is that in all cases, the capacity for α = 0.01 has a local maximum at a relative BS separation of about 0.2–0.5. This effect can explained with the aid of the interference graphs depicted in Figure 3.8(b). The ACI power for α = 0.01 is primarily determined by Imb , but for small BS separations the interference contribution from the BS, Ibb , is significant which finally results in high interference for BS separations of 0–0.2 times the cell radius. The interference from the BS diminishes rapidly before Imb starts increasing at a BS separation of 0.6–1.0. This effect of Ibb and Imb having their maximum at a relative BS separation of 0 and 1.0 respectively, results in a local minimum of Iad and consequently in a local maximum of the relative capacity M/M0 . This mechanism is inherent to a TDD system since in a FDD system only Imb has to be considered and therefore a similar local maximum does not exist. Co-siting is only feasible for κI = 35 dB and α = 0.01 without a significant capacity loss. In this case the capacity is between 59% and 85% dependent on Puu adjustments. The power adjustments are most effective for relative BS separations of 0.6–1.4. For a relative BS separation of 0.95 and κI = 30 dB the gain from increasing Puu by a factor of three is about 60%, whereas the gain is only about 10% for a relative BS separation of 0.4. The corresponding capacity increases from 85% to 95%. In Figure 3.10 the results of the relative capacity for six interfering MSs are presented. In this case an ACIR of 30dB and TS synchronisations of α > 0.1 prohibit connections at the respective TSs in the victim cell. This situation is improved significantly if the ACIR is increased to 35dB. In this case the capacity improvement is highest and is about 50% for a BS separation range of 0.7–1.0 times the cell radius. If, however, the BS’s separation is less than 0.4 times the cell radius and α > 0.1 it is still not feasible to use the same TSs at the COI when assuming a tolerated outage of 5%. Despite an almost ideal synchronisation, α = 0.01, an ACIR of 30 dB results in a significant capacity reduction within a relative BS separation range of 0–0.2 and 0.7– 1.0. From these results and the results in Figure 3.9 it can be found that within the cell boundaries of the interfering cell, the best BS separation
80
Next Generation Mobile Access Technologies
(a) κ I = 30 dB
(b) κ I = 35 dB
Fig. 3.10. Relative cell capacity with six interfering MSs. The cell load factor, χ, is 0.75.
is 30–50% of the cell radius provided that both cells transmit and receive synchronously, i.e. α < 0.01. It is recognised that the condition of α < 0.01 is difficult to satisfy, in particular, if the cells belong to two independent operators. Clearly, this case requires an ACIR greater than 35 dB, but there is obviously a trade-off between the costs associated when increasing the ACIR and the advantages obtained. In the following section the scope of investigation is extended to consider a cluster of interfering cells.
3.3.3 Multiple interfering cells One reason as to why a CDMA system has the potential to achieve greater flexibility than a TDMA or FDMA system is that in a cellular environment a frequency or TS reuse distance of 1 can be applied (Lee, 1991; Viterbi, 1995a). Despite the greater interference and the accompanying effects on capacity, the system can be operated to allow high capacity and at the same time generate great flexibility. The disadvantage of CDMA systems is that power control and handover techniques have a vital impact on interference (Viterbi et al., 1994; Viterbi and Viterbi, 1994; Wong and Lim, 1997). This holds for CCI as well as ACI. Therefore, we will examine ACI in a cellular environment using different power-control algorithms and handover margins. Two different DL power-assignment algorithms are applied: the power-assignment method as described on page 68 and a C/I-based DL power-control algorithm (Seidenberg et al., 1999). The cell topology as shown in Figure 3.11† is applied. The location of the † Figure 3.11 is reproduced with permission from: (Haas et al., 2000b).
Interference and capacity analyses
150
81
Base station Mobile station COI Cell of interest
100
R d0
50
d0
0
=
0 COI
50
100
150 150
100
50
0
50
100
150
Fig. 3.11. Multiple cells causing ACI at a cell located on top of the interfering cell c cluster. 2000 IEEE
COI is varied along the d0 -axis within the range 5 m ≤ d0 ≤ R.
3.3.3.1 Propagation model The path-loss model given in (3.19) is used. In the analyses of cellular networks it is usually assumed that lognormal shadowing on the propagation path is uncorrelated for different propagation paths. However, measurements have shown that shadowing on the desired link and on the interference link can be highly correlated (see Figure 3.5). The correlation of lognormal shadowing has been subject to many investigations by different researchers (Abu-Dayya and Beaulieu, 1991; Gudmundson, 1991; Klingenbrunn and Mogensen, 1999) In the interference analysis conducted in this book the model reported by (Klingenbrunn and Mogensen, 1999) is adopted. This paper reports that the correlation coefficient is primarily dependent on the relative distance difference of the receivers and the angle-of-arrival difference (AAD). It was found in (Klingenbrunn and Mogensen, 1999) that the AAD dependency has the largest impact on the correlation coefficient of two signal paths. The correlation coefficient is computed as follows: first, the relative difference
82
Next Generation Mobile Access Technologies
between desired path, d1 , and interference path, d2 , is calculated as follows:
d1 A = 10 log10 [dB] . (3.52) d2 Next, a threshold, X, is introduced which determines the location when the distance dependency of the correlation coefficient reaches its minimum: & A if A ≤ X, 1− X (3.53) f (X, A) = 0 otherwise. Finally, the correlation coefficient is obtained as: |φ| + 0.4 if |φ| ≤ 60, f (X, A) 0.6 − 150 r(φ, A) = 0 otherwise,
(3.54)
Note that the minimum correlation coefficient is non-zero so as to account for local scattering around the receiver. It is reported in (Klingenbrunn and Mogensen, 1999) that the threshold X is in the range of 6–20 dB. It is set to 6 dB in this evaluation because the correlation model is used in an indoor environment with a rapid change of the propagation conditions due to walls, doors and interior. 3.3.3.2 User distribution and handover The total ACI power is dependent on the transmitted powers on the adjacent carrier. Therefore, methods such as handover in the interfering network reduce ACI. The significance of handovers are demonstrated by Chebaro (Chebaro and Godlewski, 1992). In these papers it was shown that the allocation of a mobile to the closest BS rather than to the BS that offers the smallest signal attenuation can create interference that is between 4 and 20 times higher. Note, that these results were obtained for an FDD system in which only the MSs contributed to interference at the BS and a cell reuse distance of 1 was assumed. In a TDD system this effect can be more significant since the UL and DL use the same radio-frequency carrier. The severity of this problem in TDD with respect to ACI is investigated by considering handover regions as depicted in Figure 3.12. In the case that handovers are assumed, an MS that is located within the grey shaded areas chooses the best out of three BSs. The handover areas are determined by a circle with radius d1max . The most significant criterion of the selection process is the lowest path loss. In order to avoid ping-pong effects, however, handover algorithms such as the IS-95A algorithm (Holma and Toskala, 2000) incorporate a fixed handover margin. The basic mechanism of a handover margin is explained with the aid of Figure 3.13. The path loss from an arbitrarily
Interference and capacity analyses
83
Mobile station Base station
R
d1 max d0
Fig. 3.12. Handover model: within the grey shaded areas an MS is located to the best serving BS.
located MS to a set of its closest BSs (BS0 , BS1 , . . . , BSn ) is shown. It is assumed that the MS is located within the cell 0. Then the MS is served by the BS0 if, and only if, a0 ≤ ai + δ for all i = 0, 1, . . . , n. The handover margin, δ, represents the additional path loss that can be tolerated until a handover process is executed to the cell offering the lower path loss. The effects of the handover mechanism can be examined by calculating the minimum coupling loss (MCL) between the BS and MS. The results of the MCL assuming a uniform user distribution and the path loss as given in (3.19) are presented in Figure 3.14. The handover mechanism reduces the probability of high coupling losses significantly. For example, the probability that the coupling losses are less than 70 dB is about 8% if handovers are not used and about 17% when implementing handovers with a 5 dB handover margin. This results in a difference of 9%. For comparison, the difference for 90 dB MCL is 30%. This shows how that the probability of great coupling losses are reduced considerably.
84
Next Generation Mobile Access Technologies
Path loss a0 a2
δ
BS number BS0
BS1
BS2
BSn
Fig. 3.13. The MS is assigned to BS0 . The handover margin, δ, is used to model situations where a MS is not necessarily allocated to the BS that offers the lowest path loss (BS2 in this case). 1
δ = 0.1 dB 0.9
δ = 5 dB 0.8
Cumulative density function (cdf)
No handover 0.7
0.6
0.5
0.4
0.3
0.2
0.1
0 30
40
50
60
70
80 90 Path loss [dB]
100
110
120
130
Fig. 3.14. Minimum coupling losses assuming different handover thresholds.
3.3.3.3 Interference results Simulations are conducted using the parameters given in Table 3.5. For reasons of comparison the results in Figure 3.15 show the probability of ACI when handovers are used (solid curves) and when handovers are not used (dotted curves). As can be seen, handovers decrease the probability of ACI significantly. In the case of BS interference and with d0 = 10 m, the proba-
Interference and capacity analyses Interference from MSí s
1
1
0.9
0.9
0.8
0.8
0.7
0.7 Cumulative density function
Cumulative density function
Interference from BS
0.6
0.5
0.4
d =10m 0 d =20m 0 d =30m 0 d =40m 0 d =50m
0.3
0.2
0.1
0.6
0.5
0.4
d =10m 0 d =20m 0 d =30m 0 d =40m 0 d =50m
0.3
0.2
0.1
0
0 150
140
130
120 110 100 90 Adjacent channel interference power [dBm]
85
80
(a) Probability of ACI from BSs (α = 1)
0
70
0 150
140
130
120 110 100 90 Adjacent channel interference power [dBm]
80
70
(b) Probability of ACI from MSs (α = 0)
Fig. 3.15. ACI distribution assuming four simultaneously active MSs in each interfering cell. Ideal power control in the UL and the DL power-control algorithm given in (3.31) is assumed. The dotted lines represent the cdf of ACI for the case that handovers are not considered. The solid graphs depict the probability of ACI considering handover with a threshold of 5 dB. The ACIR, κI , is 35 dB.
bility that Ibb is less than the thermal noise power (about -102 dBm) reduces from 15% to 3%. For the same cell configuration the probability of ACI from the MSs, Imb , when permitting handovers resembles the distribution of ACI when disallowing handovers (Figure 3.15(b)). This effect is anticipated since the location of the victim cell is very close to an interfering BS. Since all MSs are power controlled to their respective BS the signal power of each MS is about the same at the BS receiver. Therefore, if a victim receiver is located close to a BS, the signal powers from the MSs do not vary greatly and are independent of the actual transmitted power of the MS. Hence, the reduction in the transmitted power due to handovers does not greatly affect the ACI interference from the MSs. Therefore, it becomes clear that the total interference from the MSs is minimal for co-siting of both BSs. However, the total interference from BSs is highest for co-siting and reduces as the BSs are moved apart. These mechanisms are documented in Figure 3.15. If the victim BS is located at a cell corner at d0 = 50 m the interference from the MSs is maximal – the probability that Imb is greater than the thermal noise power when considering handovers is only about 1% and increases marginally to about 2% when handovers are omitted. Given that two independent networks using an adjacent carrier are not synchronised such that cells receive and transmit at the same time, interference between BSs cannot be avoided. When comparing the worst-case probabilities of Imb and Ibb being less than the thermal noise power an im-
86
Next Generation Mobile Access Technologies
portant statement can be made: it is more advantageous to locate BSs at the cell corners of the coexisting network rather than at the same site. This finding is significant as it says that the freedom in cell planning of a TDD network is strongly limited by coexisting networks that make use of an adjacent carrier. In particular, the cost effective option of using the same location for the BSs is the worst option if the ACIR at the BS and the MS is equal. A further significant finding is that the interference from the BSs, Ibb , (Figure 3.15(a)) is affected by handovers to a greater extent than the interference from the MSs, Imb , (Figure 3.15(b)). As an example, if we let d0 be 30 m, then ACI from the MSs is in 90% of all cases less than -120 dBm assuming handovers. This threshold increases to about -118 dBm for the case that handovers are not considered. The situation for Ibb and the same cell topology is that in 90% of all cases the interference is less than -119 dBm with handovers carried out, but -110 dBm where there are no handovers. The effect that Ibb decreases by 9 dB whereas Imb only reduces by 2 dB if handovers are used highlights the significance of handovers in such a TDD system.
(a) Probability of ACI from BSs (α = 1).
(b) Probability of ACI from MSs (α = 0).
Fig. 3.16. ACI distribution assuming four simultaneously active MSs in each interfering cell, C/I-based power control algorithms in the UL and DL, non-ideal power control and handovers (solid graphs). For comparison the dotted curves show the results for ideal power control, a simple DL power-control algorithm and handovers (the same as the dotted curves in Figure 3.15). .
So far, however, the same DL power-control algorithm as used in the single-cell model and given in (3.31) is applied. This algorithm does not minimise the required code power for each MS. In a multiple-cell environ-
Interference and capacity analyses
87
ment the actual DL powers can considerably affect the performance in the neighbouring cells (Prasad et al., 1993). Therefore, C/I-based power-control algorithms are used in such environments (Seidenberg et al., 1999). On the one hand, it is anticipated that Ibb can be reduced using a more precise DL power-control algorithm. On the other hand, the assumption of ideal power control in the UL and DL (all signals arrive at the same level) is not completely realistic as demonstrated in (Viterbi and Viterbi, 1993). Therefore, simulations are conducted assuming a C/I-based power control algorithm and non-ideal power control in both direction. The standard deviation of the lognormally distributed bit-energy to interference ratios, σεu , and σεd respectively, are assumed to be 2.5 dB (data taken from (Viterbi and Viterbi, 1993). For convenience the two cases are numbered as follows: (A) This is the case of ideal power control with handovers being considered (with an handover threshold of 5 dB) and the same DL power control algorithm as in the single-cell scenario. (B) This is the case of non-ideal power control with a standard deviation σεu = σεd = 2.5 dB and C/I-based power control algorithms and considering handovers with a threshold of 5 dB. The results in Figure 3.16 underline the trade-off described previously. This trade-off is most obvious in (B) (solid curves) where the interference from the MSs is greater than in (A) (dotted curves), but this effect is reversed for the case of interference from the BSs. Therefore, from the latter it can be concluded that for the parameters used the impact of the C/I-based power control algorithm outweighs the effect of power control imperfections. In the case of interference from the MSs, of course, the C/I-based power control algorithm does not have an impact and thus the interference situation in (B) is worse than in (A). This can be expressed quantitatively as follows: assuming an interference threshold with the property that, for example, in 90% of cases the interference shall be less than the given threshold, the interference from MSs in (B) increases by about 3 dB for all BS locations observed. For the same scenario the interference from BSs decreases by about 1 dB. Thus, the capacity results assuming (A) underestimate Imb and overestimate Ibb with respect to (B). Capacity results In Figure 3.17 the relative remaining capacities in the COI are presented using (3.11), which assumes ideal power control within the COI. The scenarios described in (A) and (B) are investigated. The following properties become apparent:
88
Next Generation Mobile Access Technologies 90
100
80
90
70
80
70 Remaining capacity [% ]
Remaining capacity [% ]
60
50
40
60
50
40
30
10
0
30
α=0.01 α=0.1 α=0.5 α=0.90 α=0.99
20
5
10
15
20
25 30 distance x [m]
35
40
α=0.01 α=0.1 α=0.5 α=0.90 α=0.99
20
10
45
0
50
5
10
15
20
(a) ACIR: 25 dB. 100
90
90
80
80
100
50 95
40 90
80
10
0
5
10
5
15
10
15
20
20
25 30 distance x [m]
35
40
25 30 distance x [m]
45
50
35
(c) ACIR: 35 dB.
45
50
40
45
Zoom
99
50 98
40
30
α=0.01 α=0.1 α=0.5 α=0.90 α=0.99
85
20
75
40
60
Remaining capacity [% ]
Remaining capacity [% ]
70
Zoom
Remaining capacity [%]
Remaining capacity [% ]
70
30
35
(b) ACIR: 30 dB.
100
60
25 30 distance x [m]
97
96
α=0.01 α=0.1 α=0.5 α=0.90 α=0.99
95
20
94
10 5
50
0
5
10
10
15
15
20
20
25 30 distance x [m]
35
40
25 30 distance x [m]
45
50
35
40
45
50
(d) ACIR: 40 dB.
Fig. 3.17. Relative cell capacity with four interfering MSs. The capacity is shown for different ACIR factors. The frame synchronisation, α, is used as a parameter. The cell load factor is χ = 0.75 and the tolerable outage, Pou t , is 5%. The graphs with solid lines depict the results of scenario (A) whereas the dotted curves show the results of scenario (B). All results implicitly assume handovers.
• An ACIR of 25 dB can lead to significant capacity reductions.
• The effects of non-ideal power control (scenario (B)) on capacity are most significant for ACIR values less than or equal to 30 dB. • The greater interference contribution from MSs in (B) mostly affects the capacity for BS locations of x > 30 m and synchronisations with α ≤ 0.5. This, in turn, means that the detrimental effects of greater interference from MSs at locations of x > 30 m can be avoided by changing the synchronisation so that α > 0.5. Thus, the fundamental finding in section 3.3.2 is also valid for a multiple-cell scenario. The basic finding is that opposed synchronisation (α = 0.99) is sometimes more advantageous than synchronous transmissions (α = 0.01). Thus, this mechanism inherent to
Interference and capacity analyses
•
• • •
89
a TDD system may be exploited to minimise severe interference scenarios resulting, for example, from non-ideal power control. The mechanism discovered by this analysis is known as the TS-opposing technique. For BS separations in the range 22.5 m ≤ x ≤ 27.5 m the capacity is least sensitive to variations in the frame synchronisation factor α. Alternatively to the TS-opposing technique, the fact that the interference for BS separation in the range 22.5 m ≤ x ≤ 27.5 m is almost independent of TS synchronisation may be exploited to maintain a constant capacity in the COI regardless of the synchronisation to the interfering network. This may be important if the victim cell belongs to a different operator which does not synchronise to the coexisting network. An ACIR of 40 dB and α = 0.01 merely yields a capacity reduction between 0.1% and 6%. Location of the victim BS at x < 10 m results in a significant capacity drop due to the great interference from the close-by BS unless the TSs are synchronised with an error of less than or equal to 1%. When comparing the results of the single-cell scenario with an ACIR of 35 dB (Figure 3.9(d)) with the similar results obtained from the multiple cell scenario in Figure 3.17(c) (solid curves), it can be found that there are situations where the single interfering cell creates greater capacity losses in the victim cell than if an interfering network with handovers is considered. This finding is counter intuitive, but can be explained by the significant interference reduction due to handovers. 3.4 Co-channel interference in a CDMA-TDD system
In the previous sections various scenarios are considered to study the effects of ACI in a CDMA-TDD system. The parameters used in these investigations are closely related to the UTRA interface. The results reveal a significant property of such a system, which is that synchronous transmissions between victim and interfering cells does not yield the lowest interference under all circumstances. The characteristics of CCI are different because cells usually do not overlap, but the same carrier is used in every cell. The interference protection results from the spatial separation and methods such as antenna sectorisation and handover. If it can be shown that the basic observation that opposed transmissions are sometimes causing less interference than synchronous transmissions also holds in the case of CCI, DCA algorithms might be found that exploit this mechanism to permit cell independent channel asymmetries between neighbouring cells. Therefore, in this section the properties of CCI are inves-
90
Next Generation Mobile Access Technologies
tigated, given that synchronous as well as asynchronous transmission can occur in a CDMA-TDD system (if neighbouring cells adopt different traffic loads in the UL and DL by TS pooling, asynchronous transmissions occur inevitably). 3.4.1 Simulation platform The cell structure as depicted in Figure 3.18 is used to carry out the CCI analysis, i.e. a cell reuse distance of 1 is applied.
desired link interference link
Mobile station Base station
MS d 2 φm d1 dB S φb d1 max
R COI
Fig. 3.18. Cell model used to calculate interference in the COI.
Interference is evaluated in the COI for different user populations within the first tier of cells. The propagation model and handover model are the same as described in section 3.3.3. CCI is calculated assuming non-ideal power control with σεu = σεd = 2.5 dB and C/I-based power-control algorithms in the UL and DL. The simulations are conducted using the parameters given in Table 3.5. 3.4.2 Performance metric The six neighbouring cells of the COI are equally and uniformly populated. The transmitting entities (MSs and BSs) of these neighbouring cells cause
Interference and capacity analyses
91
interference at the COI. For this purpose a quadratic mesh is placed on top of the cell of interest in Figure 3.18 and for each grid point, (x, y), an interference vector is calculated with one component being the interference resulting from the neighbouring BSs: Ib (x, y) =
L j=1
P cdj , aj (x, y)
(3.55)
and with a second component being the interference resulting from the MSs: Im (x, y) =
Mj L j=1 i=1
P cui,j , ai,j (x, y)
(3.56)
where L is the number of adjacent cells (in the interference analysis here, L is confined to the first tier of cells), P cdj is the transmitted carrier power (3GPP, TSG, RAN, 1999b) of BS j. Furthermore, aj (x, y) is the path loss between the grid point specified by the x and y coordinates and BS j, ai,j (x, y) is the path loss between the user i in cell j and the grid point in the COI specified by (x, y). P cui,j is the transmit power of mobile i in cell j. At each grid point a binary decision is made whether Ib is greater than Im , & −1 if Ib < Im , ˜ (3.57) Ψ(x, y) = 1 otherwise. Then Q Monte Carlo runs are carried out and for each grid point a weighting factor is calculated as follows: Ψ(x, y) =
Q
˜ Ψ(x, y)i (x, y) .
(3.58)
i
Using Ψ(x, y) it is possible to calculate the probabilities,
Ψ(x, y) Z(x, y) = P r [Im (x, y) < Ib (x, y)] = 0.5 1 + . Q
(3.59)
3.4.3 Results The probabilities, Z(x, y), within the COI were calculated. The location of the BS of the COI is at x = y = 0. The radius √ that defines the handover regions (Figure 3.12), d1max , equals d1max = 3 3R/4 ≈ 65 m, which is set to 75% of the distance between two BSs. The results are depicted as three-dimensional plots in Figure 3.19 and as contour plots in Figure 3.20. It is found that Ib dominates over Im (Z(x, y) >> 50%) throughout the
92
Next Generation Mobile Access Technologies 0.66
0.68
0.7
0.74
0.64
0.68 0.72
0.66
0.7 > I MM )
0.62
0.68
0.66
BM
0.66 0.64
0.6 0.6
Pr (I
Pr (I
BM
0.62
> I MM )
0.64
0.58
0.64 0.62 0.6
0.56
0.58
0.58 0.54 50
0.62
0.56 50 50
50
0.56
0
0.6
0 0
0 0.58
y [m]
50
50
y [m]
x [m]
50
50
x [m]
(a) Three interfering users in each neighbour- (b) Four interfering users in each neighbouring cell. ing cell. 0.76 0.72 0.74
0.8 0.74
0.72 0.7 0.75
0.68
BM
0.66
0.72
0.7
0.7
Pr (I
Pr (I
BM
0.68
> I MM )
> I MM )
0.7
0.64
0.66 0.65
0.62 0.6 50
0.64
0.68
50
0.66
50 0 0 y [m]
50
50
x [m]
50 0.62
0 0 y [m]
50
50
0.64
x [m]
(c) Five interfering users in each neighbouring (d) Six interfering users in each neighbouring cell. cell.
Fig. 3.19. Three-dimensional plots of the probabilities that Ib is greater than Im within and around the COI.
entire COI. Note that the probability of the absolute interference cannot be inferred from Z(x, y). The maximum probability of Ib being greater than Im is obtained at the centre of the COI where Z(x, y) is between 58.5% and 75% (assuming three to six interfering MSs). This is of particular interest since the UL performance for all connections is determined at this location. From the contour plots it can be seen that the probability that Ib is greater than Im at the cell corners varies between 56% and 66%. At these locations Z(x, y) is lowest because the MSs transmit highest powers at the cell corners. The results in Figure 3.20 can also be used to find an upper limit for outage caused by MS↔MS interference at any given point within the COI. When assuming an ideal synchronised network, Ib and Im are mutually exclusive. This means that either Ib or Im constitutes co-channel interference. It merely depends on the TS configuration between neighbouring cells as to which component needs to be considered. It is found that the maximum
Interference and capacity analyses 50
93
50
40
58.3
56.2
60.3
58.3 56.2
30
66.4 64.3
40
60.3
64.3 62.2
62.2
56.2
30
56.2
60
60
60
.2
56
62.2
56
60
.2
.3
40
30
60.3
20
10
0 x [m]
10
20
30
40
40
.3 64
60
50 50
50
60 60
62.2 64.3 66.4
58.3
.3
60 .2
62
62.2
40
62.2 64.3 66.4
60
60.3
50 50
.3
30
56.2
56.2
58
62.2
y [m] 56.2
56.2
40
64
60 20
56.2
56.2
0
10
56.2
58.3
.2
56
20
30
60 64.3
0
10
64.3
10
56
58.3
y [m]
.2
58
58
56.2
.2 62
.3
10
.2
20 62
20
30
20
10
0 x [m]
10
20
30
40
50
(a) Three interfering users in each neighbour- (b) Four interfering users in each neighbouring cell. ing cell. 50
50 .9
67.3
.1
20
65.1
71.7
50 50
40
30
65.1
67.3
62.9
69.5 20
10
0 x [m]
67.3 69.5
10
20
30
67
30
62.9
65.1
65.1
67.8
.8
70.2
40
65.4
70.2
67.3
40
.2
y [m]
67.3
69.5
75
.6
65
69.5
.9 62
30
.8
67
.6 72
70
69.5
.2
.1 65
65.4
72
.3
.9
10 62.9
67 .8
75
.8
67
71.7
.1
65
67
62
0
.2
71.7
.6
67.8
70.2
y [m]
75
10
70
71.7
10
20
72.6
72
.1
.4
65
70.2
20
69
.9
69.5
65
0
30
.1
.5
10
70.2 67.8
.8 67
.8
62
62.9
72.6 65.4
67
65
67.3
67 .3
65.1
20
72.6 70.2
40
65.1 62.9
.8 67
.1
65
30
69.5
70
67.3
71.7
62
65.1
65.1
69.5
65.4
71.7 40
67.8
67.8
65.4 72.6
71.7
40
50
50 50
40
30
20
10
0 x [m]
10
20
30
70.2
40
50
(c) Five interfering users in each neighbouring (d) Six interfering users in each neighbouring cell. cell.
Fig. 3.20. Contour plots of the probabilities that Ib is greater than Im within and around the COI.
probability of Im being greater than Ib is only about 44% for a worst case location at any of the six cell corners. As a consequence, only in 44% of the cases will outage be caused by another mobile. Hence, for the scenario investigated, an upper bound of outage as a consequence of MS↔MS interference is given by evaluating 1 − Z(x, y). This means that if a MS at the location (x, y) experiences outage, the maximum probability that this is caused by MS↔MS interference is 1 − Z(x, y). 3.5 Conclusions This chapter considered interference issues inherent to a CDMA/TDD air interface. The analysis of ACI showed that interference resulting from BSs can cause a significant capacity loss if BSs are located in close proximity and
94
Next Generation Mobile Access Technologies
if both cells do not receive and transmit synchronously. This limits the freedom in planning such a network if the ACIR is not increased beyond 40 dB for such BS locations. It was found that the ACI resulting from an isolated cell can be greater than ACI from an underlying network in which handovers are used. This observation supports the use of TDD in a cellular network. This is further corroborated by observation that opposed synchronisation of TS’s can yield higher capacity than synchronous transmissions. In the case of ACI, cells may not be able to synchronise to cells belonging to another operator. The study on ACI revealed that for BS locations at about half the cell radius the interference power varies the least for different TS synchronisations. It was further found that a TDD system can suffer significantly if the transmitted powers in the DL are not reduced to the minimum required power determined by the C/I ratio at each MS individually. Hence, tight DL power control is required. A co-channel interference investigation was carried out. In particular, the inherent TDD property that at any given point in the network interference may result from MSs or BSs was studied. It was found that, at the centre of the COI, in up to 75% of cases the interference power from the BSs of the other cells, Ib , is greater than the interference power from all MSs in the neighbouring cells, Im . The upper bound on outage due to MS↔MS interference was found to be 44%. The investigation on CCI confirmed the novel finding of the ACI analysis: in 25%–41.7% of all investigated user distributions it is more advantageous to apply opposed synchronisation in a TDD system. These figures are obtained for a load of between three and six simultaneously active MSs in each of the six interfering cells. If this new finding is exploited systematically, cell-independent channel asymmetries between neighbouring cells using a TD-CDMA/TDD interface, such as UTRA-TDD, can be enabled whereby capacity can be gained. This finding is counter intuitive since it is commonly assumed that cell-independent channel asymmetries in such a system yield a significant capacity drop as, for example, can be found in (Holma and Toskala, 2000, Chapter 10). According to Holma and Toskala: “Cell-independent asymmetric capacity allocation between UL and DL is not feasible for each cell in the coverage area”. In chapters 4 and 5 DCA algorithms are developed to exploit those findings and to demonstrate that cell-independent asymmetries between neighbouring cells are indeed feasible.
4 Centralised DCA algorithm using the TS-opposing idea Harald Haas
4.1 Introduction This chapter aims to exploit the key finding of Chapter 3 and apply it to the TDD air interface of UMTS (UTRA-TDD). The significant finding of the previous chapter is that ideal synchronisation is not necessarily a prerequisite to obtaining the maximum capacity in a TD-CDMA/TDD network. This has led the author to develop a novel technique which is called the time slot (TS)-opposing method. In this chapter this method is used to develop a centralised dynamic channel assignment (DCA) algorithm. The approach in this chapter is as follows: first, in section 4.2 a simple centralised DCA algorithm used in a single cell is studied. This investigation aims to find an upper bound of the network performance when combining the TS-opposing technique with a DCA algorithm. Second, in section 4.3 the TS-opposing algorithm is investigated in a cellular TD–CDMA/TDD network. For this approach it is assumed that a group of BSs (following the so called bunch concept (Mihailescu et al., 1999)) is connected to a radio network controller (RNC). 4.2 TS-opposing technique applied to a single cell In this investigation an idealised deployment scenario is assumed to investigate the new TS-opposing mechanism. This means that a TS-opposing algorithm is employed with the aim of improving the capacity only with respect to a single cell. The capacity obtained thereby is then compared with the capacity of an equivalent FDD interface. A cluster of seven hexagonal cells is assumed with the cell of interest (COI) in the centre. The effects of the TS-opposing technique on the interfering cells is neglected in order to find the maximum capacity gain. In section 4.2.1 the TS-opposing technique is described mathematically. In section 95
96
Next Generation Mobile Access Technologies
Cell 2
α1,2 α1,3 α1,2
α1,3 Cell 1 = COI
Cell 3
Fig. 4.1. A cell arrangement with each cell using two successive time slots where the first begins at the same time in each cell is shown. The direction of transmission is arranged so that the cell of interest (COI) and cell 2 receive in TS 0 and transmit c in TS 1. In contrast, the BS of cell 3 first transmits and then receives. 2000 IEEE
4.2.2 a simple DCA algorithm is presented and applied to the simulation environment described in section 4.2.3. The performance of the DCA is compared with the capacity of an equivalent FDD interface with the results being discussed in section 4.2.4.
4.2.1 System model It was demonstrated in Chapter 3 that the received, power-controlled signal power from the desired user (3.2) can be found as: Puu =
Iou + N , (pg/εu ) + 1 − MFDD
(4.1)
where MFDD is the number of simultaneously active users in an FDD system, Iou the total other-cell interference power, N the thermal noise power,
Centralised DCA algorithm using the TS-opposing idea
97
εu the bit-energy to interference ratio and pg the processing gain. Note that symbols which are followed by the superscript ‘u’ are associated with the uplink channel; symbols which are followed by the superscript ‘d’ are associated with the downlink channel. It can be seen that Puu is a function of the number of simultaneously active users, M . Therefore, a factor can be defined as to how the required signal power at the receiver has to be increased as the cell load increases. This factor is commonly known as the interference margin (Shin et al., 1999): ν=
Puu (M ) = Puu (M = 0)
1 1−
(MFDD − 1) εu pg
.
(4.2)
q
Using (4.2) the desired signal power in (4.1) can be rewritten as: Puu =
ν N εu . pg
(4.3)
This allows study of the dependency of the desired signal power, Puu , on the u D −1) ε number of active mobiles in the cell. Since the term q = (M F D pg in (4.2) has to fulfill |q| < 1, ν can be interpreted as the value to which the infinite geometric series converges: ν = 1 + q + . . . + q n−1 n → ∞
(4.4)
Hence, it can be seen that Puu increases non-linearly with an increasing cell load. It is well known that CDMA is an interference-limited multiple access technique. This means that in a system with a single user detector, the capacity is primarily limited by the multiple access interference (MAI) power, which, in this book, is equivalent to the own-cell interference power. With the aid of (3.1) given as follows: εu =
Puu pg , Puu (M − 1) + Iou + N
(4.5)
Iown
the own-cell interference power yields:
Iown = Puu (MFDD − 1).
(4.6)
Since Puu increases non-linearly with the number of active users in a cell, it directly follows from (4.6) that own-cell interference also increases nonlinearly. Moreover, (4.1) contains a singularity at MFDD = εpgu + 1 which
98
Next Generation Mobile Access Technologies
defines the theoretical maximum of users that can be served. The capacity maximum is also referred to as the pole capacity (Veeravalli and Sendonaris, 1999; Holma and Toskala, 2000). At the pole capacity both the desired signal and own-cell interference approach infinity. This relationship will be useful in explaining our later results. It can be shown that (4.1) can be transformed into an equation used in a paper by Viterbi (Viterbi and Viterbi, 1993). In this paper, it was demonstrated that the maximum number of users that can have access to a FDD-CDMA system using a single-user detector can be expressed as: MFDD ≤
pg (1 − η) , εu (1 + f )
(4.7)
where η = N/Itot with Itot being the maximal total acceptable interference power and f the ratio of other-cell interference to own-cell interference at the COI. In the assumptions leading to (4.7) only other-entity interference, as defined in the previous chapter, is included because it describes the capacity of an FDD system. It was demonstrated in Chapter 3 that when using the TDD mode also BS↔BS and MS↔MS interference, characterised as sameentity interference, can exist. With the aid of Figure 4.1†, equation (4.7) will be modified to also include same entity interference. Figure 4.1 shows a cell arrangement with each cell using two successive time slots (TS 0 and TS 1) — with TS 0 synchronised in all cells. The direction of transmission is arranged such that the BS in the COI (cell 1) and the BS in cell 2 receive in TS 0 and transmit in TS 1. In contrast, the BS in cell 3 first transmits and then receives resulting in an asynchronous TS overlap‡ at TS 0 and TS 1 between the COI and cell 3, and between cell 2 and cell 3 respectively. At the COI, when comparing interference from cell 2 with interference from cell 3, two entirely different interference scenarios exist. The interference between the COI and cell 2 consists only of other entity interference. This is the same for an FDD system. When investigating interference between the COI and cell 3, it can be found that other-entity interference does not exist, but same-entity interference can be observed instead. Assuming a network with ideally aligned TSs these are the two different interference scenarios that can occur in a TDD system. A factor, α, is introduced to account for the different interference scenarios shown in Figure 4.1. Since all TSs are † Figures 4.1, 4.4, 4.7 and 4.10 are reproduced with permission from (Haas et al., 2000a). ‡ The expression ‘asynchronous TS overlap’ is used to express the state in a TDD system when the same type of entity (BS or MS) in two adjacent cells does not transmit and receive synchronously at any time. This state is created automatically if two adjacent cells adopt a different rate of channel asymmetry.
Centralised DCA algorithm using the TS-opposing idea
99
assumed to be aligned α can only take values of 0 or 1. Thus, interference from the first tier of adjacent cells can be written as: Iou
=
7 ' i=2
(1 − αi,1 ) Imbi ,1 + αi,1 Ibbi ,1
(
αi,1 ∈ {0, 1} ,
(4.8)
where Imbi ,1 is the interference power at the BS of cell 1 resulting from all mobiles in the adjacent cell, i. Similarly, Ibbi ,1 is the interference power at the BS of cell 1 caused by the BS in the neighbouring cell, i, and 1 if opposed transmission (Tx) and reception (Rx), αi,1 = (4.9) 0 otherwise. Substituting (4.8) into (4.7) yields: MTDD ≤
εu
1+
7
pg (1 − η)
i=2 [fi,1
+ αi,1 (gi,1 − fi,1 )]
,
(4.10)
where gi,1 is the ratio of other-cell interference conveyed by the BS i to own-cell interference in the COI. Similarly, fi,1 is the ratio of the total MS interference of cell i to own-cell interference. It holds that f=
7
fi,1 ,
(4.11)
7
gi,1 .
(4.12)
i=2
and g=
i=2
The value obtained by (4.11) is the same as observed in an FDD system and the result of (4.12) is inherent to a TDD system. Equation. (4.10) shows an interesting property of a CDMA/TDD system in that the capacity can, in principle, be higher than in an equivalent FDD system if the Rx/Tx direction of two neighbouring cells are chosen carefully. For example, if gi,1 is smaller than fi,1 and αi,1 is chosen such that the TSs of the respective cells overlap asynchronously (BS i is transmitting while the BS of the COI is receiving, αi,1 = 1), the total other-cell interference is smaller than fi,1 and thus smaller than in an equivalent FDD system. Moreover, the implications of these findings are counter intuitive, i.e. in a TDD system one would expect that the adoption by neighbouring cells of different rates of asymmetry (with the consequence that the TSs will overlap asynchronously) would cause a significant capacity loss. However, the results in section 3.4 reveal that in 25–41.7% of all uniform user distributions it
100
Next Generation Mobile Access Technologies
would be more advantageous if same-entity interference was effective. Since this scenario inevitably occurs in the case of asynchronous TS overlaps, the strategy is to exploit the previous finding by a DCA algorithm that reduces interference in the occurrence of asynchronous TS overlaps. This eventually enables neighbouring cells to apply different rates of channel asymmetry. Therefore, a DCA algorithm is presented which adapts αi,1 for each neighbouring cell so as to minimise interference. The DCA algorithm is assumed to only minimise interference in the COI. This optimisation process, however, has mutual effects on the interference in each of the six neighbouring cells. These effects are that the interference in the neighbouring cells may not necessarily be diminished or may even be increased. In addition, the simple DCA algorithm merely minimises the interference at the BS. Although it was found that MS↔MS interference is not a critical issue (Holma et al., 1999), the situation may arise where severe MS↔MS interference needs to be arbitrated. Hence, interference minimisation with respect to the BS of a single cell results in an idealised scenario. However, this provides a valuable bound on network performance. The aim of this analysis is to directly compare the capacity results of an equivalent FDD interface obtained using (4.7) with the capacity of a TDD system which uses a TS–opposing algorithm as described in (4.10). Note that (4.10) can be reduced to (4.7) by setting α ≡ 0 for all user scenarios. Thus, the relative capacity can be expressed as: 6 ζi,1 , (4.13) MTDD /MFDD = 1 − 6 i=1 1 + i=1 (fi,1 + ζi,1 )
with ζi,1 = αi,1 (gi,1 − fi,1 ).
4.2.2 A simple DCA algorithm From (4.10) it follows that the best strategy is to minimise ζi,1 with respect to αi,1 , which can be achieved by: 1 if gi,1 < fi,1 , αi,1 = (4.14) 0 otherwise. An interpretation of (4.14) is that, whenever the BS interference contribution from adjacent cell i is smaller than the total MS interference power from cell i, the algorithm forces an asynchronous overlap to occur at the respective TSs. The consequence is that the uplink and downlink between two cells are in opposed direction. Using the strategy as described in (4.14), ζi,1 in
Centralised DCA algorithm using the TS-opposing idea
101
(4.13) becomes: ζi,1 =
0
gi,1 − fi,1
if gi,1 ≥ fi,1 ,
if gi,1 < fi,1 .
(4.15)
As a consequence, and quite important to note, it holds that ζ=
7 i=2
ζi,1 ≤ 0 .
(4.16)
Combining this property with (4.13), it can be found that MTDD / MFDD ≥ 1 .
(4.17)
This means that by using the proposed TS-opposing algorithm the uplink capacity of a single cell is always greater than or equal to the capacity of an equivalent FDD cell for the scenario investigated. Monte Carlo techniques are used to calculate the expected value, E(MTDD /MFDD ).
4.2.3 Simulation environment A cluster of seven hexagonal cells with the COI in the centre, as shown in Figure 4.1, is applied. The interfering mobiles are distributed uniformly and allocated to the BS that offers the least signal attenuation (Chebaro and Godlewski, 1992). However, a handover margin, δ, is considered such that the following holds: ai < aj + δ for all j = i ,
(4.18)
where ai is the path loss from a MS to its serving BS and aj is the path loss from the same MS to the neighbouring cell j. The path loss is calculated using the static COST-231 indoor path-loss model with no wall or floor losses (ETSI 30.03, V3.2.0 (1998-04), 1998): a = 37 + 30 log10 (d) + ξ
[dB],
(4.19)
where d is the transmitter–receiver separation distance in metres and ξ is a lognormal random variable modelling shadowing effects. The model of correlated propagation paths as given in section 3.3.2.1 is used. In addition, a simple downlink power control algorithm as described in section 3.3.2.2 is applied. The simulations are conducted using the parameters listed in Table 4.1.
102
Next Generation Mobile Access Technologies
Table 4.1. Parameters used for the simulation of the simple DCA algorithm. Parameter
Value
Information bit rate
16 kbps
Chip rate
3.84 Mcps
Standard deviation of lognormal shadowing
10 dB
Receiver noise figure
5 dB
Max. MS Tx power
10 dBm
Max. BS Tx power
24 dBm u
bit energy to interference ratio, ε
3.5 dB
Handover margin, δ
5 dB
Cell radius, R
50 m
4.2.4 Results Monte Carlo techniques are used to calculate the expected value E (MTDD /MFDD ) of (4.13). The investigation is restricted to a single pair of TSs because this is sufficient to demonstrate the mechanism of the TSopposing algorithm. The results are depicted in Figure 4.2. The number of equally distributed users per cell and TS is varied and drawn on the abscissa. The interference analysis carried out in section 3.4 of the previous chapter has revealed that interference from neighbouring BSs dominates the interference resulting from MSs. Therefore, two cases are investigated: (a) extra 10 dB signal attenuation is considered between the static BS↔BS interference path, and (b) no extra signal attenuation is assumed. The extra attenuation may be obtained by, for example, antenna beam-forming or additional BS isolation due to walls. It can be seen that the largest capacity increase for the case of no extra shielding is a factor of 1.48 for two users per cell. In contrast, if an extra 10 dB BS↔BS isolation is considered the capacity gain is a factor of about 2.18. It can be seen that the capacity gain decreases monotonically as the number of active MSs increases. This behaviour can be explained with the aid of Figure 4.3. In Figure 4.3 the expected values E(f ), (4.11), and E(g), (4.12), as a function of the active number of users per cell is depicted. Two interesting properties can be found. First, in the case of no extra shielding between BSs, the interference from the surrounding BSs is about 10 to 15 times greater
Centralised DCA algorithm using the TS-opposing idea
103
2.4 no extra BSBS shielding 10 dB BSBS shielding
2.3 2.2
2
1.9
Relative capacity M
TDD
/M
FDD
2.1
1.8 1.7 1.6 1.5 1.4 1.3 1.2 1.1 1
2
3
4
5 6 Number of users per cell and TS
7
8
Fig. 4.2. Relative capacity of a TDD cell when using the TS-opposing algorithm compared with an equivalent FDD cell. 1.9 no extra BSBS shielding 10 dB BSBS shielding
no extra BSBS shielding 10 dB BSBS shielding
32
1.8
30 28
1.7
26 24
1.6
22 20
E(f)
E(g)
1.5
18 16
1.4
14 12
1.3
10 8
1.2
6 4
1.1
2 1
2
3
4
5 6 Number of users per cell and TS
7
8
(a) Interference from other-cell MSs normalised by own-cell interference, f
2
3
4
5 6 Number of users per cell and TS
7
8
(b) Interference from other-cell BSs normalised by own-cell interference, g
Fig. 4.3. BS to BS interference, and MS to BS interference respectively, normalised by the total own-cell interference power (Iown ) as a function of the number of active users per cell. The results assume the use of the TS-opposing algorithm.
than the interference resulting from all MSs. This is primarily due to the simple downlink power control algorithm used. As expected, however, E(g) is reduced by a factor of 10 if 10 dB extra attenuation is considered. Second, E(g) increases almost linearly whereas E(f ) decreases non-linearly as
104
Next Generation Mobile Access Technologies
the number of users in the cells increase. These effects can be explained as follows. For a given number of distributed MSs the desired signal power, Puu , can be calculated using (4.3). The power transmitted by a MS is found by multiplying the respective path loss between the MS and its BS by the signal power Puu required at the receiver. This transmitted power causes interference at the BS of the COI which can be modelled as a sum of independent lognormal random variables (Prasad et al., 1993) whose mean does not increase linearly with Puu (Schwartz and Yeh, 1982). The pdf can be calculated from (3.50). Given, however, that own-cell interference increases directly proportional to Puu , which can be found from (4.6), it follows that E(f ) decreases non–linearly. The situation is different for E(g) since the transmitted code power for each user causes interference via the same path. This causes the steady increase of E(g). Since E(g) increases at the same time as E(f ) decreases, this makes the cases when gi,1 is smaller than fi,1 less likely and thus proves that the capacity diminishes as depicted in Figure 4.2. Since FDD systems have been investigated intensively, the results for E(f ) can be compared with results of an investigation by Viterbi (Viterbi and Viterbi, 1994), in which an analysis of other-cell interference in an FDD system (only other-entity interference applies) is carried out. The interference in the victim cell is related to the desired signal Puu instead of Iown as in our considerations. However, the results for two users can be compared since in this case own-cell interference is identical to the desired signal power, Puu , which can be seen from (4.6). E(f ) was found by Viterbi to be 1.32 (Viterbi and Viterbi, 1994), whereas 1.85 (Figure 4.2) is obtained in this investigation. The difference observed can primarily be attributed to the use of a path-loss exponent in our model of 3.0 compared to 4.0 used by Viterbi. This is verified by conducting the simulation with the parameters applied by Viterbi’s approach. This results in E(f ) = 1.306, which provides the necessary evidence to support the model used and the results obtained. It is found that the capacity gains of a CDMA/TDD interface over an equivalent CDMA/FDD interface can be significant, but the assumed simplifications lead to a more detailed approach in the following section.
4.3 TS-opposing technique in a multiple-cell environment Based on the findings of the investigation in section 4.2, a new, centralised DCA algorithm applied to a multiple cell environment is developed. Once again, this algorithm exploits the discovery made in Chapter 3 that it is sometimes advantageous to oppose the TSs. In the following, a limited number of BSs connected to a RNC is assumed. The DCA algorithm con-
Centralised DCA algorithm using the TS-opposing idea
105
sidered henceforth is assumed to be operated at the RNC level. At this level considerably more system information is available than at the MS or BS level. For example, information about the state of interference in several cells is available simultaneously. The group of cells that are connected to the RNC can be considered as a higher level cell. These assumptions are used to build a mathematical framework in section 4.3.1, followed by a description of the new DCA algorithm in section 4.3.2. The simulation platform is presented in section 4.3.3. In section 4.3.4 the results are discussed. As a consequence of the novel DCA algorithm, cell-independent asymmetric capacity allocation between the uplink and downlink for each cell do not cause a significant capacity loss. In some cases, with different channel asymmetries in neighbouring cells a greater capacity can be obtained than if synchronous transmission and reception is applied.
4.3.1 System model In the following, several definitions are made to describe the system and its mutual dependencies. The set of L cells connected to the RNC is defined as: def
C = {c1 , c2 , · · · , cL } .
(4.20)
Each cell consists of one BS and as there are L cells the set of BSs yields: def
B = {b1 , b2 , · · · , bL } .
(4.21)
Moreover, several MSs are allocated to one BS which results in L sets of MSs, * def ) Mi = m1 , m2 , · · · , m|Mi | i = 1, · · · , L , (4.22)
where | · | is the cardinality of the respective set. Note that in a practical scenario each BS of B will serve a different number of MSs. Therefore, |Mi | may be different for each cell, i. A single radio-frequency carrier is assumed. Due to the TDMA component in UTRA-TDD a frame is divided into a maximum of N TSs which can asynchronously overlap with respect to the transmission direction, i.e. neighbouring cells may or may not simultaneously transmit and receive. Therefore, for each TS a symmetric synchronisation matrix can be defined
106
Next Generation Mobile Access Technologies
by:
⎛
where αi,j =
⎜ ⎜ α=⎜ ⎝
0 α2,1 .. .
α1,2 0 .. .
··· ··· .. .
αL,1 αL,2 · · ·
α1,L α2,L .. . 0
⎞
⎟ ⎟ ⎟, ⎠
(4.23)
0 if ci and cj simultaneously transmit or receive at TS n, 1 if ci and cj adopt opposed transmission at TS n.
(4.24) As α only consists of binary elements a complementary matrix α ˜ can be defined so that: ⎛ ⎞ 0 1 ··· 1 ⎜ 1 0 ··· 1 ⎟ ⎜ ⎟ α ˜=⎜ . . . −α. (4.25) . . ... ⎟ ⎝ .. .. ⎠ 1 1 ···
0
For each set of MSs, Mi , and each TS, n, a vector of transmitted powers can be established, (4.26) Pcui = P cu 1 , · · · , P cu |Mi | i = 1, · · · , L . Given that L cells are connected to a RNC, (4.26) finally yields: Pcu = (Pcu1 , · · · , PcuL ) .
(4.27)
Similarly, the slot powers transmitted by the L BSs can be denoted as: Pcd = P1d , · · · , PLd . (4.28)
As demonstrated earlier, four interference scenarios can be ascertained in a TDD system (MS↔BS, BS↔MS, MS↔MS and BS↔BS). Therefore, in general, four path loss matrices between the respective entities can be established. It can be shown that it is convenient to use the reciprocal value of the path loss, also referred to as path gain. The path-gain matrix for the MS↔MS case is a symmetric block matrix denoted as: ⎞ ⎛ 0 MMc1 ,c2 · · · MMc1 ,cL ⎜ MMc ,c 0 · · · MMc2 ,cL ⎟ 2 1 ⎟ ⎜ (4.29) MM = ⎜ ⎟, .. .. .. . . ⎠ ⎝ . . . . 0 MMcL ,c1 MMcL ,c2 · · ·
Centralised DCA algorithm using the TS-opposing idea
where MMci ,cj is a |Mi | × |Mj | matrix and represents tween all MSs in ci to all MSs in cj , ⎛ 1 1 ··· am 1 , m 1 a m 1 , m |M | j ⎜ ⎜ .. .. .. MMci ,cj = ⎜ . . . ⎝ 1 1 · · · am am ,m ,m |Mi |
1
|Mi |
|Mj |
107
the path gain be⎞
⎟ ⎟ ⎟. ⎠
(4.30)
For example, am 1 ,m 1 , is the path loss between MS m1 in cell ci and MS m1 in cell cj . With (4.29) the path gain between any MS and all other MSs within the set C is fully described and used to calculate MS↔MS interference. The path-gain matrix for the BB↔BB case is: ⎞ ⎛ 1 · · · ab 1, b 0 ab 1 , b 2 1 L ⎜ 1 0 · · · ab 1, b ⎟ ⎟ ⎜ ab 2 , b 1 2 L ⎟ ⎜ BB = ⎜ (4.31) .. .. .. ⎟ . ⎠ ⎝ . . 0 . 1 1 ··· 0 ab , b ab , b L
1
L
2
In a similar way the path-gain matrix for ⎛ 0 MBc1 ,c2 ⎜ MBc ,c 0 2 1 ⎜ MB = ⎜ .. .. ⎝ . . MBcL ,c1 MBcL ,c2
the MS↔BS case is found as: ⎞ · · · MBc1 ,cL · · · MBc2 ,cL ⎟ ⎟ (4.32) ⎟. .. .. ⎠ . . ···
0
In (4.32) MBci ,cj is a vector defined as: T 1 1 MBci ,cj = ,··· , , am 1 ,bj am |Mi | ,bj
(4.33)
where T is the matrix transpose operator, and am 1 ,bj is the path loss between m1 in cell i and the BS in cell j. For the reciprocal case that the BS in cell j interferes with the MSs in cell i the path gain matrix is: BM = MBT .
(4.34)
The interference experienced by the MSs can now be written as: Im = (Pcu ) (α ⊙ MM) + (Pcd ) (α ˜ ⊙ BM) .
(4.35)
˜ ⊙ MB) , Ib = (Pcd ) (α ⊙ BB) + (Pcu ) (α
(4.36)
same entity interference
other entity interference
Similarly, the BSs experience:
same entity interference
other entity interference
108
Next Generation Mobile Access Technologies
where ⊙ is the operator for the Hadamard Product. The generalised equations, (4.35) and (4.36), for other-cell interference in a TDD system highlight that the uplink and downlink cannot be treated independently unless α equals the null matrix, i.e. synchronous transmission and reception would result in an equivalent FDD system. If this is not the case the required uplink power of an arbitrary MS is not only dependent on the transmitted powers of the MSs in the other cells, but also on the powers transmitted by the BSs in the neighbouring cells, i.e. the power transmitted in the downlink. Note that this has an impact on the requirements for the downlink power control algorithm. In the following, the vectors of transmission powers Pcu , and Pcd respectively, in (4.35) and (4.36) are calculated. This can be achieved by writing (3.13) in matrix notations. Equation (3.13) is repeated here for convenience: εui
= M
j:j=i
pg Puui Puuj + Iou + N
,
(4.37)
Ib
The transformation using matrix notations yields: (I + diag(γ) − γJ) (Pcui )T = (γ Ibi ) ⊙ MBci ,ci ,
(4.38)
where I is the identity matrix; diag(·) is the diagonal matrix representation; T u is the vector of the required carrier to interference γ = γ1u , · · · , γ|M | i ratios at the BS in cell i with dimension: dim(γ) = |Mi |; J is also a vector of dimension: dim(J) = |Mi | with each element set to 1; Ibi is the accumulated interference from other cells and thermal noise at the BS in the cell ci . Making the substitution B = I + diag(γ), (4.38) yields: (4.39) B − γ T J (Pcui )T = γ T Ibi ⊙ MBci ,ci .
Using the Bartlett–Sherman–Morrison–Woodbury Formula (Golub and van Loan, 1996, Chapter 2) and the property of B being a diagonal matrix, (4.39) can be solved for Pcui , which results in:
and thus Pcui
−1 B−1 γ T J B−1 B − γT J , = B−1 + 1 − J B−1 γ T
Ibi B−1 γ T J B−1 γ T −1 T = Ibi B γ + ⊙ MBci ,ci . 1 − J B−1 γ T
(4.40)
(4.41)
The denominator in (4.41) is a scalar and a singularity is observed at the pole
Centralised DCA algorithm using the TS-opposing idea
109
capacity. Thus, by rearranging the two terms within the brackets of (4.41) using the lowest common denominator, (4.41) can be simplified to yield:
V u ⊙ MBci ,ci Ibi , (4.42) Pci = 1 − JV Uc i
where
V=
u γ|M γ1u i| , · · · , u 1 + γ1u 1 + γ|M i|
T
.
(4.43)
From (4.42), given the Ibi term, it can be seen that the transmission powers of the mobiles are linearly dependent on the interference received at the associated BS. The interference power at bi , Ibi , is a function of the transmitted powers of the MSs and BSs in the other cells which can be seen from (4.36). This describes a system of mutual dependencies, i.e. theoretically any change of a transmission power, regardless whether it is an MS or a BS, has an impact on the required transmission powers of all other entities. The transmitted code powers of the MSs in a single cell i is described in (4.42). The scope of this equation can be extended to the entire network. Using the vector U = (Uc1 , · · · , UcL )T where Uci is given in (4.42), the general expression for Pcu can be found as: Pcu = U ⊙ Ib .
(4.44)
A similar equation can be derived for the transmitted code powers in the downlink. It is assumed that C/I-based downlink power control is employed. Equation (2.22), repeated here for convenience: γid =
d P˜ci ai 1 M ˜ d P c j i=j ai
+ Iod + N
⎤
˜ di )T −(γ ⊙ MBci ,ci ) J⎦ (Pc e
(4.45)
Im
can be written in matrix notation to yield: ⎡ ⎢ ⎢diag (γ ⊙ MBc ,c )T + diag(MBc ,c ) i i i i ⎣ B
,
(4.46)
= (γ ⊙ Imci )T ,
110
Next Generation Mobile Access Technologies d
˜ i are the transmitted code powers at one TS in the ith cell – in where Pc contrast to the slot power, which is the sum of all code powers; Imci is T d is the the interference vector at the MSs in cell ci ; γ = γ1d , · · · , γ|M i| vector of the required carrier to interference ratios at the MS in cell i with dim(γ) = |Mi |. Using the substitutions as indicated in (4.46) the matrix inverse of (B−eJ) yields: (B − e J)−1 = B−1 +
B−1 e J B−1 . 1 − J B−1 e
(4.47)
Applying (4.47) to (4.46) the required code powers at the BS in cell ci can be found as: d
˜ )T (Pc i
= −
B−1 (γ ⊙ Imci )T + B−1 eJB−1 (γ ⊙ Imci )T 1 − J B−1 e B−1 (γ ⊙ Imci )T JB−1 e . 1 − J B−1 e
(4.48)
The pole capacity in the downlink is reached if J B−1 e = 1. It is straightforward to calculate the slot power for cell ci from (4.48), which results in: ˜ di , Pcdi = J Pc
(4.49)
which gives the final vector of transmitted slot powers: ˜ d1 , · · · , J Pc ˜ dL . Pcd = J Pc
(4.50)
From (4.48) it can be seen that the slot power in each cell ci is a function of the interference powers at the served MSs, Imci . Thus (4.50) may be more generally denoted as: Pcd = f (Im).
(4.51)
Using (4.51) and (4.44) the system equations in (4.35) and (4.36) may be rewritten as:
Centralised DCA algorithm using the TS-opposing idea
111
self-jamming
Im
=
U ⊙ Ib (α ⊙ MM) @@
++
+
Pcd (Im) (α ⊙ BM) ,
+
U ⊙ Ib (α ⊙ MB) .
cross-jamming cross-jamming
Ib
=
Pcd (Im) (α ⊙ BB)
33
self-jamming
(4.52) The annotated system equations in (4.52) highlight the interference dependencies in a CDMA-TDD system. The upper portion represents the downlink scenario while the lower represents the uplink scenario. The selfjamming effect due to co-channel interference is inherent in a cellular system with frequency reuse. This effect also appears in an FDD interface and can be reduced by power control and manipulating the path-gain matrix, BM, for example, by applying frequency reuse factors greater than 1. A notable example where this technique is employed is GSM — a second generation FDMA–TDMA/FDD system. When using TDD instead of FDD an additional cross-jamming effect can be ascertained. The magnitude of self-jamming and cross-jamming in a TDD system can be manipulated by the synchronisation matrix, α. If α is the zero-matrix cross-jamming is eliminated, but self-jamming may be increased. In contrast, if each element of α equals one, self–jamming does not exist, only cross-jamming. If, as in UTRA-TDD, the multiple access mode consists of a hybrid TD-CDMA interface an additional degree of freedom is added due to the TDMA mode. One TDMA frame is divided into N TSs where each TS can be used for either uplink or downlink traffic. Since the symmetric use of a channel can be considered as a special case of asymmetric usage, the more general term rate of asymmetry is introduced to characterise the load in the uplink and downlink more precisely. For a given rate of asymmetry several solutions for all αn , where n = (1, · · · , N ), may exist. This particular degree of freedom is exploited by the centralised DCA algorithm which utilises the TS-opposing idea. Due to the complexity of the system equations in (4.52) the model is estimated by Monte Carlo simulations. The novel DCA algorithm which is applied to these equations is described in the following section.
112
Next Generation Mobile Access Technologies
4.3.2 DCA algorithm A DCA algorithm is developed which minimises interference at the BSs by either applying opposed or synchronous transmission to the neighbouring cells. It has been demonstrated that a cellular TDD system can be exposed to more interference scenarios than an FDD system. Since CDMA, as an interference limited technique, heavily relies on low interference, the performance of UTRA-TDD can be significantly poorer compared to an equivalent FDD system. However, it was demonstrated in section 4.2 that the additional interference mechanism and the resulting flexibility in a TDD system can be exploited constructively in order to minimise interference. In a system that uses CDMA, such as UTRA-TDD, a reduction of interference is equivalent to increasing capacity. Such a capacity improvement or an increase of the quality of service (QoS) is intended to be achieved by the new centralised DCA for a TD-CDMA/TDD air interface. The basic principle of how this DCA functions can be explained with the aid of Figure 4.4. The DCA algorithm is executed for each BS of a cellular network. In the example of Figure 4.4, the DCA algorithm is executed in cell 1 (COI). In order to reduce interference in the example the DCA algorithm decides to use synchronous transmission with respect to cell 2 (α1,2 = 0), but opposed transmission with respect to cell 3 (α1,3 = 1). For each TS in Figure 4.4 a symmetric synchronisation matrix can be established. Since only two TSs are used and a symmetric service in each cell is implied the synchronisation matrices for both TSs are equivalent and yield: ⎛ ⎞ 0 0 1 (4.53) αTS1 = αTS2 = ⎝ 0 0 x ⎠ . 1 x 0
The x in (4.53) indicates that the synchronisation factors α2,3 and α3,2 between cell 2 and cell 3 are not directly manipulated when the DCA is carried out at the BS of cell 1. It however indirectly follows from the settings of α1,2 = 0 and α1,3 = 1 that α2,3 must be 1 as shown in Figure 4.5†. This can be explained using graph theory (Wilson, 1987) and Kirchhoff’s laws because the state of opposed TS can be interpreted as a potential difference between two vertices (BSs). With the definition of α in (4.23), for any circuit P as, for example, (BS1 → BS2 → BS3 → BS1) it holds that: α mod 2 ≡ 0 , (4.54) P
† Figures 4.5, 4.6, 4.8, 4.9, 4.12, 4.13, 4.14, 4.16, 4.18, 4.20 and 4.23 and Tables 4.2, 4.3 are reproduced with permission from: (Haas and McLaughlin, 2001a).
Centralised DCA algorithm using the TS-opposing idea
113
Cell 2
α1,2 α1,3 α1,2
α1,3 Cell 1 = COI
Cell 3
Fig. 4.4. A cell arrangement with each cell using two successive time slots where the first begins at the same time in each cell is shown. The direction of transmission is arranged so that the cell of interest (cell 1) and cell 2 receive in TS 0 and transmit c in TS 1. In contrast, the BS of cell 3 first transmits and then receives. 2000 IEEE
where ‘ mod ’ is the operator for modulo division. The dependency that results from (4.54) leads to: x = α2,3 = 1. This effect may cause greater interference in cells for which the DCA is currently not executed (cell 2 and cell 3) and thus cancels out or diminishes the capacity gains in cell 1. However, situations can occur where cell 2 or cell 3 can tolerate higher interference, but only minor interference at the BS in cell 1 may be permitted. Hence, the algorithm effectively improves capacity in this case. The final algorithm is explained in pseudo-code depicted in Figure 4.6. The algorithm starts when the mobile is requested to transmit with higher power than the maximum power permitted, i.e. the state at which outage or service degradation would occur. The algorithm steps in assuming that a MS uses at least two TSs for the communication to the BS (TS in Use ≥ 2). It monitors the interference in all n TSs of all neighbouring cells. Two cases can then be distinguished:
114
Next Generation Mobile Access Technologies
x=
1֒2
=
0
BS2
2֒3
= 1
BS1
BS3
1֒3
=1
c Fig. 4.5. The dependencies of α. 2001 IEEE
(i) If TS n in cell l is used for Rx (from the BS point of view) the interference from this particular neighbouring TS is caused exclusively from its MSs since α = 0 and ideal synchronisation is assumed. Furthermore, it is assumed that the MSs in the neighbouring cell are able to determine the path loss to their neighbouring BSs. This may be accomplished by a fixed transmission power on the pilot channel. The MSs report their transmission power and path-loss measurements to the BS which makes it available to the RNC†. Hence, the information about the path-gain matrix of the mobiles in cell l to the BS in cell j, MBcl ,cj , and the vector of transmission powers of the mobiles in cell l, Pcul , are assumed to be available to the DCA algorithm. (ii) If TS n at the BS is used for transmission the interference contribution from cell l results only from the BS (same entity interference as α = 1). The transmission powers at the BSs are known and can easily be reported to the RNC, as can the path loss to the neighbouring BSs, BBcl ,cj .
A check is made to examine if there is one TS n in the neighbouring cell † These measurements may already be required for handover decisions. Hence, the signalling traffic is not increased significantly.
Centralised DCA algorithm using the TS-opposing idea
115
BEGIN Request the measurement of the required Tx power, P cui , of user i in time slot k of cell cj if P cui > P cumax for l = 1 : max ( Neighbouring Cells) for n = 1 : max (TS in Use) where n ∈ set of used TSs if TSncl == Rx time slot determine interference from mobiles: n
Ibl n = (Pcul ) else
MBc l ,c j
n
determine interference from BS:
n n BBc l ,c j Ibl n = P cdl end if ⎡
Ibl n < Ibl k &
⎤
⎦ if ⎣ direction(TSnl ) = direction(TSkl ) exchange TS n for TS k in cell l
end if end for end for else
Assign channel
end if Repeat this algorithm for each user in the RNC area if necessary END c Fig. 4.6. The centralised DCA algorithm exploiting the TS-opposing idea. 2001 IEEE
cj which would cause less interference than the current TS k. If this is true and TS n is used for Rx while TS k was used for Tx, or vice versa, then the neighbouring cell, cl , interchanges TS n with TS k. This results in TS-opposing time slots with respect to the cj . Note that the algorithm is
116
Next Generation Mobile Access Technologies
only carried out if the Tx power of a MS tends to exceed a given Tx power maximum.
4.3.3 Simulation platform A propagation environment with severe lognormal shadowing, a cell reuse factor of 1 and user assignments based on the minimum path loss results in a negligible impact on system performance for particular cell shapes. Of more importance is the maximum distance to the closest BS, usually at the furthest corner of the cell. In order to overcome cell boundary effects a cell wraparound technique is used. Square-shaped cells as depicted in Figure 4.7 are applied. The reasons for this are as follows: first, the cell wraparound technique is easy to apply and, second, the square shaped cells represent a good approximation to an indoor environment. The wraparound technique
Fig. 4.7. The user distribution and user assignment based on the minimum path loss is shown for a random scenario. A wraparound technique is applied to prevent c cell boundary effects. 2000 IEEE
ensures that each cell is completely surrounded by a symmetric pattern composed of three different cells. The principle of this method is depicted in Figure 4.8. MSs are assigned to the BS offering the lowest path loss, but
Centralised DCA algorithm using the TS-opposing idea
117
Axis of Symmetry
Cell 2
Cell 3
Cell 2
Cell 3
Cell 1
Cell 4
Cell 1
Cell 4
Cell 2
Cell 3
Cell 2
Cell 3
Cell 1
Cell 4
Cell 1
Cell 4
c Fig. 4.8. Wraparound technique applied. 2001 IEEE
a handover margin as described in section 3.4.1 is considered. Ideal power control in the uplink is assumed. The path-loss model for indoor office test environment as described in (ETSI 30.03, V3.2.0 (1998-04), 1998) is used:
a = 37 + 30 log10 (d) + 18.3 p
p +2 −0.46 p +1
+ ξ [dB] ,
(4.55)
where d is the transmitter–receiver separation in metres, p is the number of floors in the path and ξ is the lognormal variable modelling shadow fading. In the simulation environment four consecutive TSs are considered. An example is presented in Figure 4.9. It is assumed that the MSs in cells 2– 4 use two TSs whereas in cell 1 an MS occupies four consecutive TSs. This enables an asymmetric communication channel to be created in cell 1 with different loading in uplink and downlink. Channel asymmetry in cell 1, in turn, inevitably results in asynchronous TS overlaps to at least one of the neighbouring cells. After a predefined number of users have been distributed randomly and uniformly in space, the power control loops in the up- and downlink are initiated (eqns. (4.44) and (4.48)). In Figure 4.10, a powercontrol snapshot of a randomly chosen mobile is shown. It can be seen that the transmission power rapidly increases, which leads to the conclusion that
118
Next Generation Mobile Access Technologies
BS4
BS1
BS3
BS2
c Fig. 4.9. Deployment scenario. 2001 IEEE
the total noise floor in the system is high, because of mutual interference effects in a CDMA system. However, a high noise floor means that some mobiles will not achieve the target C/I at the BS and so will experience outage. Removing extreme interferers in turn results in a reduction of the noise floor until the Tx power converges to a stable level as can be seen in Figure 4.10. If the required code power of a MS exceeded the maximum power threshold the novel DCA algorithm described above would step in and try to reduce interference to maintain the required bit-energy to interference ratio, εu , for the MS. If this method fails, one or more MSs experience outage, Mout . Monte Carlo techniques are used to calculate the pdf of outage, p(Mout ). The expected value thereof, E [Mout ], is used to analyse the system performance of the centralised DCA algorithm. From E [Mout ] the capacity for cell ci is determined as follows: C ci =
1 (1 − E [Mout ]) W Mtot 4
[kbps/TS] ,
(4.56)
Centralised DCA algorithm using the TS-opposing idea
119
Fig. 4.10. Dynamic uplink power control. The transmission power of the mobile is c successively adjusted. 2000 IEEE
where W is the user data rate and Mtot is the total number of MSs that are distributed to the network. Since four TSs are assumed in the model, an averaging factor of 1/4 is applied. The capacity per cell and TS can be calculated from (4.56): 4
1 C= C ci 4
[kbps/cell/TS] .
(4.57)
i
Two channel-assignment approaches are compared:
(i) Tx and Rx transmission direction are chosen such that the number of asynchronous TS overlaps is minimum and the MSs are then allocated randomly. Hence, this method is equivalent to a fixed channelassignment (FCA) strategy. (ii) The new centralised DCA algorithm developed in section 4.3.2 is applied which uses the new TS-opposing technique.
120
Next Generation Mobile Access Technologies
4.3.4 Results Four scenarios with different rates of asymmetry are investigated: two that favour the uplink, one of symmetric TS arrangement and one that favours the downlink. These scenarios are summarised in Table 4.2. Table 4.2. Simulated scenarios. The ratio of UL (uplink) versus DL (downlink) usage is shown. The first figure corresponds to the number of TSs used for the UL and the second figure shows the number of TSs used c for the DL. 2001 IEEE TSRX :TSTX
cell 1
cell 2
cell 3
cell 4
Scenario 1
3:1
1:1 / 1:1
1:1 / 1:1
1:1 / 1:1
Scenario 2
3:1
2:0 / 1:1
1:1 / 1:1
1:1 / 1:1
Scenario 3
2:2
1:1 / 1:1
1:1 / 1:1
1:1 / 1:1
Scenario 4
1:3
1:1 / 1:1
1:1 / 1:1
1:1 / 1:1
The parameters given in Table 4.3 are used for the simulations. MSs are distributed uniformly throughout the entire area covered by cells 1–4. The path loss and the handover margin determine to which cell an MS is allocated. Results of scenario 1 The results of scenario 1 are depicted in Figures 4.12(a) and 4.12(b) which show the capacity versus the number of distributed MSs. Figure 4.12(a) shows the average capacity for each cell individually (using (4.56)), while Figure 4.12(b) depicts the accumulated average capacity (using (4.57)) over all cells. It can be found that for a distributed load of less than 28 MSs the capacity in cell 1 is greater than in cells 2–4. The reason for this is that one MS in cell 1 occupies twice as many TSs as a MS in all other cells (Figure 4.9). Furthermore, the capacity in cell 1 has a maximum for a total number of 28–32 distributed MSs. This can be explained with the aid of (3.3). With the parameters applied, it can be found that the pole capacity is reached when about 8 MSs are instantaneously active (M = (pg/εu ) + 1 = (16/100.35 ) ≈ 8). This means that every additional MS (beyond the eighth user) experiences outage. This state is reached when a total number of about 28–32 users are distributed throughout the network due to the uniform user distribution. The situation is different in cells 2–4 because every MS in these cells only utilises 50% of the data rate used by an MS in cell 1. This means that approximately twice
Centralised DCA algorithm using the TS-opposing idea
121
Table 4.3. Parameters used for the simulation of the centralised DCA c algorithm. 2001 IEEE Parameter Cell radius, R
Value 50 m
Bit rate, W
16 kbps
Chip rate
3.84 Mcps
Standard deviation of lognormal shadowing
10 dB
Thermal noise density
169 dBm
Max. MS Tx power
10 dBm
Max. BS Tx power
24 dBm
Bit energy to interference ratio, εu
3.5 dB
Path loss
Indoor test environment (3GPP, TSG, RAN, WG4, 1999)
Handover margin, γ
5 dB
as many MSs can be accommodated in cells 2–4. As a consequence of the increasing number of users in cells 2–4, the interference in cell 1 also increases resulting in capacity losses in that cell. Therefore, a capacity maximum in cell 1 can be ascertained at about 28 MSs. Note that the capacity maximum is strongly dependent on the required bit-energy to interference ratio, εu , at the BS receiver. Methods that enable the same bit-error performance at a reduced bit-energy to interference ratio have a vital impact on the pole capacity and hence on the overall system performance. As expected, the capacity in cells 2–4 is almost the same due to symmetry. The centrally operated TS-opposing algorithm improves capacity in cell 1 by about 20–30% in the range between 14 and 35 distributed users. The improvement in the other cells is negligible which causes the total capacity improvement to be averaged to about 8% (48 kbps/Cell/TS instead of 52 kbps/Cell/TS for 28 distributed users). As described earlier the TSopposing algorithm dynamically readjusts the components of α in (4.52) in order to obtain the best capacity. For scenario 1, the expected values, E(α), resulting from this optimisation process are presented in Figure 4.13. In the case of using the FCA strategy all components of α are deterministic and
122
Next Generation Mobile Access Technologies TS1
TS2
TS3
TS4
cell 1
1
1
0
1
cell 2
0
1
0
1
cell 3
0
1
0
1
cell 4
0
1
0
1
1 := BS receives 0 := BS transmits (a) Scenario 1
Fig. 4.11. The initial TS assignment for scenario 1. 70
55
50 60 45 50
kbps/cell/TS
kbps/TS
40
40
c1 FCA c2 FCA c3 FCA c4 FCA c1 DCA c2 DCA c3 DCA c4 DCA
30
20
10
5
10
15
20 25 30 Number of distributed MS’s
(a) Cell capacity
35
35
30
25
20
40
45
15
FCA DCA 5
10
15
20 25 30 Number of distributed MS’s
35
40
45
(b) Total capacity
Fig. 4.12. Results of scenario 1. The rate of asymmetry, UL:DL,is 3:1 in cell 1 and 1:1 in all other cells. The graphs show: for (a) the capacity in each cell in [kbps/TS] and (b) the total capacity in [kbps/Cell/TS]. The results labelled ‘FCA’ are obtained by the fixed channel assignment procedure and the results labelled c ‘DCA’ are obtained from the novel centralised DCA algorithm. 2001 IEEE
its values are given in the sub-captions of Figure 4.13. From the results in Figure 4.13 two important properties with respect to scenario 1 can be derived: (i) The DCA algorithm does not change α significantly for TS 3 and TS 4, i.e. synchronous transmission and reception for all cells is
Centralised DCA algorithm using the TS-opposing idea 1
123
0.45
0.9 0.4
0.8
α
0.6
α2,3 α2,4
0.5
α3,4
0.4
1,3
α
1,4
α2,3
0.25
α
2,4
α3,4
0.2
0.15
0.1
0.2
0.05
0.1
5
10
15
20 25 30 Number of distributed MSí s
35
40
0
45
(a) Expected values of the components of α at TS 1. In the case of FCA: α 1, 2 = α 1, 3 = α 1, 4 = 1 and α 2, 3 = α 2, 4 = α 3, 4 = 0.
5
1
1
0.8
0.8
15
20 25 30 Number of distributed MSí s
35
40
45
0.6
α1,2
0.4
α
2,3
α
0
2,4
α3,4
0. 2
Expected value of α
α1,4
0. 6
0. 8
35
40
45
(c) Expected values of the components of α at TS 3. In the case of FCA: α 1, 2 = α 1, 3 = α 1, 4 = α 2, 3 = α 2, 4 = α 3, 4 = 0.
α
2,4
α3,4
0. 2
0. 8
20 25 30 Number of distributed MSí s
α
2,3
0. 4
15
α1,4
0
0. 6
10
α1,3
0.2
0. 4
5
α1,2
0.4
α1,3
0.2
1
10
(b) Expected values of the components of α at TS 2. In the case of FCA: α 1, 2 = α 1, 3 = α 1, 4 = α 2, 3 = α 2, 4 = α 3, 4 = 0.
0.6
Expected value of α
α
0.3
0.3
0
α
1,2
1,3
α1,4
Expected value of α
0.7 Expected value of α
0.35
α1,2
1
5
10
15
20 25 30 Number of distributed MSí s
35
40
45
(d) Expected values of the components of α at TS 4. In the case of FCA: α 1, 2 = α 1, 3 = α 1, 4 = α 2, 3 = α 2, 4 = α 3, 4 = 0.
Fig. 4.13. Scenario 1. The expected values of the components of α as a result of c the novel DCA algorithm are depicted for all four TSs. 2001 IEEE
applied. This is the same as for the FCA scheme. Furthermore, since each MS in cell 1 occupies all four TSs, and swapping TSs in cell 1 would involve TS 3 (because it is the only TS used for Tx), it can be inferred that primarily TS 1 and TS 2 in cells 2–4 are rearranged in this scenario. The results in Figure 4.13 are analysed and the probabilities of those TSs which change most frequently are depicted in Figure 4.14. The arrows indicate which TSs are rearranged with the highest probability. Moreover, with the aid of Figure 4.13 it can
124
Next Generation Mobile Access Technologies
also be found that: E TS1 (α1,j ) ≈ 1 − E TS2 (α1,j ) j = 2, · · · , 4
(4.58)
E TS1 (αi,j ) ≈ E TS2 (αi,j ) i = 2, 3 j = 3, 4 for all i = j. (4.59)
It can be summarised that for scenario 1 the DCA rearranges the TSs in cells 2–4 to improve the throughput in cell 1 without affecting the capacity in cells 2–4. (ii) From Figures 4.13(a) and 4.13(b) it can be seen that the greatest changes with respect to the FCA strategy are for a load between 20 and 30 MSs, which therefore can be considered as the optimal operation load for the centralised DCA.
TS1
TS2
TS3
TS4
Cell 1
1
1
0
1
Cell 2
0
1
0
1
Cell 3
0
1
0
1
Cell 4
0
1
0
1
1 := BS receives
:= 27% < pmax < 40%
0 := BS transmits
Fig. 4.14. This illustration shows the initial Tx/Rx configuration with respect to the BSs for scenario 1. The arrows highlight the TSs that changed most frequently. c The associated maximum probabilities are shown. 2001 IEEE
Results of scenario 2 In the second scenario, in addition to the asymmetric traffic in cell 1 (as discussed in scenario 1), the rate of asymmetry in cell 2 is not 1:1. In this scenario cell 2 uses both TS 1 and TS 2 for reception inducing an additional asynchronous TS overlap (see Figure 4.15(a)).
Centralised DCA algorithm using the TS-opposing idea TS1
TS2
TS3
TS4
cell 1
1
1
0
1
cell 2
1
1
0
1
cell 3
0
1
0
1
cell 4
0
1
0
1
125
1 := BS receives 0 := BS transmits (a) Scenario 2
Fig. 4.15. The initial TS assignment for scenario 2. The modification with respect to the previous scenario is highlighted.
In scenario 1 an asynchronous TS overlap between the following cells exists: Cell 2 → Cell 1, Cell 3 → Cell 1, Cell 4 → Cell 1.
(4.60)
In scenario 2 the situation now is: Cell Cell Cell Cell
3 4 3 4
→ → → →
Cell Cell Cell Cell
1, 1, 2, 2.
(4.61)
It is obvious from (4.61) that the TSs between cell 1 and cell 2 can be arranged without generating an asynchronous overlap between these cells. Furthermore, compared to scenario 1, the minimum number of asynchronous TS overlaps experienced by cell 1 is reduced by one. The results in Figure 4.16 reveal that the capacity improvement in cell 1 is maintained and, in addition, the capacity in cell 2 is improved over the FCA technique. Thus the improvement in cell 2 is not at the expense of capacity in cell 1. It is important to note that since cell 1 only faces opposed transmission from two neighbouring cells, the capacity in cell 1 when applying the FCA method is greater than in scenario 1. However, since cell 2 suffers from asynchronous TS overlaps the capacity in cell 2 is reduced when using the FCA
126
Next Generation Mobile Access Technologies
70
60
55 60 50
45
kbps/cell/TS
kbps/TS
50
40
c1 FCA c2 FCA c3 FCA c4 FCA c1 DCA c2 DCA c3 DCA c4 DCA
30
20
10
5
10
15
20 25 30 Number of distributed MS’s
35
40
35
30
25
20
40
15
45
FCA DCA 5
(a) Cell capacity
10
15
20 25 30 Number of distributed MS’s
35
40
45
(b) Total capacity
Fig. 4.16. Results of scenario 2. The rate of asymmetry, UL:DL, is 3:1 in cell 1 , 4:0 in cell 2 and 1:1 in all other cells (cell 3 and cell 4). The graphs show: for (a) the capacity in each cell in [kbps/TS] and (b) the total capacity in [kbps/Cell/TS]. The results labelled with ‘FCA’ are obtained by the fixed channel assignment procedure and the results labelled with ‘DCA’ are these obtained from the novel centralised c DCA algorithm. 2001 IEEE
scheme. The gain over the FCA technique in both cells (cell 1 and cell 2) is between 10–15% resulting in a total maximum improvement of about 8% (52 kbps/Cell/TS instead of 48 kbps/Cell/TS for 28 distributed users). It is interesting to note that the relative capacity improvement is similar to the one achieved in scenario 1. The results of the expected values of αi,j are depicted in Figure 4.17. The following properties and upper bounds for E T Sn (αi,j ) can be observed: 0 ≤ E TS1 (α1,2 )
0≤E
TS3
0.51 ≤ E
TS1
0.05 ≤ E
TS2
E
TS1
(α1,2 ) ≈ E
TS3
(α1,3 ) ≈ E
TS1
(α1,3 ) ≈ E
TS2
(α1,3 ) ≈ E
0.63 ≤ E TS1 (α1,4 ) ≈ E TS1 (α2,4 ) (α2,3 )
0.05 ≤ E TS2 (α1,4 ) ≈ E TS2 (α2,4 ) (α2,3 )
E TS1 (α1,4 ) ≈ 1 − E TS2 (α1,4 ) (α1,3 ) ≈ 1 − E
TS2
TS3
(α1,4 )
≤ 0.014,
(4.62)
≤ 0.93,
(4.64)
≤ 0.014, ≤ 0.91,
≤ 0.37,
≤ 0.49,
(α1,3 )
0.05 ≤ E TS1 (α3,4 ) ≈ E TS2 (α3,4 )
(4.63) (4.65) (4.66) (4.67) (4.68) (4.69)
≤ 0.2.
(4.70)
Using properties (4.62)–(4.70) it can be concluded that the DCA algorithm primarily uses TS 1 and TS 2 in cell 3 and cell 4 to minimise interference.
Centralised DCA algorithm using the TS-opposing idea 1
0.5
0.9
0.45
127
α 1, 3 and α 2, 3
0.8
0.4
α
α 1, 4 and α 2, 4
1,2
0.7
α
α1,2
0.35
α1,3
α1,4
0.6
α2,3 α2,4
0.5
α 1, 3 and α 2, 3
α3,4
0.4
Expected value of α
Expected value of α
1,3
0.3
α2,3
α 1, 4 and α 2, 4
α2,4
0.25
α3,4
0.2
0.15
0.2
0.1
0.1
0.05
0
α1,4
0.3
5
10
15
20 25 30 Number of distributed MSí s
35
40
0
45
5
10
15
20 25 30 Number of distributed MSí s
35
40
45
(a) Expected values of the components of α (b) Expected values of the components of α at TS 1. In the case of FCA: α 1, 3 = α 1, 4 = at TS 2. In the case of FCA: α 1, 2 = α 1, 3 = α 2, 3 = α 2, 4 = 1 and α 1, 2 = α 3, 4 = 0. α 1, 4 = α 2, 3 = α 2, 4 = α 3, 4 = 0. 1
0.014
0.8 0.012
0.6
Expected value of α
α α α
0.008
α 0.006
α 1, 2 , α 1, 3 and α 1, 4
α
1,2
α1,2
0.4
α1,3
1,3 1,4 2,3 2,4 3,4
Expected value of α
α 0.01
α1,4
0.2
α2,3 α
0
2,4
α3,4
0. 2
0. 4
0.004
0. 6 0.002
0. 8
0
5
10
15
20 25 30 Number of distributed MSí s
35
40
45
1
5
10
15
20 25 30 Number of distributed MSí s
35
40
45
(c) Expected values of the components of α (d) Expected values of the components of α at TS 3. In the case of FCA: α 1, 2 = α 1, 3 = at TS 4. In the case of FCA: α 1, 2 = α 1, 3 = α 1, 4 = α 2, 3 = α 2, 4 = α 3, 4 = 0. α 1, 4 = α 2, 3 = α 2, 4 = α 3, 4 = 0.
Fig. 4.17. Scenario 2. The expected values of the components of α as a result of the novel DCA algorithm are depicted for all four TSs.
Only with a maximum probability of about 1.4% is TS 3 in cell 1 exchanged with TS 1 in the same cell. These mechanisms are illustrated in Figure 4.18.
Results of scenario 3 In the third scenario it is investigated whether the TS-opposing algorithm can also achieve better results in the case when no asynchronous TS overlap exists, i.e. equal traffic in uplink and downlink applies. The results of the third scenario are depicted in Figure 4.20. First, where the TS-opposing algorithm is not employed, the maximum capacity in cell 1, for example, is increased from 50 kbps to about 63 kbps compared to the first scenario. Note that in the case of symmetric traffic and no DCA algorithm the maximum capacity in cell 1 is greater than in the other
128
Next Generation Mobile Access Technologies
TS1
TS2
TS3
TS4
Cell 1
1
1
0
1
Cell 2
1
1
0
1
Cell 3
0
1
0
1
Cell 4
0
1
0
1
1 := BS receives 0 := BS transmits
:= 37% < pmax < 49% := p max ≈ 1.4%
Fig. 4.18. This illustration shows the initial Tx/Rx configuration with respect to the BSs for scenario 2. The arrows highlight the TSs that changed most frequently. c The associated maximum probabilities are also shown. 2001 IEEE
cells. This is anticipated as the own-cell interference in cell 1 is lower than in cells 2 – 4. The results reveal that the TS-opposing algorithm does not achieve a capacity improvement if the TSs can be arranged such that no asynchronous TS overlap exists. On the contrary, it was demonstrated previously in section 4.2 that synchronous transmission and reception is not the ideal case with respect to an isolated cell. However, it is anticipated that the capacity gains observed through the investigation in section 4.2 cannot be maintained when considering a cellular network. The reason for this is that TS-opposing with respect to a certain cell also has an impact on the remaining neighbouring cells . The impact on the remaining cells may be such that the interference in one cell, several cells or even all cells but one is increased. This, in turn, means that the capacity gain obtained for a particular cell is offset by higher interference in some other cells. If the higher interference in the neighbouring cells cannot be tolerated (usually the case when assuming uniformly distributed MSs) the TS-opposing does not increase the overall system performance. Figure 4.21 demonstrates that in scenario 3 the DCA algorithm does not change the initial TS configuration which is not subject to an asynchronous TS overlap. The same configuration applies to the FCA scheme. Therefore,
Centralised DCA algorithm using the TS-opposing idea TS1
TS2
TS3
TS4
cell 1
0
1
0
1
cell 2
0
1
0
1
cell 3
0
1
0
1
cell 4
0
1
0
1
129
1 := BS receives 0 := BS transmits (a) Scenario 3
Fig. 4.19. The initial TS assignment for scenario 3. The modification with respect to the previous scenario is highlighted.
70
60
55 60 50
45
kbps/cell/TS
kbps/TS
50
40
c1 FCA c2 FCA c3 FCA c4 FCA c1 DCA c2 DCA c3 DCA c4 DCA
30
20
10
5
10
15
20 25 30 Number of distributed MS’s
(a) Cell capacity
35
40
35
30
25
20
40
45
15
FCA DCA 5
10
15
20 25 30 Number of distributed MS’s
35
40
45
(b) Total capacity
Fig. 4.20. Results of scenario 3. The rate of asymmetry, UL:DL, is 2:2 in cell 1 and 1:1 in all other cells. The graphs show: for (a) the capacity in each cell in [kbps/TS] and (b) the total capacity in [kbps/Cell/TS]. The results labelled with ‘FCA’ are obtained by the fixed channel assignment procedure and the results labelled with c ‘DCA’ are these obtained from the novel centralised DCA algorithm. 2001 IEEE
the capacity results of the FCA technique and the DCA algorithm are almost identical.
130
Next Generation Mobile Access Technologies
(a) Expected values of the components of α at TS 1. In the case of FCA: α 1, 3 = α 1, 4 = α 2, 3 = α 2, 4 = α 1, 2 = α 3, 4 = 0.
(c) Expected values of the components of α at TS 3. In the case of FCA: α 1, 2 = α 1, 3 = α 1, 4 = α 2, 3 = α 2, 4 = α 3, 4 = 0.
(b) Expected values of the components of α at TS 2. In the case of FCA: α 1, 2 = α 1, 3 = α 1, 4 = α 2, 3 = α 2, 4 = α 3, 4 = 0.
(d) Expected values of the components of α at TS 4. In the case of FCA: α 1, 2 = α 1, 3 = α 1, 4 = α 2, 3 = α 2, 4 = α 3, 4 = 0.
Fig. 4.21. Scenario 3: The expected values of the components of α as a result of the novel DCA algorithm are depicted for all four TSs.
Results of scenario 4 The first two scenarios are with channel asymmetry in favour of the uplink. In scenario 4 the performance of the TS-opposing algorithm for channel asymmetry in favour of the downlink is investigated. The results are depicted in Figure 4.23. In cell 1 the TS-opposing algorithm, since it is operated on the uplink, can minimise interference with respect to only one TS out of the four available (TS 4). The TSs can be arranged such that no asynchronous TS overlap with respect to TS 4 exists. This is the reason why the capacity results in cell 1 do not differ greatly when using the DCA or FCA assignment strategies. The benefits due to the TSopposing algorithm are now experienced by cell 2–4 as can be seen in Figure 4.23. The capacity in cell 2 for 35 initially distributed users increases from
Centralised DCA algorithm using the TS-opposing idea TS1
TS2
TS3
TS4
cell 1
0
0
0
1
cell 2
0
1
0
1
cell 3
0
1
0
1
cell 4
0
1
0
1
131
1 := BS receives 0 := BS transmits (a) Scenario 4
Fig. 4.22. The initial TS assignment for scenario 4. The modification with respect to the previous scenario is highlighted.
70
60
55 60 50
45
kbps/cell/TS
kbps/TS
50
40
c1 FCA c2 FCA c3 FCA c4 FCA c1 DCA c2 DCA c3 DCA c4 DCA
30
20
10
5
10
15
20 25 30 Number of distributed MS’s
(a) Cell capacity
35
40
40
35
30
25
20
45
15
FCA DCA 5
10
15
20 25 30 Number of distributed MS’s
35
40
45
(b) Total capacity
Fig. 4.23. Results of scenario 4. The rate of asymmetry, UL:DL, is 1:3 in cell 1 and 1:1 in all other cells. The graphs show: for (a) the capacity in each cell in [kbps/TS] and (b) the total capacity in [kbps/Cell/TS]. The results labelled with ‘FCA’ are obtained by the fixed channel assignment procedure and the results labelled with c ‘DCA’ are these obtained from the novel centralised DCA algorithm. 2001 IEEE
50 kbps/Cell/TS to about 55 kbps/Cell/TS, which results in a relative gain of about 10%. Similar results can be obtained for cell 3 and cell 4. Again, the results of the expected values of αi,j are presented in Fig-
132
Next Generation Mobile Access Technologies
(a) Expected values of the components of α at TS 1. In the case of FCA: α 1, 3 = α 1, 4 = α 2, 3 = α 2, 4 = α 1, 2 = α 3, 4 = 0.
(b) Expected values of the components of α at TS 2. In the case of FCA: α 1, 2 = α 1, 3 = α 1, 4 = 1 and α 2, 3 = α 2, 4 = α 3, 4 = 0.
(c) Expected values of the components of α at TS 3. In the case of FCA: α 1, 2 = α 1, 3 = α 1, 4 = α 2, 3 = α 2, 4 = α 3, 4 = 0.
(d) Expected values of the components of α at TS 4. In the case of FCA: α 1, 2 = α 1, 3 = α 1, 4 = α 2, 3 = α 2, 4 = α 3, 4 = 0.
Fig. 4.24. Scenario 4: The expected values of the components of α as a result of the novel DCA algorithm are depicted for all four TSs.
ure 4.24. It is useful to highlight the following properties: 0 ≤ E TS1 (αi,j )
0.42 ≤ E
TS2
E
TS2
0.07 ≤ E
TS4
i = 1, · · · 3
(α1,2 ) ≈ E
(α1,2 ) ≈ E
TS2
TS4
j = 2, · · · 4
(α1,3 ) ≈ E
(α1,3 ) ≈ E
(α1,2 ) ≈ 1 − E
TS4
(α1,2 )
TS2
TS4
for all i = j ≤ 0.033 (4.71)
(α1,4 ) ≤ 0.93
(α1,4 ) ≤ 0.59
(4.72)
(4.73) (4.74)
Using the properties (4.71)–(4.74) it can be inferred that the DCA algorithm uses TS 2 and TS 4 in cell 1 to a great extent to improve capacity in cells 2–4. Note that this mechanism is opposite to that used in scenarios 1 and 2. In scenarios 1 and 2 TSs in cells 2–4 are most frequently changed by the DCA
Centralised DCA algorithm using the TS-opposing idea
133
algorithm to improve capacity in cell 1. In Figure 4.25 the basic mechanisms utilised by the new DCA algorithm with respect to scenario 4 are illustrated. TS1
TS2
TS3
TS4
Cell 1
0
0
0
1
Cell 2
0
1
0
1
Cell 3
0
1
0
1
Cell 4
0
1
0
1
1 := BS receives
:= pmax ≈ 59%
0 := BS transmits
:= pmax ≈ 3.3%
Fig. 4.25. This illustration shows the initial Tx/Rx configuration with respect to the BSs for scenario 4. The arrows highlight the TSs that changed most frequently. c The associated maximum probabilities are also shown.2001 IEEE
4.4 Conclusions For a single cell it was demonstrated that by using a TS-opposing technique in a TD-CDMA/TDD interface, capacity can be increased significantly compared with an equivalent FDD system. Assuming, for example, a population of two users per cell, with the same power levels, the capacity in the TDD cell is up to 48% greater than in an FDD cell. The capacity gain was reduced as the population in the adjacent cells increased. The reasons for this were as follows: first, the BS↔BS interference being about 15–20 times greater than interference from other-cell mobiles, and secondly the non-linear increase of own-cell interference. BS↔BS interference might be reduced by static shielding between two BSs (location in different rooms which provide extra shielding or antenna beamforming) or by a more complex downlink power-control algorithm. It was demonstrated that by decreasing the BS↔BS interference by a factor of 10, the gain due to the TS-opposing technique increased by a maximum of 118%.
134
Next Generation Mobile Access Technologies
The limitations of the approach used in this section were that outage due to high MS↔MS interference was not considered and that the optimisation was aimed at a single cell while neglecting the mutual impacts on the adjacent cells. However, the results obtained provided an upper bound on capacity gains when using the TS-opposing technique. The simplifications applied in the initial study were eliminated in a further, more complex investigation of a TS-opposing algorithm where the scope of operation was extended to a multiple-cell environment. In this context, a novel centralised DCA algorithm utilising the TSopposing principle was operated at the RNC. This resulted in a strategy to avoid interference, with asynchronous overlaps being the prerequisite for capacity improvements. Thereby it could be demonstrated that the existence of asynchronous TS overlaps in a TD-CDMA/TDD system did not result in system degradations. The very important implication of this is that channel asymmetry between neighbouring cells in a TD-CDMA/TDD system did not cause a capacity reduction as a consequence of same entity interference (MS↔MS and BS↔BS interference). Since arranging channel asymmetry is one of the most significant advantages of TDD, the elimination of the disadvantages of different rates of asymmetry within a TDD network is an important result. Furthermore, it was found that the total maximum capacity for channel asymmetry in favour of the downlink is greater than the reverse or even synchronous case. This is equally important since it is predicted that future data applications such as Web browsing will require more downlink than uplink capacity (Chaudhury et al., 1999). The DCA algorithm applied to the model of a cellular network did not achieve greater spectral efficiency than an equivalent FDD interface. However, the main advantage was that different channel asymmetries in neighbouring cells did not result in a significant capacity loss, regardless of the actual rate of asymmetry. Furthermore, it can be concluded that MS↔MS was not a severe problem with the system parameters used, but that the uplink and downlink in TDD could be strongly coupled, as demonstrated in (4.52).
5 Distributed DCA algorithm utilising the TS-opposing idea Harald Haas
5.1 Introduction In this chapter a distributed dynamic channel assignment (DCA) algorithm applicable for the TDD mode of the UMTS terrestrial radio access (UTRA) is presented. It is closely related to the DCA used in the DECT (digitally enhanced cordless telecommunications) system (Punt et al., 1998). Once again, the discovery made in Chapter 3 is exploited; that is, that for certain scenarios opposed synchronisation of TSs between neighbouring cells is advantageous. The new distributed DCA algorithm is supported by the results of the investigation in section 4.2. In this section it was demonstrated that synchronous transmission and reception between neighbouring cells may not yield the greatest capacity that is attainable in a single cell. It was found, however, that when the centralised DCA algorithm developed in section 4.3 was applied to multiple cells it was not feasible to fully exploit the potential gains revealed by the capacity analysis of a single cell. In this chapter it is demonstrated that by applying the novel distributed DCA algorithm, which utilises the TS-opposing idea, greater capacity can result than would be obtained by synchronous transmissions. Most importantly, this is shown to be valid for a TDMA-CDMA/TDD (TD-CDMA/ TDD) network, which accounts for full spatial coverage. Channel asymmetry is assumed to be arranged by code pooling rather than TS pooling (3GPP, TSG, RAN, 2000c). As a consequence of using the TS-opposing principle, a new method is required to separately measure the total interference from mobile stations (MSs) and the total interference from base stations (BSs). It is shown that this method can also be used to prevent cases of severe MS↔MS interference. This mechanism is incorporated into the new decentralised DCA algorithm presented here. 135
136
Next Generation Mobile Access Technologies
This chapter is structured as follows: in section 5.2 the problems are stated. In section 5.3 a novel TS assignment plan is presented, followed by a new decentralised DCA algorithm in section 5.4. Subsequently in section 5.5 a brief description of the simulation model and the methodology for measuring the system performance is described. A discussion of the results follows in section 5.6 before conclusions are drawn in section 5.7.
5.2 Problem formulation A channel in a TD-CDMA/TDD interface (used for UTRA-TDD) is characterised by a combination of a carrier frequency (FDMA component), time slot (TDMA component) and spreading code (CDMA component). Since only four carriers are available in the licensed UTRA-TDD frequency band, it is anticipated that any one operator will only be able to use one carrier which would reduce the channel characterisation to the combination of a time slot (TS) and spreading code. In this case the interfaces for DECT and UTRA-TDD are similar, the basic difference being that the FDMA component in DECT is replaced by the CDMA component in UTRA-TDD. In the DECT system a dynamic channel selection (DCS) procedure initiated by the MS is used. Due to the similarities between DECT and UTRA-TDD a distributed DCS incorporating the TS-opposing idea is investigated for the UTRA-TDD interface. This combined with the basic differences between UTRA-TDD and the DECT standard means that the following problems must be addressed: (i) The MS must be able to separately measure the interference contribution from other MSs (Imm ) and the interference component from neighbouring BSs (Ibm ) in order to exploit the TS-opposing idea. (ii) As the TS-opposing method is used, cases of severe MS↔MS may arise which need to be resolved by the DCA algorithm. (iii) If a cellular TDD system is considered, the same entity interference (BS↔BS and MS↔MS interference) may require a TSs reuse distances† of greater than 1. It is aimed to develop a decentralised DCA algorithm to avoid the necessity of cell reuse distances greater than one in order to maintain high spectral efficiency (Viterbi, 1995a). In order to solve the problems mentioned above, the combination of a fixed TS plan and a TS-opposing algorithm is proposed. A fixed TS assignment † The terminology used here follows the terminology introduced for FDMA/FDD systems (Lee, 1993) where the reuse of channels separated in the frequency domain is described. The same logic can be applied to channels that are separated in the time domain.
Distributed DCA algorithm utilising the TS-opposing idea
137
means that TSs for uplink and downlink direction are preselected in a systematic manner. Thus, the target is to enable the MS to selectively and easily measure each interference component Ibm and Imm . It could be argued that the preselection of the TSs reduces the flexibility to adjust the TDD mode to varying traffic loads in the up and downlink. However, in the UTRA-TDD mode a resource unit (RU) is specified by the combination of a TS, a code and a frequency carrier. Therefore, channel asymmetry can in principle be achieved in three dimensions: (a) in the TS dimension by means of TS pooling (multislot operation), (b) in the code dimension by code pooling (multicode operation) or (c) in the frequency dimension, which is not realistic for UTRA-TDD due to the low number of carriers. In (3GPP, TSG, RAN, 2000c) this problem is taken into account. The finding is that code pooling is more advantageous using the ‘unsatisfied users’ criterion in (ETSI 30.03, V3.2.0 (1998-04), 1998) than TS pooling for UDD (unconstrained delay data) packet data. This supports the concept of preselected TSs. In addition, areas that are known for high traffic imbalances can be catered for by defining more downlink TSs than uplink TSs. In the analysis carried out, for simplicity each user is assumed to be using the same service with one TS being allocated for uplink and downlink respectively. The total other-cell interference at any MS can be found as follows:
Im =
Mj L j=1 i=1
αj,k
Pcui , j ai,m Im m j
+(1 − αj,k )
Pcdj aj,m
,
(5.1)
Ib m j
where L is the number of surrounding cells, Mj is the total number of active users in the neighbouring cell j, and ai,m is the path loss between the user of interest, m, and the interfering user i. Similarly aj,m is the path loss between the user of interest, m, and the BS number j. The transmitted carrier power of user i in cell j is described by Pcui , j and Pcdj is the total carrier power transmitted by the BS in cell j. The TS synchronisation factor between cell j and cell k to which user m is allocated is modelled by αj,k . Symbols followed by the superscript ‘u’ are associated with the uplink channel; those followed by superscript ‘d’ are associated with the downlink channel.
138
Next Generation Mobile Access Technologies
The interference observed at any BS, b, can be calculated as follows: Ib =
Mj L j=1 i=1
(1 − αj,k )
Pcui , j ai,u
+αj,k
Im b j
Pcdj aj,u
.
(5.2)
Ib b j
From (5.1) it can be seen that the interference at any MS is composed of MS→MS interference multiplied by a synchronisation factor αj,k . Due to this synchronisation factor the MS→MS interference, Imm , may only be a fraction of the maximum possible MS→MS interference. The second type of interference experienced at any MS is BS→MS interference (Ibm ) for which a similar scaling applies as for (Imm ). Hence, the magnitude of each type of interference can be manipulated by varying αj,k , where j = 1, 2, . . . , L, with respect to cell k. It is interesting to note that Ibm j and Imbj also described as other-entity interference are coupled through αj,k . The same holds for Imm j and Ibbj , which are categorised as same-entity interference. This mechanism can be deduced from (5.1) and (5.2) which means that an interference reduction at the MS, for example, through adjusting αj,k also has an impact on interference at the BS. In the worst case, the interference at the BS increases despite the interference reduction at the MS. Hence, the interference at the opposite end, for example at the BS, of a communication link may not be minimised automatically if the interference is minimised by manipulating αj,k at one end, in this case the MS. This particular property is undesirable, but it is inherent in DCA algorithms (Ahlin and Zander, 1998, chapter 8). In the following sections a novel decentralised DCA for UTRA-TDD is investigated using a TS assignment plan that enables a mobile to exploit the TS-opposing mechanism.
5.3 TS assignment plan In the following a novel TS assignment scheme is presented which allows each TS to be used in each cell of a cellular TDD network with a frequency reuse distance of 1. In order to exploit the TS-opposing mechanism, however, fixed TS assignment patterns are introduced. In this way the TDD specific property of being able to use any TS for uplink or downlink traffic is exploited such that a TS–opposing algorithm can be operated locally. In (5.1) it has been shown that any location within the network can be characterised by an interference vector with one component being the interference resulting from all MSs. The second component describes the
Distributed DCA algorithm utilising the TS-opposing idea
139
sum of the total interference caused by the BSs. If the frames are synchronised these components are mutually exclusive. It merely depends on αj,k ∈ {0, 1} whether Ibm j or Imm j becomes effective. Furthermore, if these interference components are known a priori a new mobile entering the cell may be allocated to a channel with opposed TSs to all other cells if L L j=1 Ibm j > j=1 Imm j . However, two notable problems result from the interference minimisation process described. First, at the mobile receiver a composite interference signal from BSs and MSs is received. It is therefore difficult to measure the single interference components Imm j or Ibm j since the MS and BS entities within the surrounding cells transmit at the same time. Second, adjusting αj,k with respect to cell k will require the neighbouring cell j to alter the direction of transmission (uplink versus downlink, or vice versa). If the respective neighbouring cell j has L adjacent cells itself, this adjustment procedure might have undesirable implications for the interference in cell j and its L − 1 neighbouring cells. The novel TS assignment plan depicted in Figure 5.1 is intended to mitigate these problems. It is assumed that one frame is composed of seven symmetrical, full-duplex channels†. It is designed so that at least one pair of TSs is opposed to all neighbouring cells permanently. The illustration in Figure 5.1‡ shows a multiple of seven-cell clusters. Each cell is supposed to occupy seven subsequent pairs of TSs. In cell 0, for example, the first pair of TSs is opposed to all six neighbouring cells (cell 1 to cell 6), illustrated by the capital ‘X’. Similarly, in cell 1, the second pair of TSs (TS 3 and TS 4) are opposed to all their neighbouring cells. This is repeated in cell 2, and so on. Since one pair of TSs per cell is opposed to the corresponding TSs in all its neighbouring cells, Imm can be measured easily as it can be found that: αj,k = 1 for all j and k .
(5.3)
Hence, the interference measured at one of two TSs of the opposed fullduplex channel merely results from other MSs since: (1 − αj,k ) Ibm j = 0 for all j and k .
(5.4)
Similarly, the interference at the second TS of the opposed channel (which is only considered as a downlink channel during the measurement phase; during normal operation this is used for uplink traffic) results from the † Here the combination frequency/TS is referred to as a simplex channel. In UTRA-TDD each channel can be divided into subchannels by utilising the code domain. ‡ Figures 5.1, 5.4, 5.8 and Table 5.1 are reproduced with permission from: (Haas and McLaughlin, 2001b).
140
Next Generation Mobile Access Technologies
2X 7X 1X 6X 3X
7X
7X
1X 4X
6X
Cell 6
5X
Cell 5
6X
7X
Cell 1
2X Cell 0
1X Cell 4
3X 1X
6X
1X 6X
4X 5X
Cell 2
3X Cell 3
2X 7X
4X
1X
7X
5X nX := Time slot pair n is opposite to all 6 neighbour cells
4X 5X
2X
4X
3X
6X
5X
2X
3X
7X 4X
5X
2X
2X
3X
3X 1X
6X
4X 5X
Fig. 5.1. TS assignment plan. The ‘X’ indicates that the respective TS pair is c opposed. 2001 IEEE
neighbouring BSs because when considering this TS as a downlink TS it holds that: αj,k = 0 for all j and k .
(5.5)
Consequently, the interference conveyed merely results from the surrounding BSs since: (αj,k ) Imm j = 0 for all j and k .
(5.6)
The basic mechanism of the measurement procedure described above is illustrated in Figure 5.2. In this figure the scenario is shown where cell 0 uses an opposed channel (with respect to all neighbouring cells) and measures the interference component from the MSs of the neighbouring cells, and BSs respectively. In Figure 5.3 the actual TS assignment is depicted. This plan illustrates how the uplink and downlink slots are distributed throughout a seven-cell cluster. This cluster can be repeated as often as required in order to achieve
Distributed DCA algorithm utilising the TS-opposing idea
MS: Mobile station BS: Base station
141
Reception Transmission Reception only during measurement
MS
Interference
BSn Cell 0
MS
MS BS0
Fig. 5.2. The mechanism of measuring the interference from BSs and MSs at the opposed channels.
full spatial coverage. It can be seen that in cell 0, TS11 (the superscript refers to the channel number whereas the subscript refers to the TS number of the respective channel) and TS12 are opposed to all other adjacent cells. In cell 1 it is TS21 and TS22 , in cell 2 TS31 and TS32 , etc. It can be found that the TS assignment plan illustrated in Figure 5.1 causes at least one pair of TSs to overlap asynchronously to one of its adjacent cells. Therefore, the term ‘quasi-synchronous’ channels is introduced. Once again, this can be illustrated with the aid of Figure 5.3. As an example, the TSs of channel 2 (Ch 2), TS21 and TS22 , in cell 0, are opposed to cell 1, but synchronous to all other cells j = 2, 3, . . . , 6. The same mechanism can be found for all other channels in cell 0, except channel 1. Thus, in each cell there are six quasi synchronous channels and one channel with opposed synchronisation to all other cells.
142
Next Generation Mobile Access Technologies
Ch 2
Ch 1 TS
1
1
TS
1
2
TS
2
1
TS
2
3
TS
1
Ch 7
Ch 4
Ch 3 2
TS
3
2
TS
4
1
TS
4
2
7
TS7 TS2 1
Cell 0 Cell 1 Cell 2 Cell 3 Cell 4 Cell 5 Cell 6 BS transmits BS receives
The unfilled arrow marks an X-TS
Fig. 5.3. TS configuration in a seven-cell cluster.
The interference measured at any of the ‘quasi-synchronous’ channels can be found as: Ibm
=
5
Ibm i + Imm n ,
(5.7)
i,i=n
Imm
= Ibm n +
6
Imm i ,
(5.8)
i,i=n
where n indicates the single, opposed-duplex channel. In order to resolve severe MS↔MS interference it is necessary to determine Imm n and Ibm n respectively. This can be achieved if an idle frame is introduced. This idle frame is also necessary to enable the opposed pair of TSs to track the measurements of Imm and Ibm after the resource units (RUs) have been allocated. The duration of a multi-frame including one idle frame will depend on the maximum specified time for ‘service establishment’ and the maximum speed of a mobile in the TDD deployment environment. Let the maximum speed of a mobile be 3 km/h and assume that the shadowing will not change its characteristic within a range of about 1.5 m (obtained using the correlation model in (Klingenbrunn and Mogensen, 1999) and assuming an average transmitter–receiver distance of 25 m and a correlation coefficient of r = 0.95), the time interval within which an idle frame is required
Distributed DCA algorithm utilising the TS-opposing idea
143
will be 1.8 s. With a frame duration of 10 ms, the idle frame would need to occur every 180th frame. If a higher mobility environment is considered with, for example, a maximum speed of 50 km/h under the same propagation conditions, the idle frame would have to occur after every 20th frame. In the event of an idle frame in cell n, which is the only cell with opposed synchronisation, (5.7) and (5.8) become: Ibm i d l e
=
5
Ibm j ,
(5.9)
Imm j .
(5.10)
j,j=n
Imm i d l e
=
5
j,j=n
Subtracting (5.9) from (5.7) and (5.10) from (5.8) gives: Imm n Ibm n
= Ibm − Ibm i d l e ,
= Imm − Imm i d l e ,
(5.11) (5.12)
and consequently: Imm = Imm i d l e + Imm n ,
(5.13)
Ibm = Ibm i d l e + Ibm n .
(5.14)
5.4 TS-opposing algorithm The concept presented in the previous section represents the foundation for the decentralised DCA algorithm depicted in Figure 5.4. This DCA algorithm is operated as a fast DCA according to (3GPP, TSG, RAN, 2000c). Due to the novel TS arrangements, the DCA algorithm is enabled to exploit an additional degree of freedom which is generated by the opposed TSs. This means that an arbitrary location is characterised by an interference vector of two components as depicted in Figure 5.5. The property of each interference component is that the source of the interference (BS or MS) is the same. The DCA algorithm basically chooses a channel for which the lower interference component applies. Moreover, the DCA algorithm (in particular the second condition) in Figure 5.4 ensures that MS→MS interference is always smaller than BS→MS interference. Hence the problem of MS↔MS interference inherent in a TDD system is eased so that its effects are less of an issue than BS→MS interference. Each time a MS enters the network Ibm and Imm are measured. First, it is calculated which of Imm and Ibm is smallest, for the channel with opposed
144
Next Generation Mobile Access Technologies
Begin
new user, k, enters the network
determine Ibm and Imm for the opposed channel n
if Imm < Ibm
no
yes assign new user, i, to opposed channel n
End
determine Imm and Ibm for quasi synchronous ch. l try next channel l = l + 1 if Ibmn < Immn and Ibm < Imm
no
yes l = max(channels)
This ensures that severe MS-MS interference is avoided.
no
assign new user, i, to channel l yes End
assign new user, i, to channel for which I = min(Ibm ) holds
End
Fig. 5.4. Decentralised DCA algorithm exploiting the TS-opposing technique. c 2001 IEEE
synchronisation. If Imm is smallest, the MS requests to be allocated to the opposed channels provided that spare capacity is available. If spare capacity is not available, however, or Imm is greater than Ibm , the MS assesses one of the ‘quasi-synchronous’ channels. This means that exactly one neighbouring cell, n, transmits and receives in opposition to the cell of interest (COI). The interference components conveyed by cell n, Imm n and Ibm n , are calculated as described in section 5.3. If Ibm n < Imm n and Ibm < Imm the MS is allocated to the observed channel. In addition, this mechanism ensures that severe MS↔MS is prevented. If the two conditions are not fulfilled, however, the same procedure is carried out for the next ‘quasi synchronous’ channel, l + 1, until the end is reached. If the two conditions above are still not fulfilled, the MS allocates itself to that ‘quasi-synchronous’ channel for which I = min (Ibm ) holds.
Distributed DCA algorithm utilising the TS-opposing idea power
145
Im m 3 Ib m 3
Ib m 1 Im m 4
Ib m 2
min(I)
Im m 6
Im m 1
Ib m 4
Im m 2
TS1
Ib m 6
TS2
TS3
TS4
TS6
time
Fig. 5.5. Interference vectors for an arbitrary location within cell 0. It is assumed that a channel consists of at least two TSs, one for the uplink and one for the downlink.
5.5 System model The simulation platform as described in section 3.4 is used to carry out Monte Carlo simulations of the DCA algorithm depicted in Figure 5.4. The cells will be populated under the following conditions: • • • •
non-optimal power control; mobiles allocated to the best serving BS; handover margin of 5 dB; correlated shadowing according to a model by Klingenbrunn (Klingenbrunn and Mogensen, 1999); • symmetrical speech service; • seven-cell cluster with the COI in the centre. In a CDMA system it is important that a MS is allocated to the best serving BS (in terms of received signal power) rather than the closest BS as otherwise the interference may increase by up to a factor of 20 (as reported in (Viterbi and Viterbi, 1994)). In this research, therefore, a simulation model is created where the MSs are allocated to a BS based on the minimum path-loss criterion. In a real system, however, a MS cannot always be assigned to the BS offering the lowest path loss due to handover imperfections. Thus, a handover margin as described in section 3.3.3 is considered. Furthermore, when assigning a MS to the best-serving BS regardless of its location, situations may arise where a MS actually is located outside the cell from which it is served due to lognormal shadowing. This behaviour is accounted for by the introduction of a second cell radius, d1m a x , as depicted in Figure 5.6. With the radius d1m a x the scope of a cell in terms of coverage can be expanded to cater for more realistic scenario in which the MSs or
146
Next Generation Mobile Access Technologies
BSs do not have knowledge about the actual cell boundaries. Instead, the path loss is the factor that determines which BS is to be used.
Mobile station Base station
R
d1 m a x
d0 COI
Fig. 5.6. Simulation environment: the DCA algorithm is operated at the cell of interest (COI). The first tier of cells is equally populated and the handover regions (grey shaded area) are considered so that MSs can be allocated to the best-serving BS.
Again, as in Chapter 4, the indoor office test environment described in (ETSI 30.03, V3.2.0 (1998-04), 1998) is used which gives the following relationship for the path loss, a, in terms of the transmitter–receiver separation distance, d, the number of floors, p, and the lognormal random variable (RV), ξ, to model shadowing effects :
a = 37 + 30 log10 (d) + 18.3 p
p +2 −0.46 p +1
+ ξ [dB] .
(5.15)
Distributed DCA algorithm utilising the TS-opposing idea
147
5.5.1 Uplink For an arbitrary user i the bit-energy to interference ratio ,εui , at the BS may be denoted as: εui
P u ui R
= M
P u uj j:j=i W
+ Iou + N0
,
(5.16)
where P uui is the received signal power from the desired user i, R is the information bit rate, W is the total bandwidth, Iou is the other-cell interference density and N0 is the thermal noise density. It has been demonstrated in (Viterbi and Viterbi, 1993) that εui needs to be considered as a RV due to multipath propagation and power control imperfections. In addition, it was shown that εui can be approximated by a lognormal RV with mean μuε and standard deviation σεu (Viterbi and Viterbi, 1993). Each time a new user enters the cell, the bit-energy to interference ratio of each user will be affected and some MSs may not be able to maintain the required threshold, and hence will experience outage or a degradation in performance. Therefore, dynamic power control is assumed to maintain the required εui for each user i. Solving (5.16) for P uui and multiplying with the respective path loss yields the carrier power for user i as follows: ⎛ ⎞ M u P u j (5.17) + Iou + N0 ⎠ . P cui = ai εui R ⎝ W j:j=i
From (5.17) it can be seen that any change in the desired power, P uuj , affects the transmitted power of all other MSs. By substituting γ = R εdi /W and Ib = W (Iou + N0 ) and using matrix notation, (5.17) has been solved in section 4.3.1 resulting in (4.42).
5.5.2 Downlink It has been demonstrated in (Prasad et al., 1993) and (H¨ am¨al¨ainen et al., 1997) that downlink power control in a cellular CDMA network is required in order to reduce interference, in particular for MSs located at outer regions of a cell. In addition, in (St¨ uber and Yiin, 1991) it has been reported that downlink power control is only effective if it is combined with some handover strategy. For a TD-CDMA/TDD cellular network in general, and UTRA-TDD in particular, downlink power control seems to be required even more as BSs can interfere with each other and thereby affect the uplink performance (this
148
Next Generation Mobile Access Technologies
is more precisely described in section 4.3.1 with (4.52)). In an investigation by Povey (Holma et al., 1999) it was found that BS↔BS interference is a severe problem in a TD-CDMA/TDD system. This, however, is primarily because only a simple downlink power control mechanism was assumed in that study. In this analysis two different downlink power control algorithms are applied in order to assess their impact on system performance when using the new decentralised DCA: • The first downlink power control approach is C/I based in which the code power for each mobile is adjusted so that εdi ≥ εd for all i = 1, · · · , M , where εd is the required bit-energy to interference ratio at the MS. • The second downlink power control algorithm is related to the distancebased algorithm as proposed in (Prasad et al., 1993) and (Lee, 1993, Chapter 9). Since distance based algorithms only function in low shadowing environments a modification is made to incorporate lognormal shadowing. Hence, the carrier power in the downlink is determined by that MS which experiences the highest path loss. The same code power is then applied to every other user. This algorithm is more precisely described on page 68.
5.5.3 Capacity and Blocking Definitions The novel TS assignment as described in section 5.3 ensures that every cell has the same number of quasi-synchronous and opposed channels. Due to this symmetry, the performance of the DCA algorithm is evaluated for a single cell, the COI, whilst the neighbouring cells are being uniformly populated. The traffic load in the first tier of cells is varied. For a constant load in the neighbouring cells, the capacity (number of instantaneously active MSs) in the COI is determined by successively adding new users. The strategy as to how a new MS is allocated is such that an already established connection has higher priority than a new access attempt (resulting in user blocking in favour of outage). Therefore, if for any i, i = 1, 2, . . . , M it is found that P cui > P cumax or P cdi > P cdmax , the new MS cannot be assigned and is blocked for the respective TSs (function of the admission control). Due to the interference limited nature of CDMA a hard limitation of the number of users seems inappropriate. Therefore, in this analysis the ‘maximum cell load criterion’ (Figure 5.4) is defined as follows: the cell is considered to be fully loaded if a predefined number of consecutive blocking events have occurred. The permitted number of consecutive blocking events can be considered as a measure of the quality of service (QoS) and is set to 10. This
Distributed DCA algorithm utilising the TS-opposing idea
149
BEGIN Add new user, i, to the COI if user i is blocked CellLoadMaxIdentifier++ increment blocking counter, Bl: Bl = Bl + 1 if CellLoadMaxIdentifier == 10 LoadStatus = MaxCapacityReached Bl = Bl − 10 end if else CellLoadMaxIdentifier=0 M =M +1 end if END Fig. 5.7. Blocking and maximum cell load criterion.
parameter can be used to trade-off QoS and capacity. The mechanism to determine the maximum capacity and the resulting blocking is described with the algorithm in Figure 5.7. From the total number of users obtained by the procedure described by the pseudo code above, the average capacity in bit/(s·TS·cell) can be calculated. It is assumed that the data rate of each MS is the same. Then, the average capacity yields: R Q i+1 Mi , (5.18) C= Q where Q is the number of Monte Carlo runs and Mi is the maximum number of users for a single Monte Carlo run. Similarly, the average blocking is determined as: Q Bli Bl = i=1 . (5.19) Q It must be stressed that blocking as defined above is dependent on the total maximum number of distributed MSs at the COI. The maximum number of distributed users at the COI is held constant, but it is varied in the first tier of cells. The average capacity and blocking figures are used to compare the performance of the new DCA strategy with a system having a TS configuration such that all entities throughout the entire network transmit and receive at
150
Next Generation Mobile Access Technologies
the same time. This means that only other entity interference is present and thus this configuration represents an equivalent FDD system. The parameters used for the simulation are based on (ETSI SMG2, 1998), but modified as a consequence of the UTRA-TDD standardisation process. The principle parameters are summarised in Table 5.1. Table 5.1. Simulation parameters used to assessing the performance of the combination of the novel TS assignment plan and the decentralised DCA c algorithm. 2001 IEEE Parameter
Value
TDD cell radius
50 m
Max MS TX power, P cum ax
15 dBm
Max BS TX power, P cdm ax
21 dBm
Mean of ε, μuε
3.5 dB
Std. dev. of ε, σεu
2.5 dB
Std. dev. of lognormal shadowing, ξ
10 dB
Information bit rate, R Bandwidth, W
16 kbps 4.68 MHz
Handover margin
5 dB
Total effective thermal noise density, N0 Number of floors, p, in (5.15)
169 dBm 2
5.6 Results When a new MS enters the network the aim is to assign it to the bestserving BS. Due to handover imperfections and lognormal shadowing the best serving BS may not be located in the cell in which the MS resides. Two scenarios are investigated in order to account for this mechanism, . The first scenario is where the MS allocated to a cell is bound within a circle specified by the radius d1min (Figure 5.6) which has an area approximately equal to that of the hexagon, e.g. a hexagon with radius of 50 m corresponds to a circle with radius d1min ≈ 45 m. The second, more realistic, scenario is that a MS can be located outside the respective cell, but still be served by the BS in the centre. This is arranged by setting d1min such that it exceeds the cell radius. Thus, in the second scenario d1min is set to 75% of the distance
Distributed DCA algorithm utilising the TS-opposing idea
151
between two BSs, which for a cell radius of 50 m results in d1min ≈ 65 m. The results of the DCA algorithm for these scenarios and two different DL (downlink) power control methods are depicted in Figure 5.8. 85
95 Opposing algorithm Equivalent FDD scenario (all TS aligned)
Opposing algorithm Equivalent FDD scenario (all TS aligned)
90
Average Capacity [kbps/TS/Cell]
Average Capacity [kbps/TS/Cell]
80
75
70
65
60
55
85
80
75
70
2
3
4 5 Number of interfering users per cel l
6
65
7
(a) Path loss based DL power control, d1min ≈ 45 m.
2
3
7
90 Opposing algorithm Equivalent FDD scenario (all TS aligned)
Opposing algorithm Equivalent FDD scenario (all TS aligned)
85
Average Capacity [kbps/TS/Cell]
75
Average Capacity [kbps/TS/Cell]
6
(b) C/I based DL power control, d1min ≈ 45 m.
80
70
65
60
55
50
4 5 Number of interfering users per cel l
80
75
70
65
2
3
4 5 Number of interfering users per cel l
6
7
(c) Path loss based DL power control, d1min ≈ 65 m.
60
2
3
4 5 Number of interfering users per cel l
6
7
(d) C/I based DL power control, d1min ≈ 65 m.
c Fig. 5.8. Average capacity results of the decentralised DCA algorithm. 2001 IEEE
For all cases investigated the combination of the novel TS plan and the new distributed DCA algorithm performs better than an equivalent FDD scenario. This is a most important result. It was shown for a single cell scenario (section 4.2) that the capacity of a CDMA/TDD interface can, in principle, be greater than that of an equivalent CDMA/FDD interface. However, the disadvantage was that the DCA algorithm changed the uplink and downlink assignment of a TS, which, in turn, effected the interference in the neighbouring cells. This disadvantage is eliminated here by the use of fixed TS assignment. For the simple DL power control scheme the proposed method increases
152
Next Generation Mobile Access Technologies
the capacity by up to 8%. This is reduced to 5% if the more sophisticated DL power control method is applied. Therefore, not only does the decentralised DCA algorithm achieve better capacity results, but it also can compensate for a poor DL power control algorithm. Note, as the DCA is operated in a decentralised manner, the decision of which TSs are to be used is made at a single end of the communication link ignoring the interference situation at the other end (at the BS). In addition, if the maximum path loss based DL power control method is applied, it can be derived from the results in section 3.3.3 that the BS↔BS interference at the opposed TSs will on average increase. This means that for the opposed pair of TSs the probability P r(Imm < Ibm ) decreases. As a consequence the relative capacity improvement associated with the new DCA algorithm is expected to be less compared with the results of the C/I-based DL power control technique which minimises the transmitted power on the downlink. As the results reveal the opposite (8% improvement for the simple DL power control compared with 5% for the C/I-based DL power control algorithm), it can be concluded that the interference caused at the MS by a greater downlink power in the quasi-synchronous case is more detrimental than the greater BS↔BS interference at the opposed TSs. This is obvious because, with the cell topology applied, the distance between an interfering BS and any MS is always smaller than the BS–BS distance. Nevertheless, the total capacity improvement of the C/I-based DL power control algorithm, in all cases, is significantly greater than that of the simple path loss based DL power control technique. As an example, consider four interfering MSs using the decentralised DCA algorithm. The capacity as shown in Figure 5.8(a) and Figure 5.8(b) for the C/I-based DCA algorithm is about 18% greater. This highlights the importance of a well designed DL power control algorithm as was also found in Chapter 3. In the case that d1max is increased from 45 m to 65 m, similar behaviour is found. In this case, however, the absolute capacity is reduced by about 5–10% due to an increased interference. Although in this case the MSs are still assigned to the best-serving BS (only limited by the handover margin) the probability of a greater path loss to the serving BS increases, resulting in a greater TX power and thus more interference. Note that in all scenarios investigated, the decentralised DCA algorithm performs better than an equivalent FDD system. This result is significant as a novel TS assignment plan also enables the TDD system to account for full spatial coverage. The blocking results are depicted in Figure 5.9. The noticeably high blocking is a consequence of the offered load being seven simultaneously active users in the COI. In addition, the high blocking can be attributed to
Distributed DCA algorithm utilising the TS-opposing idea 20
153
14 DL blocking (TS opposing) UL blocking (TS opposing) DL blocking (all TS aligned) UL blocking (all TS aligned)
18
DL blocking (TS opposing) UL blocking (TS opposing) DL blocking (all TS aligned) UL blocking (all TS aligned)
12
Number of blocked attempts (average)
Number of blocked attempts (average)
16
14
12
10
8
10
8
6
4
6 2 4
2
2
3
4 5 Number of interfering users per cel l
6
0
7
(a) Path loss based DL power control, d1min ≈ 45 m.
2
3
4 5 Number of interfering users per cel l
6
7
(b) C/I-based DL power control, d1min ≈ 45 m.
30
20 DL blocking (TS opposing) UL blocking (TS opposing) DL blocking (all TS aligned) UL blocking (all TS aligned)
DL blocking (TS opposing) UL blocking (TS opposing) DL blocking (all TS aligned) UL blocking (all TS aligned)
18
25
Number of blocked attempts (average)
Number of blocked attempts (average)
16
20
15
10
14
12
10
8
6
4 5 2
0
2
3
4 5 Number of interfering users per cel l
6
7
(c) Path loss based DL power control, d1min ≈ 65 m.
0
2
3
4 5 Number of interfering users per cel l
6
7
(d) C/I-based DL power control, d1min ≈ 65 m.
Fig. 5.9. Blocking results of the decentralised TS-opposing algorithm.
the use of the ‘blocking in favour of outage’ strategy. It is interesting to note that uplink blocking in each scenario is higher than downlink blocking and that it increases with the load in the neighbouring cells, whereas downlink blocking is not a strong function of the number of distributed users. This is primarily due to the uplink being a multi-point to single-point transmission scheme and, in contrast, the downlink being a single-point to multi-point transmission scheme. Therefore, if the interference at the BS is at a high level all users are affected and this increases blocking as defined here. On the other hand, a single MS might require the BS to transmit high power, but the probability that this is the case for all randomly located MS is small and thus blocking is less likely than in the uplink direction. These results confirm the findings by other researchers that the uplink in a CDMA system limits the capacity (Gilhousen et al., 1991; Viterbi and Viterbi, 1993;
154
Next Generation Mobile Access Technologies
Shin et al., 1999). However, the uplink and downlink cannot generally be treated as independent if TDD is employed. Since uplink and downlink are on the same frequency carrier both directions are coupled depending on synchronisation and asynchronous TS overlaps. This effect can be studied using the blocking results. If, for example, the results in Figure 5.9(a) and Figure 5.9(b) are compared it can be found that, although only the downlink power control algorithm is changed, the uplink blocking is greatly affected by the performance of DL power control. The reason for this is that the TS assignment plan used generates asynchronous TS overlaps which yield the strong coupling between uplink and downlink direction. However, it needs to be stressed that blocking and capacity are not independent, and as the capacity increases the probability of blocking is also affected. This is also why uplink blocking also decreases with a better DL power control algorithm in the case where all TSs are quasi-synchronous. As expected, and demonstrated in Figure 5.9, the downlink blocking is significantly reduced by the decentralised DCA algorithm developed in this chapter. However, it also becomes clear that the total blocking is dominated by the uplink blocking, which is about 4–10 times greater. Due to the nature of the decentralised DCA algorithm used, interference at the BS is not explicitly reduced. Therefore the uplink blocking is not improved. This proves to be a remaining problem. In a further experiment it is assumed that each user requests a symmetric 64 kbps service. In this case a maximum of two users per cell and TS can be supported. The capacity results are summarised in Table 5.2 and blocking in Table 5.3. Table 5.2. Average capacity for the case of two users per cell/TS, each with a data rate of 64 kbps. d1min ≈ 45 m path loss based
d1min ≈ 45 m C/I-based
d1min ≈ 65 m path loss based
d1min ≈ 65 m C/I-based
DCA
equiv. FDD
DCA
equiv. FDD
DCA
equiv. FDD
DCA
equiv. FDD
118.5
119.4
118.2
118.2
111.4
111.7
111.1
111.4
The blocking in the uplink (UL) is presented in the first row of Table 5.3. Similarly, the row marked with DL presents the blocking in the downlink. As own-cell interference decreases with the number of users, the total capacity increases. Moreover, for the same reason, the effect of a tighter downlink
Distributed DCA algorithm utilising the TS-opposing idea
155
Table 5.3. Average number of users blocked for the case of two users per cell/TS, each with a data rate of 64 kbps. d1min ≈ 45 m path loss based
d1min ≈ 45 m C/I-based
d1min ≈ 65 m path loss based
d1min ≈ 65 m C/I-based
DCA
equiv. FDD
DCA
equiv. FDD
DCA
equiv. FDD
DCA
equiv. FDD
UL
5.1
3.5
4.5
3.8
6.7
5.8
6.0
6.0
DL
0.1
0.7
0.4
0.7
0.2
1.0
0.3
1.1
power control algorithm is reduced. The gain of the proposed DCA algorithm with respect to capacity is diminished for the case of two high data rate users per cell. Once again, the reason that the DCA algorithm for a symmetric data rate of 64 kbps performs worse than for a symmetric rate of 16 kbps can predominantly be attributed to outage in the uplink due to a decreased processing gain. Again from Table 5.3 it can be seen that uplink blocking is greater in the case of opposed TSs, but downlink blocking is significantly reduced.
5.7 Conclusions A new distributed DCA algorithm exploiting the TS-opposing technique has been investigated. In addition, a fixed TS assignment was developed. Special emphasis was placed on ensuring ease of implementation. It was predominantly assumed that channel asymmetry is achieved by code pooling rather than TS pooling. The very important result is that this technique can result in higher capacity than an equivalent FDD system, even for a multiplecell environment. A further important result is that the DCA algorithm ensures that severe MS↔MS is avoided. The proposed DCA algorithm helps to abate a tight requirement on DL power control, but the relative capacity improvement obtained by using the DCA algorithm is less than the improvement that can be obtained by using a C/I-based DL power control algorithm instead of a path loss based DL power control method. Furthermore, the uplink and downlink direction are strongly coupled in a TD-CDMA/TDD interface if asynchronous TS overlaps occur. Therefore, an isolated treatment of either direction may produce unrealistic results. A possible drawback is that the decentralised DCA algorithm does not
156
Next Generation Mobile Access Technologies
minimise the interference at the BS. However, it has been demonstrated that the impact of interference at the BS can be more severe than at the MSs since several links are affected simultaneously. Hence, the uplink direction prevents to fully exploit the potential interference reduction capabilities of the decentralised DCA algorithm. This important finding may be used in further investigations of decentralised DCA techniques which make use of the novel TS-opposing principle. It may, for example, be interesting to study the effects of a combined decentralised and centralised DCA algorithm where the decentralised DCA algorithm is basically the same as presented in this chapter. When using the same algorithm at the MS and BS, conflicting decisions regarding the best channel selection may arise which would need to be arbitrated by an additional algorithm. If own-cell interference is reduced by joint-detection, for example, it is anticipated that the performance of the DCA concept presented in this chapter will be further improved. This may be addressed in further research.
6 UTRA-TDD Opportunity-Driven Multiple Access (ODMA) Tom Rouse Stephen McLaughlin
In this and the following two chapters the focus moves away from networks which are controlled centrally by a base station to a hybrid cellular network which permits cellular operation as well as peer-to-peer operation. Essentially we consider multi-hop wireless networks based on opportunity-driven multiple access (ODMA) which will be shown to reduce the overall transmission power in a system, to be resilient to shadowing and to potentially increase the coverage compared with single-hop transmission. However, for simple receivers and low user density, the actual capacity of UTRA-TDD may be marginally reduced from the maximum non-relaying capacity. This chapter begins the study of ODMA based systems by analysing the implications of relaying in a cellular scenario compared to a conventional nonrelaying system. Initially the interference is analysed by investigating the effect of reduced transmitted power resulting from reduced path loss for a link. The effect of shadowing is considered and it is shown that a relaying system is able to benefit from increased zero mean lognormal shadowing by utilising the diversity of paths available. A correlated shadowing model is developed from a previous model considering both distance and angle of arrival (Klingenbrunn and Mogensen, 1999) to include the shadowing correlation between all transceivers, as they may all be available to receive in a relaying environment. It is shown that while this affects the interference pattern the perturbation is not significant. Further analysis is made of the impact upon the capacity of ODMA in relation to the coverage of a cell comparing relaying performance to the analysis made for a non-relaying system by Veervalli (Veeravalli and Sendonaris, 1999). It is shown that, after the coverage limit of non-ODMA UTRA-TDD has been reached, ODMA will provide enhanced coverage. As the number of calls and/or quality of service is decreased the cell coverage can be increased beyond conventional coverage capacity trade-offs, allowing operators a far greater degree of flexibility. 157
158
Next Generation Mobile Access Technologies
6.1 Introduction Opportunity-driven multiple access (ODMA) is a misnomer as it is not a true multiple-access technique in the sense that FDMA, TDMA and CDMA are. It is actually a relaying protocol which has the potential to provide benefits such as reduction of transmission power, overcoming dead spots, and a more even distribution of interference with reduced mean received power. The basic principle underlying ODMA is the argument that it is more efficient to break the path into smaller hops, as shown in Figure 6.1, in contrast to the conventional approach, where a mobile station (MS) in a cell communicates directly with the base station (BS), or vice versa, in a single line-of-sight transmission. This is achieved by making use of other MS in the cell to relay the signal. The optimal routing is calculated using intelligence in the MS and BS to try and achieve the minimum total path loss for the transmission. Originally the UTRA-TDD standard included provision for ODMA (3GPP, TSG, RAN, 2000a), as a modification of a patent by Salbu Pty. Ltd. (Salbu Research and Development (Proprietary) Ltd, 1978), although the implementation was far from finalised. The main features of the standard covered signalling slot allocation and methods for building neighbour lists, but a routing protocol was noticeably absent. ODMA has now been removed from UTRA-TDD, with suggestions that it was dropped over concerns of increased power consumption, especially from users not involved in calls being used as relays, and the unfinished state of the protocol exacerbated by a lack of resources available within companies desperate to recoup the expenditure of 3G. This chapter will show that the issue of increased battery drain for non-calling users is not a problem as all scenarios examined only consider relays available if they are already using one or more of the TDD time slots. For those involved in calls an increase in battery life will result on average for all users if only the power transmitted by that mobile is considered. Users close to the BS may lose out, but for non-relaying they benefit from the lowest required power. MSs at the edge of cells will see the greatest benefit, with battery life being more consistent and longer on average. Several papers have covered the overall power consumption of a MS, additional factors to transmit power being mainly attributable to CPU cycles. For this book, because of the general trend in reducing feature size in ICs consistent with Moore’s law (Moore, 1965), and the commensurate reduction in voltage, and hence power, it is considered that transmitted power will be the dominant factor on battery life by the time any 3G evolution system is released.
UTRA-TDD Opportunity-Driven Multiple Access (ODMA)
159
It was shown by Harrold and Nix (Harold and Nix, 2000b)(Harold and Nix, 2000a) that a relaying system with distributed intelligence can show an average reduction of 21 dB in required transmission power or increased coverage (Harold and Nix, 2000b), and that under certain circumstances, with a sufficient density of relaying users there may be a capacity enhancement over conventional TDD (Harold and Nix, 2000a).
Mobile Stations
Base Station
Obstacle
Fig. 6.1. ODMA scenario showing routing with path broken into shorter links, and avoiding shadowing.
To begin with we will review the background to UTRA-TDD ODMA and then follow its evolution to the point where ODMA was dropped from the UTRA-TDD standard. The structure then investigates path loss, shadowing, and finally coverage-capacity trade-offs. In the former, path losses are examined in the context of a simple two-hop routing algorithm to minimise the path loss. The resulting interference pattern throughout the cell is then compared with a non-relaying system and theoretical maximum gains calculated. The interference pattern is analysed for different values of shadowing variance. It is shown that whilst in the conventional system shadowing
160
Next Generation Mobile Access Technologies
increases overall interference, a relaying system is able to exploit paths improved by zero mean log-normal shadowing, as it selects the minimum path loss route, hence choosing those benefiting from reduced path loss due to shadowing. The second half of the chapter analyses the capacity coverage trade-off within ODMA in comparison to a conventional TDD system. The connectivity of ODMA is sufficiently different to traditional CDMA MS↔BS communication that it is not reasonable to use the conclusion of Veeravalli and Sendonaris (Veeravalli and Sendonaris, 1999) that for a maximum transmission power the coverage of a cell is inversely proportional to the number of users. This information may allow for a coverage based admission control. As it is likely that the greatest bottleneck in an ODMA relay link will be the final MS↔BS hop, it may be advantageous to allow cell breathing, or dynamic cell geometry through BS assignment. This would provide for a more even loading of BSs, optimising system requirements.
6.2 UTRA-TDD ODMA Background ODMA was first proposed by the South African company Salbu Research in a 1978 patent (Salbu Research and Development (Proprietary) Ltd, 1978). In its initial form the motivation was for a packet based radio system using TDMA principles. Together with Vodafone, and to a lesser extent Siemens, Salbu introduced ODMA to ETSI SMG2 in 1996 as a proposition for the 3rd Generation (3G) mobile system (Vodafone Ltd., 1997b) (Vodafone Ltd., 1997a) (Vodafone Ltd. and Salbu (pty) Ltd., 1996) (Vodafone Ltd. and Salbu (pty) Ltd., 1997a) (Vodafone Ltd. and Salbu (pty) Ltd., 1997b). The idea on its own was a poor contender compared to the several CDMA variants and was never going to succeed in its own right. The epsilon group, however, carried out investigations into its feasibility, with the outcome that ODMA was proposed to other working groups as an extension or enhancement to their existing access methods (Alpha Concept Group ETSI SMG2, 1997) (Delta Concept Group ETSI SMG2, 1997). The proposal only required hooks to be put in and ODMA could be enabled or disabled by the operator, hence implementation was not required in the initial hardware, delaying roll out. General opinion was that ODMA may as well be included even though many issues were still unresolved.
UTRA-TDD Opportunity-Driven Multiple Access (ODMA)
161
6.2.1 Features The original patent (Salbu Research and Development (Proprietary) Ltd, 1978) describes a basic routing strategy. First a neighbour list is built up, this can be either from listening to other mobiles or by probing. The latter is a request for information from surrounding mobiles. A request is sent out for identity and quality of received signal (path loss, noise level). There is no ability to tell if the transmission is going in the optimal direction, the strategy is to transmit to the MS with the best signal quality that has communicated with the BS. Some power reduction seems to be the only requirement, but it is suggested that the route will be known up to three hops ahead. The proposals to the Delta Concept Group (Delta Concept Group ETSI SMG2, 1997) add to this basic idea in relation to UMTS. The neighbour list is required to contain at least five entries. If this is not met after transmitting at the lowest power and highest data rate the power is increased. If the condition is still not met at maximum power, the next lowest data rate is used and the power reset to a minimum, and so on. This adaptation also works if the neighbour list is too large. Two connectivity types address the routing. Local connectivity is available up to two hops away, with all path loss and noise information available to the MS, and a link budget analysis is made to minimise transmission power. End to end connectivity is used for more than two hops, using an origin and destination ID. There is a ‘time to die’ criterion after which the packet is deleted, so delay time is considered in the routing algorithm for this mode. There is also an interesting comment in the standard concerning delay and number of hops. Considering an increase in hops, one would assume that the delay would increase; however, it is pointed out that the lower power allows a higher data rate, which will balance the hop-based delay. The basic intention of all previous routing algorithms is to minimise the mean transmitted power along the route, although this is often implemented by minimising path loss. The main difficulty is to make the correct decision whilst achieving a minimum of network overhead. UTRA TDD uses only short orthogonal spreading codes, the longest being length 16, corresponding to 16 kbps (3GPP, TSG, RAN, 1999a). From the limit of the pole capacity (Veeravalli and Sendonaris, 1999) as shown in Equation (6.12), the small processing gain limits the number of users that may be routed through a node, and ultimately to the BS. There is no explicit routing algorithm in (3GPP, TSG, RAN, 1998). The routing strategies discussed previously in combination with the ODMA principles indicates the following requirements: (i) optimise node loading according to pole capacity;
162
Next Generation Mobile Access Technologies
(ii) minimise overall transmission power; (iii) minimise interference at nodes. Although TDD allows diversity in time, optimising this allocation is beyond the scope of this chapter. The initial step for routing is to assess the path loss to and interference at other MS. This is achieved by probing neighbours. The description of probing is probably the most complete part of the standard (3GPP, TSG, RAN, 1998), and consists of several modes designed to get an initial neighbour list which is then kept updated with a minimum of interference. UTRA-TDD has low mobility due to the open-loop power control’s time constant being governed by the slot length. Its advantages include terminal simplicity, and asymmetric uplink and downlink bandwidth through time slot allocation. This indicates possible uses for LAN or other data transfer such as mobile IP.
6.2.2 Gathering ODMA network parameters In order to implement routing protocols it is necessary to build an information database that contains details of nodes available for routing, and the parameters required to calculate the routing metric (path loss from calling MS to BS, interference, etc.). The scope of the information requirement is dependent upon the nature of the routing protocol, and this overhead is described in more detail in section 7.6. The option exists to perform the routing either at a central system such as a BS, or in a distributed or local fashion where each MS holds a list and performs the metric calculation itself. Centralised network lists With a centralised list system all the routing is determined on a local cellular group basis with information gathered at the BSs using a dedicated signalling slot. Much of the information still needs to be gathered by the mobiles, such as MS↔MS path loss. This list inherently includes all the users in the local cellular group, i.e. the central cell and bordering neighbour cells. The routing information is processed at a central processor and then sent to the appropriate MSs though another dedicated time slot. Advantages of centrally processed network lists include the ability to assess the entire system, potentially resulting in the optimal routing for a particular algorithm. This is because all the required parameters are held simultaneously. The algorithm will not be required to make any assumptions of redundancy or approximation. In addition, algorithms such as those
UTRA-TDD Opportunity-Driven Multiple Access (ODMA)
163
used by (Zander, 1992; Foschini and Miljanic, 1993; Hanly and Tse, 1999) may be used to simultaneously solve a system of local requirements in order to obtain a globally minimal solution through linear algebraic techniques (Strang, 1976). The central solution allows for the possibility of integrating admission control and routing in a way analogous to that used by several proposed power control systems (Yates and Huang, 1995; Hanly, 1995). The motivation for this approach is that, as opposed to the idealised scenario of a single receiver for the uplink (considered to be the limiting factor on capacity by many investigations (Gilhousen et al., 1991; Viterbi and Viterbi, 1993; Veeravalli and Sendonaris, 1999)), the maximum capacity for a scenario involving multiple receivers is dependent on the interaction of parameters such as transmitted power and the path gain matrix, even when every user is perfectly power controlled to its receiver (Gilhousen et al., 1991; Veeravalli and Sendonaris, 1999). Using this technique, allocation of a resource (such as a time slot) may be made in order to provide the maximum capacity at a particular instance instead of the more usual first-come-first-served approach (Goodman et al., 1989; Foschini and Miljanic, 1995). As the information needs to be centrally collected, a likely mechanism is through the messaging slot/channel such as for the proposed UMTS system (ETSI, 1998) where coverage allows (Veeravalli and Sendonaris, 1999). This has the benefit of allowing for system wide synchronisation thus improving overall throughput. Disadvantages of such ‘all knowing’ include a delay from information gathering to informing the MS though messaging propagation. This occurs through one or both of two mechanisms. First, after the parameters have been obtained, the resulting connectivity information needs to be imparted to the relevant MS. Second, if the MS and BS lie beyond the others non-relaying range, the system may experience relaying delay, though this may depend on the system architecture (see section 7.6). Either of these factors will introduce errors due to differences between data used in the computation and the current parameters. The result being a degradation in system performance especially in a fast fading or highly mobile environment. Indeed this situation is not the worst case. It is possible that mobiles wishing to communicate solely on a LAN basis, but out of range of a BS, or those out of range with a single hop but in range with two or more hops, may be unable to establish any routing information whatever.
164
Next Generation Mobile Access Technologies Begin
Send probe signal at lowest power and highest bitrate
List metrics to neighbors and request their lists
Go to lowest power and decrease bitrate no yes
Minimum list size achieved?
no
Maximum TX power reached
Increase TX power Raise own list minimum by 1
yes
Request neighbours increase list by 1
Metric for target in ≥2 lists
yes Select route by minimizing overall metric
no
Target appears in list
no
no
yes
Neighbour list size ≥ own list size
yes
Take direct route
End
Fig. 6.2. Flow diagram for local list routing
Local network lists A local list system is based on the idea of distributed intelligence, such that the MSs are self-organising, utilising the BS purely as a sink for information that needs to travel beyond the scope of the local network. The list is
UTRA-TDD Opportunity-Driven Multiple Access (ODMA)
165
established by probe signals in a dedicated time slot, as shown in Figure 6.2 (3GPP, TSG, RAN, 2000a), and routing formed as described below. In this investigation an initial minimum list size of five per MS is used and, as suggested in the initial protocols for ODMA, can be increased in certain circumstances (3GPP, TSG, RAN, 2000a). Scenarios where this minimum list size is exceeded may be caused by more users coming within the current quanta of the path loss threshold, after the minimum list size was not met for the previous probe level, or where a practicable target is not contained within the lists of the members of a list with a minimum size. The implications of locally organised networks mean that this type of protocol may fulfill different requirements of, and hence of applications, in a centrally organised system. One positive aspect of local organisation is the ability to establish networks in areas with poor or no BS coverage. This was the original and purest meaning and motivation for ‘ad-hoc’ networks (Perkins, 2001). As it is not necessary for a central organising node to facilitate network operation, a BS need be neither within range or even communicated with. Initial applications were mainly military (Jubin and Tornow, 1987) though it is rapidly being found to benefit requirements as diverse as wireless LANs, ‘smart’ homes (Wallich, 2002) and self-organising sensor arrays. For MSs with restricted power resources or systems where bandwidth is precious, a reduction will be seen in peak transmitted power and hence in potentially damaging interference. This comes from restricting messaging overheads to communication only with the closest MSs and thus requiring the lowest power transmissions. In scenarios where a limitation of routing information latency is desirable, due to rapid changes in the path matrix, or where a high degree of sensitivity is evident, e.g low processing gain, local routing may be advantageous. This is because the lag can be made to be dependent only on the time difference between Tx and Rx slots in a TDD system, i.e on slot length and allocation of transmit slots between users. This is possible because path loss and other channel parameters are available from the reciprocal channel, as shown in section 2.3.1, which may be further simplified with a header containing information such as initial transmit power. A locally routed system may be less desirable where the sub-optimal routing for a particular algorithm due to an incomplete path-gain matrix may make substantial impacts on the power requirements or available capacity. This results from a minimised database of the local lists, which may result in a different routing than if the algorithm was able to use all users. Such an instance is where a high processing gain is used with the system operating close to theoretical capacity. As there is no shared slot containing metric
166
Next Generation Mobile Access Technologies
data - and indeed one network of users may be completely unconnected to another save though the impact of interference - imperfect synchronisation is a potential result within the network and a near certainty between cells and between networks. Imperfect synchronisation between cells results in a required increase in guard bands between time slots, and between both cells and networks in a likely reduction in the potential benefits available from multi-user detection. Any system with distributed intelligence requires a like dissemination of processing power to the MS, which requires increased MS complexity. Whilst the requirement per MS will be less than that for the central processing model, the MS simply being required to calculate its own route rather than everyone else’s, the reduction cannot be expected to be linear with number of users. This is due to calculations being replicated by different MSs, e.g. the reciprocal path gain values and routes. An implication of this requirement for increased processing by each user will be a like increase in power usage - indeed, routing algorithms have been proposed that include processing as part of a routing metric (Michail and Ephremides, 2000; Toh, 2001). Probing methods The ODMA protocols included in early specifications of UTRA-TDD mainly concern the probing cycle and use of the ODMA Random Access Channel (ORACH) (3GPP, TSG, RAN, 2000a), see Figure 6.3. There are three levels of activity: full probing, where the relay constantly monitors and transmits probes on the ORACH; duty maintained probing, the ‘normal’ mode, allowing flexibility in scheduling; and relay prohibited, where all probing is ceased and normal TDD or FDD operation is resumed. The level of probing activity is governed by parameters such as number of neighbours, gradient to base of neighbours, terminal speed and battery level. All routing strategy information is still missing, apart from a note that all MS should have at least one gradient to a node B, where a gradient is a cost function in terms of propagation conditions, number of hops, and other parameters. 6.2.3 Other work in ODMA Vodafone have produced the majority of the industrial simulation results (Vodafone Ltd., 1997a) (Vodafone Ltd. and Salbu (pty) Ltd., 1997a) (Vodafone Ltd. and Salbu (pty) Ltd., 1997b), with work from Bristol University (Harold and Nix, 2000b) (Harold and Nix, 2000a), and from the National University of Taiwan, in the use of ODMA in military applications. For the first simulations(Vodafone Ltd. and Salbu (pty) Ltd., 1997a) Vodafone
UTRA-TDD Opportunity-Driven Multiple Access (ODMA)
167
Fig. 6.3. ORACH Superframe Selection
used three propagation models: inverse square law, inverse fourth power law, and a Manhattan model. These all use seeds to relay the signals instead of relaying by MS. The capacity results given were calculated by increasing the number of calls until the ‘resource’ usage was 100% at a point in the cell. These results however are simply the number of supported calls, and do not consider any particular access method. Later results(Vodafone Ltd., 1997a) consider indoor office, outdoor/indoor/pedestrian and vehicular models. These models are more advanced, and are based upon the path loss models in the UMTS selection procedures(ETSI 30.03, V3.2.0 (1998-04), 1998), incorporating COST 231 path loss and a distance dependent expo-
168
Next Generation Mobile Access Technologies
nential autocorrelation function for the log-normal shadowing. They use MSs as relays, and only transmission power is considered, with no capacity calculation. All these simulation results show that the reduced overall path loss results in a lower overall transmission power, although this is shared by MSs not directly involved in the call. This will result in a reduction in interference, and a reduced average battery consumption. The papers discuss the situation with respect to shadowing. If there is high shadowing on the line-of-sight path, transmission power will need to be increased to achieve the required Eb /N0 . If the maximum transmit power is reached before this condition is attained a dead spot occurs where no communication is possible. Using ODMA it will be possible to go around the area of high shadowing resulting in a further gain on top of the shadow-free example. It is worth noting that due to the reduced transmission power and the subsequent lower interference, a system can be devised such that the radio resources, be they time slots, frequency bands, etc., can be reused. This effectively creates many pico-cells within the main cell and should serve to increase capacity.
6.3 Path loss investigation Let us now consider a simple simulation platform and routing algorithm to investigate the interference properties of a relaying system within a direct spread spectrum context. The effect upon path loss is examined by comparing a simple two-hop relaying system that attempts to minimise the path loss with a conventional system. The resulting interference pattern throughout the cell is used to show bottlenecks in the system, and hence the limiting factors for potential relaying gains, suggesting the new network topologies proposed in section 6.3.1. The interference pattern is analysed for different values of shadowing variance for correlated and uncorrelated shadowing variables. This indicates how consistent relaying will perform when moving from environments such as rural to indoor, and whether the systems become feasible or not.
6.3.1 Simulation model The COST 231 model was used for the indoor office test environment: L = 37 + 30 log10 d + 18.3n((n+2)/(n+1)−0.46) + ξ [dB],
(6.1)
UTRA-TDD Opportunity-Driven Multiple Access (ODMA)
169
where d is the transmitter-receiver distance in metres and n is the number of floors in the path (the recommendation is to set this to 3). ξ is lognormal shadowing in dB, with a standard deviation of σ and a zero mean; no correlation is applied. The minimum loss is never allowed to be less than that for free space. The MSs were placed randomly, their polar co-ordinates normalised to give a uniform distribution by taking the inverse of the required pdf (Proakis, 1995), in this case 1/r2 . The routing used in the initial simulations is a simple form of ODMA, corresponding to local connectivity only. There is an allowed maximum of two hops, and the metric is calculated using the minimum overall path loss of the transmission paths, an example is shown in Figure 6.4. The required transmission power is calculated as shown in equation (6.2). T xP ower [dB] = sensitivity [dBm] − gain [dB] + L [dB] + Imargin [dB] (6.2) Where gain is the total overall antenna gain and transmitter losses, and Imargin is a value based upon the number of users to try and achieve the required Eb /N0 . This is derived from the perfect power control requirements for a direct transmission signal, i.e. there is no power control system, in fact this means that the ODMA transmissions use a margin that is higher than necessary, so these results are actually worse than could be expected from a real system. No upper limit is applied to the transmission power. The interference power is based on a similar concept: I = T xP ower + gain − L [dBm].
(6.3)
This is used to plot an interference surface for the uplink, with the grid uniformly distributed in the same way as the MS. One problem with using this grid, as opposed to simply measuring the received power at nodes is that the power is averaged over the space in the grid, so the values around the base station are higher than the actual received signal, and may not be completely indicative of the interference at the receiver. 6.3.2 Characterisation of interference The interference pattern for the simulation was averaged from the surface plots as shown in Figure 6.5 over equal radius and several hundred runs to produce an interference cross-section for different values of σ in the lognormal shadowing. The results are shown in Figure 6.6 for a 20 m cell size, and in Figure 6.7 for a 100 m cell size. It can clearly be seen that,
170
Next Generation Mobile Access Technologies
90
20 60
120 15 10
150
30
5
180
0
330
210
240
300 270
Fig. 6.4. Example of mobile placement and routing
as expected, the increase in σ causes the conventional system to require increased transmission power, hence increased interference. The ODMA system, however, is not only resilient to the increase in shadowing, but it actually benefits from the increased shadowing deviation. Although it is not completely clear why this occurs, it is hypothesised that this is due to the distribution being zero-mean, resulting in lower path loss. There may also be shielding of some MS by the shadowing, resulting from the increase average difference of Tx to Rx transmissions and Tx to other MS path losses. This is caused by choosing the lower path for the former and the latter being zero-mean. The general resilience seems to be due to the choice in signal path, hence adaptation, allowed by ODMA. It is important to note that if more than one cell was considered the conventional system would continue to have an increase in interference power with distance, instead of tailing off as shown for the single-cell case. It is not
UTRA-TDD Opportunity-Driven Multiple Access (ODMA)
171
Fig. 6.5. Example of the interference surface experienced by MS
possible to make capacity calculations purely from the above information as the capacity of an ODMA system is not dependent just on the interference at the receiver, but is limited by the worst link in the system. It may be possible to overcome this by sending data over multiple routes, but even then there is the problem that if the weakest link is the last one before the BS, multiple routes may all suffer equally. As can be seen in Figures 6.6 and 6.7 an effect of ODMA is that the interference slope is almost the opposite of the conventional system, increasing towards the centre of the cell, due to the increased load on the central MSs. This will accentuate the bottleneck on the weak link, as almost all traffic will need to pass through these nodes. This would suggest a possible target for a routing algorithm, to produce a flat interference pattern from centre to edge of the cell. With the bottleneck in mind it would seem prudent to investigate BS diversity techniques such as antenna diversity, and analyse the effects of higher data rates on the central MS, once they become power limited.
172
Next Generation Mobile Access Technologies
Fig. 6.6. Interference with distance for a 20m cell with 10 users.
6.3.3 Correlated shadowing The initial simulation uses uncorrelated shadowing, simply assuming a lognormal distributed random variable can represent the path-loss fluctuations due to shadowing. For the case with no available path diversity in the transmission, such as a conventional non-relaying CDMA system, this is a reasonable assumptions In an ODMA system where there are several choices of path available for routing however, ignoring correlation in the shadowing may cause potentially optimistic results. This is due to the low shadowing paths being available in all areas of the cell, e.g. two users in close proximity may be given completely different shadowing values. In this case the high shadow path will not cause any difficulty, as routing can be made through the other user. It is more likely that there will be a correlation between the two users as the reason for the shadowing, building, wall, etc. will probably affect the path of both users. Klingerbrunn & Mogensen (Klingenbrunn and Mogensen, 1999) proposed a method for modelling cross-correlated shadowing
UTRA-TDD Opportunity-Driven Multiple Access (ODMA)
173
Fig. 6.7. Interference with distance for a 100m cell with 10 users.
with regard to the base station. Previous models have assigned correlation coefficients according only to the angle of arrival. The method in (Klingenbrunn and Mogensen, 1999) includes provision for correlation using both angle of arrival and distance correlation, the idea being that the correlation is greater if more of the propagation path is common between the users. The transformation between uncorrelated X, and correlated Y shadowing matrices is provided by a weight matrix, C, of the form; Y = CX.
(6.4)
C is derived by performing Cholesky factorisation on a correlation matrix, Γ, where ⎤ ⎡ 1 ρ12 . . . ρ1N ⎢ ρ21 1 . . . ρ2N ⎥ ⎥ (6.5) Γ=⎢ ⎣ . . . . . . . . . . . . . . .⎦ , 1 ρ N 1 ρN 2 . . .
174
Next Generation Mobile Access Technologies
which maintains the variance of the data when the correlation is performed. The correlation coefficients, ρij , are generated according to the correlation model, of which two are presented for distance and angle of arrival correlation. This model is adapted to ODMA by extending the model from only considering the shadowing on paths to the BS, to applying the correlation to all path losses between transceivers. The operation is performed n + 1 times where n is the number of mobiles in the cell. Starting with the base station as receiver, the correlation matrix is calculated with the desired receiver being omitted from the shadowing and weight matrix, then the first mobile station as receiver and so on until the correlation for all transmissions to all users have been calculated. Positive Definite Matrices In order to perform the Cholesky factorisation it is necessary for Γ to be a positive definite matrix. With large numbers of elements in the matrix, highly related variables such as the result grid and the base station, and rounding errors in computation, it often occurs that the matrix is not positive definite. Positive definiteness is satisfied if all of a matrix’s eigenvalues are positive (Wothke, 1993). This may be qualified to some extent by the determinant of the matrix. For symmetric matrices, such as our correlation matrix, if the matrix and every principle sub-matrix (formed by removing row and column pairs) does not have a positive determinant then the matrix is not positive definite. Rounding errors that create these problems need to be modified. The solution is to add a quantity to the diagonal of the matrix, until it becomes positive definite, or multiply the off-diagonal by a factor as close as close as possible to 1 until the eigenvalue condition is satisfied (Yung and Bentler, 1994). This works by attenuating the estimated relations between the variables. It means, however, that it is not always possible to indicate the degree of correlation in the system, e.g. a value of less than 1 is often required on non diagonal positions in order to satisfy positive definiteness in the matrix. It is therefore desirable to perturb the original matrix as little as possible. The sensitivity of a matrix to rounding errors can be due to the decomposition technique, e.g. pivoting around small numbers as they suffer higher percentage error for a fixed error, or some property of the matrix. Assuming that most decomposition techniques include a sufficient degree of intelligence to choose the optimal starting point, the matrix sensitivity will dominate. Hence it is useful to find a measure of this sensitivity (Strang, 1976). If we introduce an error vector δb into the original positive definite matrix A, this
UTRA-TDD Opportunity-Driven Multiple Access (ODMA)
175
will be amplified in a resulting eigenvector x by a factor 1/λ1 , the largest eigenvalue of A−1 . In order to make this amplification factor invariant to any matrix scaling, it is necessary to normalise by instead using eigenvalue bounds: λn δb δx ≤ . (6.6) x λ1 b The number c = λn /λ1 = λmax /λmin is called the condition number of A. Hence, the larger the condition number, the greater the scaling of the error. A matrix with a large condition number is termed ill-conditioned. The norm of A is defined by: A = max x=0
Ax , x
(6.7)
which bounds the amplifying power of the matrix (Strang, 1976)
6.3.4 Results Figures 6.8 and 6.9 show a comparison of ODMA using correlated and uncorrelated shadowing models for 20 and 100 m cells respectively. It can be seen that in general for the 5 dB shadowing case, the correlated model produces slightly higher interference statistics than the equivalent uncorrelated system. For the 10 dB case however, the correlated model gives lower values of interference. It would seem that the correlated model does influence the interference pattern, but in all cases the performance is not significantly decreased below the level of the 0 dB shadowing case.
6.4 Capacity Coverage Analysis A relaying system is sufficiently different from a conventional CDMA system for it to be unreasonable to assume that, for maximum transmission power, the coverage of a cell is inversely proportional to the number of users. This is due to the availability of many paths to the users, i.e. there are many possible receivers, effectively creating hand-offs within the cell. The number of users transmitting to a particular receiver is therefore not fixed. Analysis of the coverage capacity trade-off for a relaying system may allow for a coverage based admission control. As shown in section 6.3 the bottleneck in terms of interference for an ODMA relay link will be the final MS↔BS hop, so it may be advantageous to allow cell breathing, or dynamic cell geometry through BS assignment. This would provide for a more even
176
Next Generation Mobile Access Technologies
loading of BSs, optimising system requirements. Knowledge of the coverage
Fig. 6.8. Correlated and uncorrelated shadowing for ODMA: 20 m cell
capacity is also essential for effective cellular planning.
6.4.1 Simulation model Time slots need to be allocated as part of the routing process. This is because a TDD system is not able to transmit and receive simultaneously. There are only two alternating slots used in this model, A and B. The slots are allocated using the following criterion: (i) The final hop must correspond to the receive slot of the target. (ii) The slot allocation must alternate (the relay cannot transmit and receive on the same slot). (iii) The uplink and downlink (as opposed the nomenclature for the slots) are simulated simultaneously. This removes the need for any argument as to which dominates the capacity. Indeed with a relaying
UTRA-TDD Opportunity-Driven Multiple Access (ODMA)
177
-91 -92 -93
P ower in dBm
-94 -95 -96 -97 5dB Correlat ed S hadow 10dB Correlated Shadow 0dB S hadow 5dB Unc orrelated S hadow 10dB Unc orrelated S hadow
-98 -99 -100 0
10
20
30
40
50
60
70
80
90
100
Dis tanc e from B S in m
Fig. 6.9. Correlated and uncorrelated shadowing for ODMA: 100 m cell
system the distinction is blurred as a large proportion of communication is MS↔MS.
Fig. 6.10. Example allocation of A and B timeslots for a three-hop route
An example allocation structure is shown in Figure 6.10. The routing is allocated according to the minimum path loss, and more than 1 call/route is provided by increasing the data rate with a lower processing gain.
178
Next Generation Mobile Access Technologies
Table 6.1. UTRA-ODMA-TDD simulation parameter. Parameter
Value
Maximum transmit power
10 dBm
Target Eb /I0 at MS
5 dBm
Target Eb /I0 at BS
2 dBm
Logn. standard deviation, σ
5 dB
Noise figure (receiver)
5 dB
Bit rate
16 kbps
Considering UTRA-TDD’s suitability for low mobility data use, the path loss model for the indoor office test environment, equation (6.1), is used (ETSI 30.03, V3.2.0 (1998-04), 1998). A correlated shadowing coefficient is used, even though the interference pattern is not dramatically affected due to the path diversity available in ODMA. However, non-correlated shadowing could produce over-optimistic results with low shadow paths available in all areas of the cell. The simulation is performed in a single cell with MS distributed in a random fashion with a uniform distribution. It is considered that joint detection is available to the BS but not to the MS due to complexity. This is modelled by different target signal to interference ratios at the respective targets.
6.4.2 Power control Power control is implemented as a simple-step increment if the desired signal to interference ratio is not achieved at the receiver. A link is in outage if the maximum transmit power is reached at any stage in a link. This is not the optimal method for power control/call admission as capacity may be increased by more selective pruning; however, the intention is to model a simple distributed algorithm. Monte Carlo analysis is applied to assess the capacity.
6.4.3 Capacity limitations As a user may relay to any other user, the ODMA cell can be considered as a network of pico-cells, the effective BS in each case being the relaying
UTRA-TDD Opportunity-Driven Multiple Access (ODMA)
179
MS. In a conventional system the bit-energy to interference ratio, ǫj , in the presence of Gaussian interference may be denoted as: ǫj =
pgPu j M
i:i=j
,
(6.8)
Pu i + I + N
where pg is the processing gain, Pu j is the received strength of the desired signal, M is the number of calls in the cell, I the other cell interference, N the thermal noise, and Pu j the unwanted own-cell interference. In the multi-hop scenario, where each relay is a receiver, interference comes from all the hops sharing the same time slot. Ignoring interference that may occur from hops in the same time slot for the same route, the interference at each receiver becomes: pgPu j , (6.9) ǫj = H M ni Pu i k + I + N i:i=j k=1
where Hn i is the number of hops for user i in TS n, and Pu i k is the power at node j from the route for user i, hop k. This applies to each relay and the receiving node in a link. Thus the link for each user is limited by the lowest value of ǫj on route. This means that interference throughout the cell affects the uplink, not just at the ‘real’ BS. In the majority of cases, however, the most problematic link is the final hop to the BS. This is due to power warfare occurring when there is heavy cell loading. MS nodes do not suffer too severely from this issue as they will normally only relay a fraction of the total number of users, whereas everyone has to route to the BS, unless the target MS is in the same cell. By splitting the interference into intra-cell interference, IODM A , corresponding to interference at the BS due to in-cell MS↔MS transmissions, and adjacent cell interference, Iad , the received bit energy to noise for each user with perfect power control at the BS for the uplink can be reformulated from (6.9): ǫj =
pgPu , Pu (M − 1) + IODM A + Iad + N
(6.10)
where M is the number of users transmitting directly to the BS, and Pu is the received power at the BS. If pg is constant for all links, i.e. calls are not aggregated together as two or more calls are at a higher data rate and all
180
Next Generation Mobile Access Technologies
users are transmitting at the same rate, it can be seen that M is an upper limit on links/time slots for calls involving an uplink to the BS. Rearranging with respect to the number of users, M gives:
M=
pg IODM A + Iad + N − + 1, ǫj Pu
(6.11)
which can be compared with the pole capacity for a CDMA system from (Gilhousen et al., 1991)
Mmax =
pg +1 ǫj
(6.12)
It can be seen from equation (6.12) that the only variables we have control over that limit the number of users that can route to the BS are IODMA and Pu . This means that the limiting interference at the relaying nodes is linearly related to the transmission of the M users. The effects of adjacent-cell interference and receiver noise can be mitigated against, but the reduction in capacity due to IODM A cannot be minimised by increasing Pu , as this will result in a corresponding increase in IODM A . There is, however, more scope for optimising Pu as the effective reduction in cell size for MS↔BS transmissions means that the average MS↔BS path loss is reduced, and hence the outage due to maximum transmit power being reached will be lower. IODM A is dependent in several factors, including shadowing, routing, power control and call admission. The routing in conjunction with shadowing will determine the relationship between Pu and IODM A . The number of hops per link is a balance between minimising transmission power, as discussed in the next section, and the number of interferers, albeit at a lower transmit power. Power control in conjunction with call admission is a trade-off between the benefit of reduced transmission power, and increasing capacity by increasing power. It is important to note that the above equations only hold for single user detectors where the interference from other users is Gaussian distributed (Verdu, 1998). Multi-user detection and selective use of spreading codes invalidates the assumption of Gaussian distributed interference. Indeed, with a more sophisticated receiver, IODMA could be used as a form of antenna diversity with MSs not on the original route retransmitting the signal, and thus actually improving system capacity (see Figure 6.11).
UTRA-TDD Opportunity-Driven Multiple Access (ODMA)
181
Relaying MS
BS
Tranceiving MS
Fig. 6.11. Retransmission of signal over multiple paths to introduce antenna diversity.
Transmission Power One of the main cited benefits for ODMA is the reduction of transmission power for a link. This is achieved by splitting the transmission into a series of hops using other MS as relays. The transmission power is reduced due to the non-linear nature of path loss, as shown in Figure 6.12. The gain shown is the reduction in overall path loss over a single transmission. The path loss model is of the form (without shadowing): L = k1 + k2 log10 d [dB],
(6.13)
where k1 is a constant loss, k2 is the path loss exponent, and d is the transmission distance in m. The larger the path loss exponent, the greater the potential gains. Linear path loss would not show any gains for relaying. The gains are the maximum available for the non-shadowing scenario, with MS spaced equidistantly along the line-of-sight path from the MS 100 m from the BS. It can be seen that gains of almost 30 dB can be achieved in the vehicular model, and 15 dB typical for the indoor model. Although the overall transmission power is reduced, this lower burden is shared between users not directly involved in the call. Although it might initially appear that this will be unpopular with these users, this potential issue can be removed by introducing criterion as to which MS are available for relaying. Indeed in all the simulations in this book uses the approach detailed below, such that the battery life of users on standby will remain unaffected. In order to achieve this transparent sharing of resources only mobiles
182
Next Generation Mobile Access Technologies
Fig. 6.12. Path loss reduction against number of relays for 100 m transmission.
that are transmitting in other time slots are allowed to be viable relays, for example in a UTRA-TDD context with each user transmitting one slot per frame on average. This would allow a routing pool 15 times the size of available calls per slot, and remove the need to run down the batteries of users not making calls. A potential drawback is that users near the BS may experience a higher power requirement than with a conventional system, but this should be contrasted with the overall reduced power/call requirement, and a more predictable and consistent power usage. In fact this arrangement means that battery consumption will have both a lower mean and deviation. Conventionally, low Tx power users (near BS) will see an increase in required Tx power; conversely users near the edge of a cell, normally experiencing the highest required Tx power, will see a reduced power requirement. This means that on average everyone should benefit.
6.4.4 Results and discussion Figure 6.13 shows the average number of supported calls for each time slot in the TDD frame. Until the cell size reaches 30 m the conventional TDD
UTRA-TDD Opportunity-Driven Multiple Access (ODMA)
183
system achieves greater capacity. This is due to the combination of IODMA and the simple receiver architecture, where all received signals must be at the same power for optimum capacity. At 30 m all systems are equivalent. After this point the non-relaying system quickly loses the ability to reliably support calls, which is due to many of the MSs lying in a region where the path loss is too great for the signal to reach the BS with the required signal to interference ratio above the receiver noise.
Fig. 6.13. Supported number of users per time slot against the coverage of the cell for less than 5% outage. n is the number of users available for routing.
Reducing the target number of calls allows for a small increase in the coverage, but ultimately the MSs cannot increase their power sufficiently. The ODMA scenario where the number of calls is initially the same as the number of users available for routing offers a gain in capacity of two users at 40 m and will support five users when the conventional system cannot cover that radius. When the number of users available for routing is double the initial limit for call admission the capacity is conserved until 40 m and almost full capacity is offered when the conventional system has failed. After this point a reliable system is available for another 30 m - albeit
184
Next Generation Mobile Access Technologies
with a reduced number of users - three times the coverage in terms of area than a non-relaying system. Figure 6.14 shows the probability of outage for different numbers of allowed calls against the coverage of the cell. For the conventional system, once outage starts to occur the gradient is such that the number of dropped calls is too great for service to be maintained, which corresponds to users outside a particular radius getting no service. As the number of allowed calls for ODMA is decreased, not only does the coverage increase for a particular probability of outage, the gradient of the curve decreases. For the four user case there is a 4% probability of outage in a 70 m cell. Coverage is still available, however, at 90 m with a 16% probability of outage, much less than the 40% outage for the greater area in the conventional scenario, i.e. in the conventional scenario all MSs between 70 and 90 m would be in outage.
6.5 Conclusions Through the use of ODMA relaying to minimise the mean power transmitted, it has been shown that for a simple model, using only a single cell and local connectivity, the relaying system shows a reduced level of interference in the cell compared to a conventional CDMA system. When shadowing is taken into consideration, the conventional system shows an increased level of interference for higher variances in shadowing. An ODMA system may actually exploit the lower path losses made available by zero-mean log-normal shadowing, reducing interference in some higher shadowing scenarios. In order to ensure that the model itself is not responsible for the gain in high shadowing environments, a correlated shadowing model was extended to an ODMA situation. This showed that correlation between shadowing variables for distance and angle of arrival to determine shared paths shows different interference characteristics, but the benefit of path diversity is not significantly diminished in the worst case, and may actually be improved for high shadowing. Through the use of multi-hop transmissions, ODMA will extend the coverage of a cell, but only after the cell size becomes large enough to prevent a comparable single-hop system from achieving its maximum capacity. When coverage is a more important criterion than capacity ODMA will provide a reliable service far beyond the coverage of a conventional TDD system. As the number of users available for relaying increases, the coverage of the ODMA cell increases, and the gradient for coverage capacity decreases. Re-
UTRA-TDD Opportunity-Driven Multiple Access (ODMA) 0.4
P out - P robability of outage
0.35
185
UTRA -TDD, T= 10 O DM A , T= 10 O DM A , T= 7 O DM A , T= 5 O DM A , T= 4
0.3
0.25
0.2
0.15
0.1
0.05 10
20
30
40
50
60
70
80
90
R - Cell s iz e (m )
Fig. 6.14. Probability of outage against the coverage of the cell, for different numbers of allowed calls, T , with 20 users available for routing.
ducing the number of allowed calls means that ODMA can provide a reliable service for an area greater than the normal capacity coverage trade-off. By allowing a reduced quality of service the operator can further extend coverage, the outage being shared between all users, not just those outside the transmission radius limit of the non-relaying system. For unevenly loaded adjoining cells, ODMA could be used to even out the loading by dynamically adjusting the cell size. The use of ODMA could provide operators with a far greater degree of flexibility in cell planning, and to go beyond the conventional trade-off between coverage and capacity. This comes at the cost of increased complexity and signalling overheads, and relies upon a sufficient user density to maintain available relays.
7 Routing strategies in multi-hop CDMA networks Tom Rouse Stephen McLaughlin
Multi-hop relaying routing protocols have been investigated for CDMA air interfaces in conventional cellular scenarios, as described in Chapter 6 and in (Harold and Nix, 2000a)(Harold and Nix, 2000b). This chapter compares the performance of ODMA with direct transmission for cases where links maybe required directly to other nodes, as well as to a controlling (backbone) node, and presents two new routing algorithms. For an interference-limited system, it is shown that the topology cannot be supported by a conventional (single-hop) system, but that a relayed system is able to provide service. As an enhancement to path loss routing, a new admission control and routing algorithm based on receiver interference is presented which is shown to further enhance performance. A second new routing algorithm, which considers the interaction between all receivers in the system by means of a ‘congestion’ measure is presented. This approach allows for routing that is optimised for the entire system, not just a particular route under arbitrary starting conditions. This is possible under both central and local parameter gathering scenarios. Through formulating this measure into the power control equations it is possible to determine system feasibility, although this is a conservative criterion due to approximations in the formulation. This congestion-based routing is shown to outperform non-relaying and any previous routing technique in available capacity for the new network topologies, and has the lowest transmitted power requirement of all investigated methods.
7.1 Introduction Previous ODMA systems have utilised the path loss between terminals as the metric to determine the routing for the relayed packets (3GPP, TSG, RAN, 2000a)(Harold and Nix, 2000b)(Harold and Nix, 2000a). This is suit186
Routing strategies in multi-hop CDMA networks
187
able for a single-user system, or one that is not interference-limited such as frequency or time multiplexing where no simultaneous resource sharing is required. For an interference-limited system with multiple users, route selection purely from path loss does not take into account the degradation in performance that will be suffered by other users, either in the form of outage or reduced data rate. The degradation in performance is caused by similar factors discussed previously, e.g. in networks not exclusively of a star topology (Section 6.4.3). An ODMA system is effectively the same, even when there is no peer-to-peer requirement, due to the relaying nodes. Unlike the non-relaying system, however, we have a selection of paths available to us, so it is possible to minimise the interference effects by careful selection of the route. This section investigates several systems for reducing the detrimental effects of interference, with various trade-offs between complexity, transmitted power, latency, signalling overheads and capacity, and the differences between a locally or centrally organised system.
7.2 Multi-hop network architectures In Chapter 6 it was shown that there may be capacity gains due to relaying when non-relaying systems start to lose system capacity through an extended coverage requirement. Due to the relaying interference IODM A in Equation (6.11), however, the supported number of users for a conventional cellular network within the non-relaying coverage area is reduced if a relaying system is employed. This section investigates new network topologies that attempt to utilise the features of a relaying system to allow increased capacity over a non-relaying system, especially where there is a high degree of peer-peer communication, e.g. as shown in Figure 7.1.
7.2.1 Topologies The motivation for introducing peer-to-peer communications is two-fold. First the pole capacity of a receiver, (6.12), limits the number of users that may be received by a single receiver. If all calls go via the BS then this limits the number of calls in a cell. Secondly the interference patterns shown in section 6.3.2 indicate that the highest levels of interference are towards the centre of the cell. It is therefore considered that it would be advantageous to reduce the number of calls via the BS if possible. The network topologies presented here allow for peer-t0-peer communication if required to try and achieve this, and hence utilise the lower interference and spare capacity experienced by the MS towards the edge of a cell.
188
Next Generation Mobile Access Technologies
Fig. 7.1. Cellular communication utilising both BS and peer-to-peer transmissions, for relaying and non-relaying systems
In this section the allocation of a target from a user attempting a call is considered in two ways, as illustrated in Figure 7.2†. The first is in a cellular or BS centred fashion, where if the desired recipient is within the cell allocated to a user by handover procedures then the user will attempt a direct link without the use of the BS, otherwise the transmission is relayed via the BS to the cell containing the end receiver. A similar topology has been independently presented by Lin and Hsu (Lin and Hsu, 2000), and routing protocols investigated (Ananthapadmanabha et al., 2001). The motivation for this architecture is not motivated by interference, as no air interface is considered, hence the routing protocols are not relevant here, and solely concern route discovery and throughput. The advantage of this BS-centred availability region is that no extra signalling is required to establish the end-to-end terminals. Unfortunately the transmission may be from one side of a cell to the other, which causes infeasible interference for the critical BS reception. The second approach is † Figures 7.2, 7.3, 7.10, 7.11, 7.12, 7.13, 7.14, 7.15, and 7.16 and Table 7.6 are reproduced with permission from: (Rouse et al., 2005).
Routing strategies in multi-hop CDMA networks
189
MS centred region
BS centred region
c Fig. 7.2. Regions governing local target selection 2005 IEEE
to centre the selection region on the MS and then use the same criterion detailed above, this reduces the BS interference problem but requires increased signalling. This MS availability region is merely a revision based upon the findings of section 6.3.2, to attempt to reduce the BS interference, and to optimise transmission distance. This approach is still ignorant of other system parameters, analogous to the path loss routing which will be presented in section 7.3.1. It is presented in order to show that for an interference limited system at least, whilst it may be reasonable to centre a cell on the BS for calls via the BS, this is not a reasonable assumption for mixed peer-peer and BS communication. An optimal approach would be to decide upon BS or end recipient to maximise capacity or throughput. Perhaps this could be achieved though comparison of the congestion measure, as will be shown in section 7.5 for the two possibilities. However, this is beyond the scope of this book.
190
Next Generation Mobile Access Technologies
7.2.2 Impact of CDMA The conventional technique for networking in interference limited systems is a star topology in order to ease the problem of power-control. The topology used in this paper consists of sources and sinks, with communication targeted outside a cell and routed through a BS, while that within cell is targeted at the desired recipient. Whilst this minimises the number of links for maintaining a cellular paradigm, there is no longer a single point to which the power must be controlled. As there is more than one receiver it is not guaranteed that all the transmissions can be received at the same level by all receivers, and in some cases the interference generated by one user may mean that others close by may not be able to receive the desired signal, no matter how much the source increases power. In these cases it may be better to revert to a star topology, though it must be considered that two calls are now required where one sufficed previously.
7.3 Using path loss and interference-based metrics to route Within the scenarios users are required to communicate either with each other or to achieve a link onto a backbone. The criterion for the local routing is that the target is in the same cell as the transmitter. It is likely that greater capacity would be achieved if the region for local routing were centred instead on the user, however, a simple mechanism for call assignment was selected. Previous work on ODMA (Harold and Nix, 2000a)(Harold and Nix, 2000b)(Rouse and McLaughlin, 2001) uses path loss between terminals in the metric to determine the routing. This is a quick and efficient way to establish routing although it does not take account the bottlenecks in the system due to interference caused by many users routing through a particular area, and indeed it will encourage bottlenecks as users will all try to route via regions of low path loss. In this chapter three new algorithms are investigated, based on either minimising interference or a ‘congestion measure’ (Hanly, 1999), so that in combination with local and central lists seven network topologies may be investigated and compared.
7.3.1 Path Loss Routing The simplest form of routing is to attempt to minimise the path loss in the system. In non-interference limited systems this can also be considered optimal routing as far as power consumption is concerned, as long as no
Routing strategies in multi-hop CDMA networks
191
other power control requirements are imposed other than a minimum received signal power. For the CDMA system considered here, path loss is not an indication of transmitted power. The required Eb /I0 ratio is altered as shown in (6.10), the IODM A term being dependent on the interaction between all path losses for all transmissions in the system. Minimising the path loss for a route may well reduce required transmission power, and hence IODM A , but there is no guarantee that this will be the case. This is because the path loss routing may generate a situation where the required Eb /I0 means that a higher transmitted power is required than for the single hop routing due to the lack of power control to a central point. Initiating node Least distant node from initiator
Most distant node before target Target node 1 hop
2 hops
(n-1) hops
n hops
c Fig. 7.3. ODMA routing trellis2005 IEEE
Central routing calculation For in-cell calls, the user probes to find the path loss to other nodes, and their path loss to the target. The routing is determined by choosing the route with the minimum path loss overall. For out-of-cell calls a similar probing is performed, but the target may be any BS. This means that the user may be outside the handover region of the BS that it linked with. The basic routing structure is performed in a trellis as shown in Figure 7.3 to reduce the number of calculations required; with a path loss-based system this causes no degradation in performance as the minimum path loss route is selected at each node in the trellis. The trellis is formed by ordering the nodes with respect to the path loss from the initiating node. It is truncated
192
Next Generation Mobile Access Technologies
at the target distance plus a margin to include 90% of the users within the shadowing distribution of the system. The routing is performed at each stage by storing an array of the current distance and previous nodes visited at each node. A transition is available to any node at the next stage provided that node has not been previously visited, giving a number of paths from each node at stage i of number of nodes −i − 1, when i = 0 at the initiating node. The winning entrant to a node is decided by choosing the minimum distance. The routing for i hops is held from the path information of the winning entrant at the target node at stage i. The trellis is truncated at stage n if the distance at the target node stage n is greater than distance at the target node stage (n − 1) + (the shadowing margin used above), or if n is equal to a maximum allowed number of hops. The minimum distance at a target node is chosen, and the routing information taken from the path used to reach it. The routing is considered for two-hop and multi-hop cases. With the centrally assessed mode, the minimum path loss route in both these cases. Distributed routing calculation Local routing is performed using the trellis structure as described in section 7.3.1. The difference is the method of discovery and the subsequent reduced availability of path loss information. The basic methodology for gathering the path loss data is as described in section 6.2.2. The parameters are a minimum list size of five, with a maximum of two allowed hops. 7.3.2 Interference-based routing Routing using path loss is non-optimal due to multiple access interference. This section describes a routing technique that initially uses a path loss metric, but proceeds to account for interference once this information is available, after power control iterations have begun to converge. Routing metric Although path loss routing is a simple and instantaneous routing method, it suffers from assuming that the minimum path loss gives minimum transmission power. This is not the case in an interference-limited system in the multi-user case, unless other users’ interference is below the margin of Eb /I0 − (processinggain). The required transmission power for user i transmitting to user j is Pij =
Eb − pg + Γ + IM AI + IODM A + Iad + N, N0
(7.1)
Routing strategies in multi-hop CDMA networks
193
Begin
PC counter since re-route > n (=5)?
no
Re-order=TRUE Node_re-route=ALL
Re-order =TRUE?
no
yes
yes Node_re-route=UE_ctr inc UE_ctr
Calculate interference at each node
yes
Metric = path loss + interference above floor
yes
no
Interference > receiver floor?
Metric = path loss + receiver floor
All nodes calculated?
no
Route with modified metric, removing txed power from current links
End
Fig. 7.4. Interference-based routing flow
where pg is the processing gain, Γ is the path loss from user i to user j, IM AI is the interference from other users transmitting to user j, IODM A is the interference from other users in the cell, Iad is the adjacent cell interference, and N system noise. IM AI is dependent on the number of users transmitting to user j. IODM A depends on the path loss between other transmissions and user j, and their solution to (7.1), hence this is a highly interactive system. It can be seen that a lower transmission power will result by moving to
194
Next Generation Mobile Access Technologies
another relay, k, with a larger path loss, as long as Γik + IM AIk + IODM A k < Γij + IM AIj + IODM A j .
(7.2)
This re-routing according to interference will reduce the transmission power for an identical system, as this is a condition of the routing. Hence it is fair to also consider this technique as transmission power routing. It is important to note that this system will only find a local minimum from an initial condition, and for a single route within a highly related system. This technique reduces required transmission power, and hence outage from exceeding the maximum transmit power. Unfortunately the power control problem is not necessarily asymptotic for every local minimum (Hanly and Tse, 1999), so this is not an optimal solution. Initial routing is performed as in section 7.3.1 for central routing and section 7.3.1 for local routing. Once the power control has reached a steady state, the system is re-routed according to the modified metric, minimising overall P , (7.1), for each entire route. Any interference above the receiver floor plus processing gain at each receiver is added to the path loss matrix between users. This gives an indication of the necessary transmission power for that link, but this will just be an approximation as the re-routing changes IODM A and IM AI for other users in the system. The re-routing is performed on a route by route basis in order to prevent many users jumping to a low interference route at once and causing instability in the routing. The predictive outage calculation described in section 7.4 is suspended until all routes have had the chance to re-route, avoiding unnecessary outage. 7.4 Interference-based admission control With a star topology based CDMA system the number of allowed calls is relatively constant due to the single receiver in the uplink. In an ad-hoc environment, the capacity will vary depending upon the position of users and the links that are required. This means that admission cannot be based simply on the number of calls currently in progress. The technique used in this paper is to start with a desired number of calls and attempt to make all of them. Instead of waiting until the maximum permitted transmit power, in this case set at 10 dBm, is reached and the link involving the offending transmitter removed from the current calls, a prediction technique as shown in Figure 7.5 is used. The convergence rate of the power level is used to analyse which receiver suffers the worst interference and then a decision based on a metric to determine which call to terminate in order to make the greatest reduction in interference for that user.
Routing strategies in multi-hop CDMA networks
195
Begin
New power → running average
Perform one power control iteration
No
Increment prediction counter
Prediction counter>5 Yes Find average dP / dt and d 2 P / dt 2
0 ≥− Yes
Extrapolate
−
dP / dt d 2 P / dt 2
≥-100 No
dP / dt d 2 P / dt 2
Extrapolate 100 iterations ahead
iterations ahead
No
Prediction >max Tx power
Yes Remove user creating most interference
Fig. 7.5. Interference-based admission flow
The power level p(t) for each user is extrapolated by approximating the first and second derivatives by their respective time differences, taken from a running average for iteration i. The power level at some later iteration, j, (when the power control may reasonably be expected to have converged), can then be predicted using a Taylor expansion, as in (7.3) below. If the second derivative is of an opposite sign to the first, the iteration when the power is expected to have converged is predicted to be j = i − (dpi /dt)/(d2 pi /dt2 ). A modification of this is introduced if the predicted iteration j is greater than
196
Next Generation Mobile Access Technologies
(i + 100). The procedure is repeated for j = i + 100 instead of the predicted value. This capping of the iteration is also applied when dp/dt and d2 pi /dt2 are of the same sign pj = pi + (j − i)
j−i dpi d2 pi k. + 2 dt dt
(7.3)
k=1
If any of the predicted powers exceed the maximum transmit power the interference is analysed. All of the interference from interfering transmissions that is greater than the desired signals is summed, and the link with the greatest contribution removed. After any links are removed, or the routing is changed, the power control will need to take account of the new scenario. To prevent any users who will have acceptable power variables being unnecessarily put into outage, prediction is switched off until the averaging takes account of the new situation. This extrapolation technique has two advantages over a wait-and-see approach. First, convergence of the power control is reached earlier in a system that does not have a feasible solution below the maximum transmit power without one or more users being placed in outage. Second, the greatest interferers are removed, allowing increased capacity. Power control is performed on an iterative basis according to the target C/I ratio at the receiver at a rate of one iteration per time slot. As the simulation is static, the main requirement is a convergent solution. In order to achieve this the loop uses exponential convergence, which results in slow but stable and predictable performance. For macroscopic congestion-based ODMA, discussed in section 7.5, the system produces enough information to directly calculate the required power from the Perron-Frobenious eigenvector (Gantmacher, 1959), and a more advanced form of admission control using the power control feasibility requirement of λ < 1 may be implemented. In order to produce a comparison between the routing effectiveness, however, all systems share the same power and admission control. 7.5 Congestion-based routing The routing methods considered in section 7.3 use path loss for each route to initialise the routing. This technique is appropriate where there is little or no interaction between users, such as where a single user is allocated all of the available bandwidth at a particular instant. This is often the case for the packet radio systems where the relaying concept originated. For civilian CDMA systems this is rarely the case. In order to mitigate destructive interaction between users, this section considers a measure of this interaction,
Routing strategies in multi-hop CDMA networks
197
termed congestion, which can be assessed both locally and centrally with the same result. A simple algorithm is presented that attempts arrange the routing in order to minimise this interaction.
7.5.1 Congestion measures and routing A measure of congestion in a spread-spectrum system has been formulated by Hanly (Hanly, 1999) as a development on his work on cell-site selection (Hanly, 1995) and has also been independently proposed as a routing metric. The measure is in the form of the lower bounds on the PerronFrobenious eigenvalue (Gantmacher, 1959), λ, for a positive M × M matrix of the form A[i, u] =
αi Γ[u, ci ] , W Γ[i, ci ]
(7.4)
where αi is a QoS requirement, Γ[x, y] is the path loss between x and y, ci is the BS for user i and W is the bandwidth. In the above scenario the only operational potential to change A is to change ci through cell-site selection. In the multi-hop scenario the situation is more akin to sources and sinks than a fixed BS receiver, hence our problem formulation may be minimised in several dimensions. Generalising (7.4) to allow for all users in the cell to be available as receivers we obtain the following: ⎞⎛ ⎞ ⎛ αu Γ[T Xu , T Xj ] ⎟ α ⎟⎜ ⎜ Rxu =Rxj u A[j, k] = ⎝ ⎠⎝ ⎠ , (7.5) W αi Γ[T Xu , T Xk ] RX u =RX k
RX i =RX j
where T Xi and RXi are the transmitter and receiver for link i, respectively. The size of the matrix is governed by the number of links, not the number of users, although it can be shown that as the eigenvalues for all links transmitting to the same node are identical, λ is the same as if it were formulated for number of users (potential receivers) in the cell. Initially λ allows us to assess whether our current connectivity has a feasible power control solution (λ < 1), see section 8.2.3. The advantage of this formulation over the interference-based algorithm is that not only is interference assessed at the intended receiver, the interaction between all users is accounted for, hence optimisation of the system is possible. By reformulating A for potential routing candidates and using approximate eigenvalue methods, we perform routing by minimising λ.
198
Next Generation Mobile Access Technologies
7.5.2 Local congestion metric The formulation in (7.5) is a near-optimal local solution with a congestionbased metric. It is not optimal due to the approximation of counting selfinterference (in the i = u terms in (7.4)) and a possible restriction of the routing candidates to reduce the necessary computation. It is still quite computationally intensive, and suffers from the disadvantage that all the path losses in the system need to be known at the central database, a non-trivial task requiring a large signalling overhead and the disadvantages detailed previously for central lists. Fortunately a result also presented in (Hanly, (n) 1999) is that for a decentralised power control algorithm, where Qk is the interference at receiver k for iteration n, λ can be assessed locally due to the convergence of (n+1)
Qk
(n)
(n)
− Qk
(n−1)
Qk − Qk
→ λ as n → ∞,
(7.6)
as long as the path gains remain fixed between iterations. Hence we can assess the impact on congestion for different routes by utilising a channel probing technique, avoiding the need for collection of all path loss information and facilitating a decentralised congestion-based routing algorithm. Routing is performed by attempting to minimise λ by evaluating other routes. This takes place in the slot that would otherwise be used for signalling overheads. There is no reason not to transmit desired packets during this process so, if the target is reached during the optimisation, little capacity is sacrificed by the probing. The disadvantage of this technique is the necessity for static path loss values, increasing the error in approximation, and the instantaneous empirical nature of the algorithm, not allowing for system-wide optimisation. It is likely, however, that in a peer to peer situation users will be slow-moving or static, so this should not generally be a problem.
7.5.3 Central congestion metric Routing is initially performed as in section 7.3.1 to provide a reasonable starting point for optimising λ. This gives us a point of reference as to whether our re-routing in improving system performance. The trellis is then re-navigated for each user attempting a call, except that instead of summing the required metric distance at each stage, A is reformulated allowing λ to be recalculated. λ is then used as the distance to determine the winning entrant at each node. The same selection criterion is used at each stage to determine whether to proceed to the next, i.e. is λn−1 > λn at stage
Routing strategies in multi-hop CDMA networks
199
n, with the exception that no shadowing deviation allowance is made, the same approach is used to select the appropriate final routing. In conjunction with the cell-site selection procedures detailed in (Hanly, 1995) another level of optimisation is available on the network organisation level through cell breathing as detailed in section 7.2. This approach is an extension of hand-over regions to include a region centred on all users, not just the BS. This region determines whether peer to peer transmission takes place, and if not which BS handles the call. This may be integrated into a routing algorithm if the target is not fixed (i.e. it may be the destination MS or any BS), and the computation encompasses several cells.
7.6 Signalling overheads and latency All of the above routing algorithms require signalling that will serve to reduce the useful available resources of the system. This needs to be offset against any capacity gains that may result from the use of these algorithms. The amount of signalling required varies according to the routing algorithm used and information-gathering technique used. Without detailing a specific signalling protocol it is not possible to determine the exact overhead required; however, Table 7.6 shows the required information for each protocol and the proportional increase in required signalling bandwidth for each system. A common question regarding relaying systems concerns system delay. This is dependent on the relaying technique in question, some proposed systems (Boyer, 2001) simply use the relay as a signal booster with a resultant degradation in the C/I (Boyer et al., 2001) and hence system capacity; the delay will then be dependent purely on the circuit design. Another approach is to piggy-back data by reducing the processing gain, hence increasing the bit-rate. It has been argued that this will actually reduce system latency due to the increased bit-rate (Delta Concept Group ETSI SMG2, 1997); however, this does not account for decoding, encoding and delays through the RF hardware. The TDD system investigated in this book does not support concurrent transmission and reception, hence neither of these schemes are feasible. The system instead uses the next time slot for relaying, hence the system delay is the length of the time slot times the number of hops used. With the exception of the interference based ODMA and the local congestion ODMA, the protocols produce a final routing table purely from the initial information. This makes them suitable for highly mobile and/or bursty data. For the other systems, however, the protocol will converge towards the routing at the same rate as the power control convergence, and are thus
200
Next Generation Mobile Access Technologies
Table 7.1. Required information and proportionality of signalling load per c user 2005 IEEE Routing protocol
Required information
Signalling amount per user proportional to n users/cell
Direct
Path loss to target
Central ODMA
Path loss between all users
n2
Local ODMA
Path loss to neighbour list
invariant
Central interference ODMA
Path loss between all users, interference at all users
n2 + n
Local interference ODMA
Path loss to neighbour list, interference for neighbour list
invariant
Macroscopic congestion ODMA
Path loss between all users
Local congestion ODMA
Power control probe to test route
invariant
n2 invariant
unsuitable for high mobility and bursty traffic unless fast power control is in use.
7.7 Results While three scenarios were investigated; four square cells, seven hex cells, both shown in Figure 7.6, and the 3GPP indoor office shown in Figure 7.7, the plots shown are all for the square scenario. The hex scenario is the closest to a conventional cellular plan, and the 3GPP indoor office is applicable for indoor peer to peer communications with repeaters/backbone nodes placed in some rooms. The square cell is a good approximation of the hex scenario, but closer BS placement to the 3GPP indoor office, and is used in most results due to the reduced simulation complexity. Table 7.2 shows the affect on capacity of the different scenarios. The scenarios were modelled using a MATLAB based scenario generator (Rouse et al., 2002). The model randomly generates the users with a uniform distribution and uses the COST 231 model for path loss (COST231, 1999).
Routing strategies in multi-hop CDMA networks
201
Fig. 7.6. Square and hex scenarios
Table 7.2. Supported users per BS for various scenarios for 1:1 ratio of local to non-local traffic Number of supported users/BS Direct ODMA ODMAint Square, pg = 12.9 dB
4
5.5
8
Square, pg = 15.9 dB
5
7.5
11.5
Hex, pg = 12.9 dB
4.5
6.1
8.5
Hex, pg = 15.9 dB
6
9
13.5
3GPP, pg = 12.9 dB
1.5
3
4
3GPP, pg = 15.9 dB
2
5
6
7.7.1 Capacity The capacity of systems limited to two hops and multiple hops were examined separately. The two-hop systems are simpler to implement and should be more robust and predictable in fast-fading environments, especially for local lists, as the whole route can be determined simply from communication with the relaying node. Systems not limited to two hops have more possible routes available, and should be able to use lower power on average as shown in Figure 6.12. Apart from section 7.7.3 all the results presented are for the BS or cell-centred network allocation as shown in Figure 7.2. For the two-
202
Next Generation Mobile Access Technologies
25
20
15
10
5
0
-5
-10
-15
-20
-25 -50
-40
-30
-20
-10
0
10
20
30
40
50
Fig. 7.7. 3GPP indoor office scenario
hop strategies the congestion results presented are interchangeable for local and macroscopic methods, as the results are identical. For the multi-hop routing only the macroscopic congestion method was implemented. Two-hop Strategies Figure 7.8 shows the number of supported calls in all four cells for 5% outage against different ratios of local to non-local traffic. The star curve represents the number of supported calls that could be expected for a star topology with either the direct or interference-based ODMA. This is calculated for the number of supported calls for all non-local traffic and doubling the number of required calls for local traffic to account for the BS being required to forward the data onwards to the local target. It can be seen that for all cases, except interference-based ODMA, at higher ratios of local to non-local traffic the star topology delivers greater capacity. It is important to note that much of the outage is due to the simplistic selection of local targets from BS handover regions as is shown in section 7.7.3. For example, it is very likely that
Routing strategies in multi-hop CDMA networks
203
a target will be selected that requires the user to transmit in a path that crosses the BS, inherently creating very high interference at the BS and thus the user may well be put into outage as it is a problem interferer. A more intelligent selection of users available for local traffic would be achieved by letting the user make local decisions based on neighbour lists; however, this approach would not be possible for the direct approach without an extended signalling complexity. The approach investigated can be considered as a worst case for local routing. Intelligence in the selection of local traffic should reduce outage in all scenarios, meaning that peer-to-peer communication will only be established if it will enhance capacity, power levels, etc. 40
direct ODMA ODMAint star local localint congest
Number of supported calls for 5% outage
35
30
25
20
15
10
0
0.1
0.2
0.3
0.4 0.5 0.6 0.7 Proportion of traffic via backbone
0.8
0.9
1
Fig. 7.8. Number of supported calls for 5% outage for different proportions of local and non-local traffic, cell-centred network, two-hop routing strategies.
It can be seen that for the non-relaying scenario that the capacity is about half that which could be expected if a star topology was used, apart from when all traffic is via the backbone when the situation is identical. This is because the use of peer-to-peer communication introduces a power control problem that is likely to be insoluble. This is due to users transmitting to
204
Next Generation Mobile Access Technologies
other users on the other side of the cell, hence power controlling the peer instead of the BS. This will introduce high interference levels at the BS that cannot be compensated for with power control, as increasing the received power has a feedback effect on the required power to be received at the local target due to the increased interference at that user. The results for the direct transmission indicate that there is no reason to implement the presented local/non-local topology in a conventional CDMA system unless there is a severe problem with coverage. Although the path loss-based ODMA shows about a 50% improvement over the non-relaying system for all scenarios involving local traffic, it is still significantly worse than the conventional star topology. The gains arise because ODMA systems generally involve lower transmission power than conventional systems, hence interference problems are more localised so they will not have such a detrimental effect on other users due to feedback in the power control. Interference-based ODMA shows a gain where the local traffic is at least 50% of the total, performing best when all traffic is local. When non-local traffic is dominant it is roughly equivalent to the conventional system. Path loss-based routing does not take account of the congestion due to interference, mainly in the centre of the cell, when routing, the interference-based system will avoid these problem areas unless absolutely necessary. This means that transmission, even to the other side of the cell, is possible without destroying desired signals at the BS as the signal is routed to avoid relays in this area. On the downside any ODMA system requires increased complexity both in signalling to establish routing, and in the handset’s ability to relay the signal. It also takes longer to establish the routing as the dynamic routing requires the power control to converge after each iteration. Congestion based two-hop relaying shows a slightly reduced capacity from the interference based routing, but will be as up to date as the path loss information that is gathered, as no prediction is required. The local list based routing methods show a comparable, and sometimes lower, capacity to the simple ODMA. As can be seen from Table 7.2, the greatest number of users are supported by interference-based ODMA, in all scenarios. The most dramatic improvement is for the 3GPP scenario as this involves high losses in certain paths due to walls. A non-relaying system has no path diversity and is forced to use this transmission path. The relaying system has several choices available to it, and thus is likely to be able to avoid these areas of high signal attenuation. In this scenario the wall losses are the dominant problem so path loss and interference-based ODMA have similar performance.
Routing strategies in multi-hop CDMA networks
205
Figure 7.9 shows the number of supported calls against processing gain for the square cell scenario. The ratio of local to non-local traffic is 1:1. The systems all show a reduced performance over that which could be expected from increasing the processing gain; where the performance should roughly double, and the data rate halve for each 3 dB increase in the processing gain. This is due to the increased number of users creating greater likelihood of strong interactions between local transmissions thus creating power control difficulties. This can be avoided by intelligent local target selection. Both ODMA systems show similar increases in the supported number of users with increasing processing gain, which is not too far below conventional systems; however, the direct routing shows greatly reduced performance with just over double the number of users for 1/16 of the data rate for the lowest processing gain. 70 direc t O DM A O DM A int
Num ber of s upport ed c alls for 5% out age
60
50
40
30
20
10 12
14
16
18
20
22
24
26
P roc es s ing gain (dB )
Fig. 7.9. Number of supported calls for 5% outage for various processing gains.
Multi-hop Strategies Figure 7.10 shows the supported calls for various ratios of local to non-local traffic. In this case however, the routing strategies are allowed to utilise
206
Next Generation Mobile Access Technologies
more than two hops, with a maximum of six. No local list routing schemes are analysed, as it is considered that for more than two hops the reduced signalling advantage of this system will be compromised. The highest capacity two-hop system, interference based ODMA, and two-hop ODMA are shown for comparison. 40
ODMA ODMAint2 star 6hop 6hop int 6hop congest
Number of supported calls for 5% outage
35
30
25
20
15
0
0.1
0.2
0.3
0.4 0.5 0.6 0.7 Proportion of traffic via backbone
0.8
0.9
1
Fig. 7.10. Number of supported calls for 5% outage for different proportions of c local and non-local traffic, cell-centred network, multi-hop routing strategies 2005 IEEE.
It can be seen that for all traffic ratios the increase in allowed number of hops gives the benefit of slightly increased capacity for the simplest path loss routing strategy. This increase is slightly reduced as the majority of traffic moves from local to central. The best performer for two-hops, interference-based ODMA, is severely compromised by the move to an increased allowed number of hops, with performance little or no better than the non-interference based system. It is likely that this is due to the system being locked in by the initial starting condition, which comes from the path loss routing. This starting condition will utilise more and smaller hops on
Routing strategies in multi-hop CDMA networks
207
average than for two-hops, hence the iterative probing for low-interference routes is likely to involve introducing larger hops which will not result in lower interference. The problem is that the local minimisation for a larger number of hops generally finds a locally higher minimum than the two-hop scenario. A solution to this would be a search algorithm that transcends the local problem. The routing method that shows the greatest improvement with the move to multiple hops is congestion-based routing. It now outperforms all other routing strategies except for sending all traffic via the BS, where the pole capacity is the limiting factor. The improvement is due to the trellis search employed by the algorithm being allowed more possible routes, and these routes only being used if lower congestion results. The use of more hops allows for greater capacity from this increased diversity and also the nature of the smaller hops. With a smaller hop, the path loss will be lower, and without interference this will result in a lower transmitted power and hence interference. The congestion algorithm inherently takes interference into account, thus these shorter hops will be chosen in order to minimise the overall system impact.
7.7.2 Power The results in this section are all for two-hop systems with a BS or cell centred network as shown in Figure 7.2. Power results for multiple hops and MS/user-centred network allocation are presented in section 7.7.3. Figures 7.11-7.13 show the average power levels for the duration of a call, i.e. for a relaying scenario with two hops, this is the sum of the power for the two links. Figure 7.11 shows the situation for all local traffic. Both the direct and interference based ODMA have similar power levels, though it must be considered that the direct system has been required to put far more users into outage, e.g. to achieve 19 calls the interference based ODMA has placed one call into outage where the direct system has removed 13 calls. With the admission control described in section 7.4, this outage will have removed the worst interferers, creating a more favourable scenario than the interference based ODMA is experiencing as ODMA is more able to cope without forcing problem users into outage. In Figure 7.12, for equal local and non-local traffic, interference based ODMA shows a transmission power reduction over direct transmission for most scenarios. In Figure 7.13, with all non-local traffic, the transmission power is much less than the scenarios with local traffic due to power controlling only to one point. The interference
208
Next Generation Mobile Access Technologies
based ODMA requires increased transmission power at high user densities as paths are chosen with greater path loss to avoid high interference regions. Congest
0
Local 2-hop Int 2-hop
Total power per c all (dB m )
-20
Local Int Direct
-40
-60
-80
-100
14
16
18
20
22
24
26
28
30
32
34
Num ber of supported calls
Fig. 7.11. Total power per call against number of supported calls for all local traffic, c cell-centred network 2005 IEEE.
The local list-based strategies for ODMA show a greatly reduced transmission power over the centrally implemented systems for most situations. This is because, inherent in the list-gathering method, communication with other nodes is established according to a maximum transmit power. Hence users will only appear as available candidates on the list if the transmit power required for a link is below this maximum. This means that links that require a high-power transmission will not be considered even if they would be preferred solely on the basis of path loss. In all traffic conditions the simple ODMA produces the lower power than interference based ODMA and direct transmission for low user densities, due to some outage of problem users and the reduced transmission power and interference of ODMA. At higher user densities the required power is increased due to the higher number of links used.
Routing strategies in multi-hop CDMA networks
209
-10 -20
Total power per c all (dB m )
-30 -40 -50 -60
Congest Local
-70
2-hop Int
-80
2-hop Local Int
-90
Direct
-100 -110 16
18
20
22
24
26
28
30
32
34
Num ber of supported calls
Fig. 7.12. Total power per call against number of supported calls for 50% local c traffic, 50% non-local, cell-centred network 2005 IEEE.
The congestion based routing results in the lowest required average transmission power in all conditions, and shows very little increase with higher user densities. As will be shown in section 8.2.3 the congestion measure based on the Perron-Frobenious eigenvalue of (7.4) or (7.5) is directly related to the power control equation. Hence minimisation of this eigenvalue directly translates into a minimum power solution, which means that this technique may also be considered as minimum power routing.
7.7.3 Network allocation The results presented in the previous sections for capacity and power are all for BS/cell-centred target allocation as shown in Figure 7.2. This is a simple technique using the same criterion as would be used for a direct transmission system’s cellular hand-over. For a relaying system this is not necessarily ideal, as cellular loading may be uneven and as such it may be desirable to route a call via an adjacent cell’s BS, minimising cell interference
210
Next Generation Mobile Access Technologies -30
Congest
-40
2-hop Int
Local
Total power per c all (dB m )
2-hop Local Int
-50
Direct
-60
-70
-80
-90
-100
-110 20
22
24
26
28
30
32
34
36
38
Num ber of supported calls
Fig. 7.13. Total power per call against number of supported calls for all non-local c traffic, cell-centred network 2005 IEEE.
through the use of a relay in the adjacent cell. Cell-centred target allocation also reduces the number of potential relays if a MS is located near the edge of a cell compared to one near the centre as all MS in the adjacent cell are unavailable for relaying. This section presents comparative results for a simple alternative system centred instead on each MS, which is also shown in Figure 7.2. Figure 7.14 shows the number of supported calls for 5% outage for different ratios of local and non-local traffic, with the target allocation centred on the user instead of the cell, and a maximum of two hops. The direct transmission system shows an improvement over the BS-centred allocation where the traffic is 50% local or more. This is because there will be fewer peer-to-peer transmissions from one edge of the cell to the other. The interference based ODMA only shows improvement for 75% non-local traffic, showing that this system is already operating close to the limit for path loss initialised systems, managing to avoid the areas of high interference when they occur.
Routing strategies in multi-hop CDMA networks
211
The most dramatic change is with the congestion based routing. This now shows very little decrease in the number of supported users from the all non-local scenario down to 25% non-local traffic. For all local traffic there is an increase of over 40% in the number of supported users than for the BScentred target allocation. This demonstrates the extra capacity that may be available from a more appropriate choice of targets. 40
direct ODMAint congest star
Number of supported calls for 5% outage
35
30
25
20
15
0
0.1
0.2
0.3
0.4 0.5 0.6 0.7 Proportion of traffic via backbone
0.8
0.9
1
Fig. 7.14. Number of supported calls for 5% outage for different proportions of local c and non-local traffic, user-centred network. 2005 IEEE
Figure 7.15 shows a comparison between the power for multiple-hop systems for BS-centred target allocation (n0 ) and MS-centred allocation (n1 ) with 50% local 50% non-local traffic. It can be seen that the choice centred on the user results in a lower average power in all circumstances. This is a result of less congestion and hence power warfare at the centre of the cell. It should be noted that the multi-hop ODMA routed systems show a greatly reduced transmit power over the two-hop systems due to the combination of non-linear path loss and the greater localisation of interference resulting from shorter hops. Figure 7.16 shows the total average power for all local
212
Next Generation Mobile Access Technologies
traffic and the MS-centred target allocation. Unlike the BS-centred allocation, there is very little require increase in power with the move away from power control at one point (the BS). The plot for all non-local traffic is not shown as this is identical for both forms of traffic allocation. -60 Congest n1
-65
Total power per c all (dB m )
-70
Congest n0
*
-75
Trellis
n1
Trellis
n0
TrellisInt n1 TrellisInt n0
-80 -85 -90 -95 -100 -105 -110 20
22
24
26
28
30
32
34
36
38
Num ber of supported calls
Fig. 7.15. Total power per call against number of supported calls for 50% local c traffic, 50% non-local, comparing network types. 2005 IEEE
7.8 Conclusions In this chapter, a network topology has been investigated that allows both peer-to-peer and non-local traffic. We have presented a new admission control that allows congested areas to be identified and problems users to be removed. It has been shown that a conventional CDMA system is unable to produce performance comparable with a star topology. Path loss-based ODMA shows a capacity improvement over the non-relaying system and in most cases reduced transmission power. A new ODMA algorithm based on interference is presented that allows for greater capacity than a star topology
Routing strategies in multi-hop CDMA networks
213
Congest
-60
Trellis
*
n0
Total power per c all (dB m )
TrellisInt n0
-70
-80
-90
-100
-110 20
22
24
26
28
30
Num ber of supported calls
Fig. 7.16. Total power per call against number of supported calls for all local traffic, c user-centred network. 2005 IEEE
when scenarios involve at least 50% local traffic and comparable capacity when non-local traffic dominates. This provides the highest capacity of the presented strategies where there is a maximum of two hops and target allocation is based on conventional cellular handoff. A congestion-based routing algorithm is presented that attempts to minimise the overall power of the system as well as providing a measure of feasibility. This technique provides the lowest required transmit power in all circumstances, and the highest capacity in all cases except those outlined above for interference-based ODMA. A simple alternative to conventional handoff-based target allocation is presented. This shows capacity and power benefits in all circumstances. When combined with the congestion-based routing, a capacity increase of up to 44% over a conventional star topology is shown when peer-to-peer communication is involved.
8 Multi-hop DCA Tom Rouse Stephen McLaughlin
8.1 DCA techniques Congestion-based routing, as developed in the previous chapter, is shown to require the lowest transmitted power, and in most cases achieves the highest capacity of all the routing algorithms examined in Chapter 5. All of these routing algorithms have allocated TDD time slots on a first-comefirst-served basis and according to the rules outlined in section 6.4.1. This allocation only serves to ensure that the limitations of the TDD hardware are considered. It makes no attempt to optimise time slot allocation. The allocation of time slots with regard to system performance has been shown to be an effective technique to mitigate interference (Haas, 2000). Integrating slot allocation, or DCA, into the routing algorithm would appear to be the most effective approach due to the interactive nature of interference. This approach will also need to conform to the extra limitations imposed by relaying. In addition to the rules in section 6.4.1, it is obvious that the slot allocation must be in the same order as the relays. A combined DCA will allow minimisation of the desired measure, in this case congestion, simultaneously in routing and slot allocation. This chapter develops a combined routing and resource allocation algorithm for TDD-CDMA relaying. It starts by reviewing one such algorithm applicable to TDMA and FDMA. A novel method of time slot allocation according to relaying requirements is then developed. Two measures of assessing congestion are presented based on matrix norms. One is suitable for current interior point solution, the other is more elegant but is not currently suitable for efficient minimisation. 8.2 Combined routing and DCA algorithm This section presents a development of a simultaneous routing and resource allocation algorithm (Xiao et al., 2001) to include CDMA and explicit DCA. 214
Multi-hop DCA
215
8.2.1 Simultaneous routing and resource allocation An algorithm has been published that allows for simultaneous routing and resource allocation for multi-hop networks using FDMA or TDMA air interfaces (Xiao et al., 2001). CDMA is not considered as the authors could not adapt the capacity for an interference limited system to their model. Within TDMA, time slot allocation is only considered as a discrete resource, i.e. there are so many time slots each with a fixed capacity. There is no explicit allocation of time slots, especially with respect to the sequential nature required by a relaying system. The basic development of their algorithm consists of three stages. Firstly a network flow model is presented, then a relevant capacity model is coupled to the flow problem. This creates a linear program with convex constraints which can be solved globally by recently developed interior-point methods (Nesterov and Nemirovskii, 1994; Ye, 1997; Boyd and Vandenberghe, 2003). The algorithm concludes with an analysis of the associated dual problem (Strang, 1976) in order to improve algorithm efficiency; however, this final stage is not considered in this chapter. This section describes the formulation as described in (Xiao et al., 2001). Subsequent sections are additional to anything contained in that paper. The network topology is represented by a node-link incidence matrix A ∈ RN ×L , where N is the total number of nodes and L is the total number of links. This matrix has entries such that ⎧ ⎨ 1 : node n is transmitting on link l, Anl = (8.1) −1 : node n is receiving on link l, ⎩ 0 : otherwise
For example the system as shown in Figure 8.1† would have an incidence matrix given by ⎡ ⎤ 1 −1 −1 1 0 0 A = ⎣ −1 (8.2) 1 0 0 1 −1 ⎦ . 0 0 1 −1 −1 1
In order to specify the nodes between which communication is desired a source-sink vector s(d) ∈ RN is introduced for each destination d, where d = 1, ..., D. The nth (n = d) entry is the flow injected into the network by node n. Due to the conservation of flow, the sink flow at d is defined as (d) s(d) (8.3) sd = − n . n,n=d
† Figures 8.1, 8.2, and 8.3 are reproduced with permission from: (Rouse et al., 2005).
216
Next Generation Mobile Access Technologies
3
3
1 4 2 5 1 6
2 c Fig. 8.1. Link connectivity example. 2005 IEEE (d)
For each link, a flow vector xl determines the amount of flow destined of destination d where x(d) ∈ RL+ . Hence the total amount of traffic for link (d) l is given by tl = d xl . Requiring that the traffic does not exceed the capacity, the following minimisation problem is defined: minimise f (x, s, t, r) subject to Ax(d) = s(d) , d = 1, ..., D x(d) 0, s(d) d 0, d = 1, ..., D , (d) l = 1, ..., L tl = d xl , l = 1, ..., L tl ≤ φl (rl ),
(8.4)
where φ(r) is the capacity as a function of the communications variables r, means component-wise inequality and d means component-wise in-
Multi-hop DCA
217
equality except for the dth component. If the function φ is concave and monotone increasing in r and the objective function is convex then this is a convex optimisation problem. This type of problem can be solved globally and efficiently by recently developed interior-point methods (Nesterov and Nemirovskii, 1994; Ye, 1997; Boyd and Vandenberghe, 2003).
8.2.2 Time slot allocation The allocation of time slots is not explicitly addressed by the formulation in (8.4). This problem is more difficult than for the single-hop DCA case outlined previously. This is because, unless it is tolerable to introduce sufficient buffering and hence extra packet delay, the next routed hop needs to take place in the subsequent time slot. Previous implementations of a TDDCDMA multi-hop routing algorithm (Harold and Nix, 2000a), and Chapters 3 and 4 have used arbitrary time slot allocation subject to the restrictions outlined in section 6.4.1. This section presents a formulation of the linkincidence matrix that allows for integrated routing and time slot allocation, or DCA. In the formulation (8.4) there is no way of determining in which time slot the link occurs. If the link-incidence matrix were to be explicitly represented for each time slot this would become possible. An important feature of (8.1) is that for the equality contained in (8.4), Ax(s) = s(d) d = 1, ..., D ,
(8.5)
Axl = sl ,
(8.6)
or, more generally,
where xl is the collection of flow vectors and sl the collection of sourcesink vectors. If node n is not the destination, sln will be zero. This can be interpreted as the necessity that if there is flow incident to a node then there must be a corresponding and equal flow leaving the node, hence flow is conserved. This is achieved by having 1’s and -1’s on the same row with appropriate values in the flow vector. Therefore, in order to define the time discrete link incident matrix, and require flow into the next time slot, it is necessary to separate (8.4) into transmitting and receiving matrices, A+ and A− respectively, where A+ contains only the positive entries of A and A− only the negative entries. Thus the links between the nodes can be
218
Next Generation Mobile Access Technologies
represented in a time discrete fashion as ⎧ ⎨ A+ c = 1, ..., C ζ= c = 1, ..., C A ⎩ − 0 otherwise,
(8.7)
where
A+ = ζ((c−1)×N +1):(c×N ),((c−1)×L+1):(c×L) , A− = ζ(c×N +1) mod (C ×N ):((c+1)×N mod (C ×N ),((c−1)×L+1):(c×L)) , and C is the total number of ⎡ A+ ⎢ A− ⎢ ⎢ 0 ⎢ ⎢ ζ=⎢ 0 ⎢ . ⎢ .. ⎢ ⎣ 0 0
(8.8)
time slots. Therefore ζ will take the form ⎤ 0 0 ··· 0 A− A+ 0 ··· 0 0 ⎥ ⎥ 0 ⎥ A− A+ · · · 0 ⎥ 0 ⎥ 0 A− · · · 0 (8.9) ⎥. ⎥ .. .. .. .. .. ⎥ . . . . . ⎥ 0 0 · · · A+ 0 ⎦ 0
0
···
A− A+
Hence the receiving nodes are aligned with the same node transmitting in the next time slot. This means that the only way to satisfy (8.6) is by transmitting in the next time slot if there is flow incident into a node. A problem with this formulation is the specification of s. If a non-zero value is allocated, then for a positive value the packet initiation, and for a negative value the packet reception is tied to a specific time slot. Clearly this is incompatible with our objective of combining the routing algorithm and time slot allocation. To overcome this failing we can break the problem down further than the time slots. There are four possible classifications of a link: (i) Neither transmitter or receiver is an initiating source or a terminating sink, i.e. both are relays. This is described by ζ as the corresponding value sln for node n is 0. (ii) The transmitter is an initiating source, the receiver a relay. sln is positive for the transmitter, 0 for the relay. (iii) The transmitter is a relay, the receiver the destination sink. sln is 0 for the transmitter, negative for the relay. (iv) Transmitter is an initiating source, the receiver the destination sink. This corresponds to the non-relaying scenario. sln is positive for the transmitter, negative for the relay.
Multi-hop DCA
219
If these modes can be incorporated into our link incidence matrix as separate entities, then it is possible to specify sl such that any time slot is available for route initiation or termination. This may be achieved by breaking each link into the four types described above. This split flow is described by x, the sum of which is equal to xl , and the sum of each group corresponding to link l is equal to the value of xl for that link. Thus the link incidence matrix is expanded column-wise four times for the different modes, has an extra 2N rows to facilitate the specification of sl for required source and sink nodes. The required link incidence matrix may be broken down into four sub-matrices, T u , T t , T r , and T d , respectively, for each of the cases above: TNu +1:(C +1)×N = ζ, u (8.10) T = = 0 otherwise, ⎧ t ⎪ ⎨ T1:N,((c−1)×L+1):(c×L) = A+ c = 1, ..., C, Tt = TNt +1:(C +1)×N = ζ− , ⎪ ⎩ =0 otherwise,
(8.11)
⎧ r ⎪ = ζ+ , ⎨ TN +1:(C +1)×N r r T = T(C +1)×N +1:(C +2)×N,((c−1)×L+1):(c×L) = A− c = 1, ..., C, ⎪ ⎩ =0 otherwise, (8.12) ⎧ d ⎪ = A+ c = 1, ..., C, ⎨ T1:N,((c−1)×L+1):(c×L) d d T = T(C +1)×N +1:(C +2)×N,((c−1)×L+1):(c×L) = A− c = 1, ..., C, ⎪ ⎩ =0 otherwise, (8.13) where ζ+ contains only the positive entries of ζ, and ζ− only the negative entries. The composite time discrete link incidence matrix, T , can be written as T = [T u T t T r T d ],
(8.14)
though the change of order of the sub-matrices, or indeed any column exchange, is allowed. The only change resulting from such a transformation will be that the corresponding value for flow in x will appear in the exchanged position when the solution to the system is found. Reformulating sl as s for the time discrete link incidence matrix T : ⎧ = sl+ , ⎨ s1:N s= (8.15) = sl− , s ⎩ (C +1)×N +1:(C +2)×N =0 otherwise.
220
Next Generation Mobile Access Technologies
The condition Tx = s
(8.16)
can replace (8.6) in (8.4), the resulting x in the solution will contain the DCA allocation, and the solution will be the time slot allocation such that the solution is a global minimum.
8.2.3 A CDMA feasibility condition The initial work on congestion (Hanly, 1999) has been mainly concerned with determining system feasibility. We can show that a congestion measure is a sufficient condition for feasibility by starting from the initial power control equality for a single receiver: ⎞ ⎛ αi ⎝ (8.17) Pu Γ[u, ci ] + η[ci ]pg ⎠ , Pi Γ[i, ci ] = pg u=i
where η[k] is the power of the unspread noise in the system. The power control problem takes the form: (I − A)P = bM ×1 ,
(8.18)
where bM ×1 is the M × 1 vector with ith entry b[i] = η[ci ]αi /Γ[i, ci ]. As it is necessary for all elements of p and b to be ≥ 0, it is a necessary and sufficient condition (bar maximum power constraints) that (I − A) be invertible and (I − A)−1 be non-negative. Substituting the eigenvalue problem Ax = λx
(8.19)
(1 − λ)P = b,
(8.20)
into (8.18) gives
hence the condition that the largest, in this case Perron-Frobenius, eigenvalue is < 1. In order to integrate the congestion matrix into the link-based formulation of (8.4) we need to be able to modify the matrix according to the traffic flowing on a particular link. Examining (7.4) we can break the matrix into two parts, the path loss ratio, G, for all links; and the QoS element, Q. G is formed from the path gain matrix for a single time slot: pg : Rxi = Txu or Rxu = Txi , v (8.21) H[i, u] = Γ[Txu ,Rxi ] : otherwise, Γ[Txi ,Rxi ]
Multi-hop DCA
221
where Txn and Rxn are respectively the transmitter and receiver for link n and v is the minimum quanta of flow. The term for conflict in Tx and Rx means that simultaneous Tx and Rx infeasible in the formulation. G is formed with H on the diagonal for each time slot: ⎡ ⎤ H 0 ··· 0 0 ⎢ 0 H ··· 0 0 ⎥ ⎢ ⎥ ⎢ .. .. .. ⎥ .. G = ⎢ ... (8.22) . . . . ⎥ ⎢ ⎥ ⎣ 0 0 ··· H 0 ⎦ 0 0 ··· 0 H The QoS element, Q, can be formed directly from the required traffic flow: Q=
1 diag(t), pg
(8.23)
where t is the vector containing the traffic for all the links. Hence the congestion matrix for the link-based approach may be written as A = QG.
(8.24)
The Perron-Frobenious eigenvalue for this matrix is given by the spectral norm |•|2 (Horn and Johnson, 1985) since it is defined as √ (8.25) |A|2 ≡ max λ : λ is an eigenvalue of A∗ A. Since it is an induced norm, and λ can be calculated from x∗ A∗ Ax = λx22 .
(8.26)
Hence we can reformulate (8.4) to include CDMA as follows: minimise x∗ (QG)∗ (QG)x subject to T x = s x(d) 0, s(d) d 0, d = 1, ..., D (d) l = 1, ..., L tl = d xl , (I − QG)p = b pmin p pmax .
(8.27)
Unfortunately this formulation(8.27) cannot be solved by current interiorpoint methods, although if the progress on minimising the eigenvalues of asymmetric matrices, e.g. (Burke et al., 2001; Burke and Overton, 2001), continues at its current rate this may soon be possible. Until that time we can break the problem down somewhat.
222
Next Generation Mobile Access Technologies
The Euclidean, l2 , or Frobenius norm, •F , is defined (Horn and Johnson, 1985) as ⎛ ⎞1/2 n (8.28) |aij |2 ⎠ . AF ≡ ⎝ i,j=1
• F is not a vector bound norm, as |•|2 is, but it is compatible with |•|2 (Meyer, 2001), hence the inequality |A|2 ≤ AF .
(8.29)
Therefore, we can calculate an upper bound on λ without the complexity of calculation for an induced norm. One implication of this simplification is that we can no longer use the pg/v factor to eliminate simultaneous Tx and Rx for a node. This is because, with the removal of the vector term, these factors may no longer be multiplied by zero, and will dominate the Frobenius norm. One approach to this problem is to allocate nodes to odd or even time slots prior to the optimisation. This removes some degree of freedom from the routing algorithm, hence the DCA may not be as favourable as in (8.27). There is still a great deal of integration between routing and DCA as the formulation in (8.16) is still applicable. The preallocated time-slot congestion matrix may be formed from H ∗ , where $ o % H 0 ∗ . (8.30) H = 0 He
Initially receiving nodes are arbitrarily split between H o and H e and the matrices formed as in (8.21) with the proviso that a node is only available as a transmitter if it is not a receiver in that time slot. This means that no terms to prohibit simultaneous Tx/Rx appear. There follows an exchange of receiving nodes between H o and H e until the Frobenius norm for this matrix is minimised. G is now formed from T /2 entries of H ∗ (as H ∗ covers two time slots) as ⎤ ⎡ ∗ H ··· 0 ⎥ ⎢ . .. (8.31) G = ⎣ ... ⎦. . .. 0
···
H∗
We can now write the minimisation problem n n 2 minimise i,j=1 | k=1 qik gkj | subject to T x = s x(d) 0, s(d) d 0, d = 1, ..., D (d) l = 1, ..., L, tl = d xl ,
(8.32)
Multi-hop DCA
223
which may be solved globally and efficiently through interior-point methods as a semidefinite program (SDP) (Helmberg et al., 1996). The power allocation may then be performed as a minimisation problem: minimise P subject to (I − QG)P = b Pmin P Pmax .
(8.33)
There may then be a feedback loop between (8.33) and (8.32) which alters s according to requirements for outage probability, power allocation, or traffic maximisation.
8.3 Results The results presented in this section compare the congestion-based routing explained in Chapter 6 with the DCA outlined in this chapter. All results are for the square scenario described in Chapter 6, a processing gain of 12.9 dB, a bandwidth of 2048 kHz, user-centred network allocation, and a required signal to noise ratio of 3 dB. The throughput is calculated for all users maintaining the same data rate, whilst maximising this data rate by stacking transmissions. The limiting factor is that no user may exceed the maximum transmit power with its required overall transmit power. The results showing the performance of the DCA compared to the nonDCA congestion based routing are shown in Figure 8.2. These results use a system with eight separate time slots available to 28 users. It can be seen that the throughput of the DCA-based routing outperforms the nonDCA-based routing in every situation, except where all traffic is routed via the BS. In this case the throughput is limited by the pole capacity of the BS. In Chapter 4 this non-DCA-based routing is shown to have the highest capacity of all the non-DCA systems investigated. The performance gain of the DCA increases as the proportion of ad hoc traffic increases. For entirely ad hoc traffic there is an increase of about 140% over the congestionbased routing, and while not shown, there is almost a ten-fold increase over a non-relaying star topology (where two links are required for in-cell communication). This increase can be entirely attributed to resource reuse and interference avoidance inherent in the minimisation problem of (8.32). Results for different numbers of timeslots available to the DCA are shown in Figure 8.3. This corresponds to the number of H terms in (8.22). It can be seen that the non-DCA system is not able to exploit an increased number of time slots. This is due to the arbitrary time slot allocation of this routing algorithm outlined in section 6.4.1. As would be expected for the DCA, extra
224
Next Generation Mobile Access Technologies
2500 dca no dca
Throughput per slot (kbps)
2000
1500
1000
500
0
0
0.1
0.2
0.3
0.4 0.5 0.6 0.7 Proportion of traffic via backbone
0.8
0.9
1
Fig. 8.2. Throughput against proportion of traffic via backbone for 8 timeslots, 28 c active users, and 12.9 dB processing gain. 2005 IEEE
time slots allow for an increase in throughput, with about a 20% increase per slot by going from 2 to 12 slots. After this point there does not seem to be any considerable increase in throughput. The most significant increase in throughput with number of slots occurs just before this plateau. It would seem reasonable to assume that the increased throughput is generated by the DCA having more options/flexibility in reducing congestion with an increased number of time slots. It is interesting to note that the DCA performs almost as well with two slots as with six, which could either be taken as the two-slot case coping well or the six-slot not taking full advantage of the flexibility due to the compromises made in (8.32). The results in Figure 8.4 show throughput against number of users, again with eight time slots. For the non-DCA routing it can be seen that the increase in user numbers causes a reduction in the available throughput. This is considered to be due to the extra terms not controlled to the same point introduced into the power control problem, and subsequent increased interference. Moving from 20 to 24 users the DCA system also suffers this
Multi-hop DCA 3000
225
dca no dca
Throughput per slot (kbps)
2500
2000
1500
1000
500
0
2
4
6
8 number of slots
10
12
14
Fig. 8.3. Throughput against number of time slots for 28 active users and 12.9 dB c processing gain. 2005 IEEE
drop in throughput. After this point, however, an increase in throughput occurs. It is likely that there is a balance between number of active links creating interference, and the number of different available links allowing for more routing opportunities. The DCA seems more able than the non-DCAbased routing to exploit these opportunities and thus utilise the increased number of active users in its favour. 8.4 Conclusions A novel DCA algorithm has been developed exploiting the novel congestion based routing from Chapter 4. Simultaneous routing and resource allocation was formulated for TDD-CDMA. Specifically, time slot allocation has been performed in a novel manner, which produces specific sequential allocation in the same order as required for relaying. The novel formulation of the time slot matrix, with discrete allocation for each slot, enables the minimisation of a congestion measure, through
226
Next Generation Mobile Access Technologies
1500 dca no dca
Throughput per slot (kbps)
1250
1000
750
500
250
0 20
22
24
26 28 30 Number of active users
32
34
36
Fig. 8.4. Throughput against number of number active users for 8 timeslots and 12.9 dB processing gain.
combined routing and time slot allocation, in order of hop. This allocation reduces delay and terminal complexity, and ensures a feasible relaying configuration. The matrix structure, in addition to performing time slot allocation and routing, enables the selection of any BS for optimal relaying hand-over, according to congestion. A congestion measure was presented that is suitable for minimisation though interior-point methods, which find a global solution. This formulation requires an initial partitioning of relaying nodes, and a final phase of power allocation. Another formulation was presented, which, whilst unsuitable for minimisation with current interior-point methods, requires no pre-partitioning, integrates power allocation for maximum C/I, and should further improve performance beyond the interior-point-compatible version. Results for the multi-stage DCA showed increased throughput compared to the non-DCA congestion based routing, except when all traffic is routed via the BS, where performance is identical. For entirely peer-to-peer traffic, throughput was shown to increase by 140% over the best non-DCA routing,
Multi-hop DCA
227
and by almost 1000% over a non-relaying system. Increasing the number of time slots considered in the allocation improves throughput, at the cost of computational complexity. Furthermore, the increased opportunity provided by a higher number of active users was exploited by the DCA. Other routing approaches in this book have actually shown slightly impaired performance for an increased numbers of users, through failure to mitigate the increased number of interference sources.
9 Radio resource metric estimation Yeonwoo Lee Stephen McLaughlin
Until now, the focus of the book has been on interference analysis and management for either a cellular TDD-CDMA system or an ODMA enhanced TDD-CDMA system. Interference management can be viewed as a form of resource management. In this chapter the issue of link adaptation is addressed with a focus on what metrics are appropriate to enable radio resource management in a cross-layer manner. In conventional systems, the decision on how to choose the ideal physical mode (PHY-mode) is primarily based on the knowledge of the interference encountered. This information is reported to higher-layer entities that deal with the radio resource management. Subsequently, the radio resource management entity makes a decision as to which physical resource will be used (e.g. by employing a DCA algorithms as described in Chapter 5 and 8) and this information is reported back to the physical layer. It is apparent that this process can take quite a long time. Meanwhile the channel conditions may have changed significantly. Hence, the previously chosen radio resource may no longer be the ideal choice. Therefore, new methods (e.g. resource metric estimation) are discussed here that (a) base the decision as to which radio resource to use, not only on interference, but also, for example, on the statistics of the channel state information, and (b) make the decision as to which radio resource to use already at the physical layer. For (a) it is shown that the TDD mode is ideally suited due to the reciprocity of the channel. The performance of the new methods presented is assessed in terms of throughput and delays. This also includes reports on the interaction between new resource metrics and other link adaptation techniques such as adaptive modulation/coding and power control. This chapter consists of two major sections ;the first section deals with a radio resource metric mapping function and the second with a radio resource metric region. In section 9.2, the effect of wireless internet traffic 228
Radio resource metric estimation
229
on the functionality of a radio resource metric estimation (RRME) applicable to a TDD-CDMA system which incorporates a multirate transmission scheme (such as multicode and multislot) with simple call admission control algorithm is investigated. A solution for efficient inter-working between the physical layer and the higher layers is proposed; a resource metric mapping function (RMMF) that can deliver information about the state of the current channel load condition by the monitoring of signal-to-interference ratio bursts and also the availability of a code pool by estimating the mean and the standard deviation of SIR bursts. This function is defined and demonstrated by means of an average raw BER mapping diagram as a function of the mean and the standard deviation of SIR bursts according to the multirate transmission method. By using this kind of mapping function combined with throughput estimation, the radio resource allocation algorithm can successfully reflect the current interference and mobile radio channel characteristics. In section 9.4, as a solution for efficient cross-layer inter-working, the radio resource metric region (RMR) concept is described, which can provide the acceptable resource region where QoS and acceptable link quality can be guaranteed with an achievable resource margin to be utilised in terms of the resource metric mapping function (RMMF).
9.1 Radio resource metric estimation applied to radio resource allocation The growing demand for multimedia services over wireless communication systems has accelerated a need for more efficient utilisation of the scarce and expensive wireless resource. In addition, it is important for future differentiated services which will be delivered via mobile networks that can adapt (adaptation used in the context of resource management) the utilisation and allocation of resources. In next generation wireless communication systems the optimisation of resource use in the radio resource management (RRM) will be key. This optimisation can only be made possible with the development and adoption of efficient algorithms. Current RRM functionality in 2G and 3G systems acts from the network layer down to the data link and physical layer, thus the wireless physical resources are utilised directly without intermediate layers. Future RRM in next generation networks will be much more complex. Therefore, research work incorporating an RRM that takes account of the conditions of the physical layer resources suggests the need to focus on its potential impact in the intermediate layer; that is, a radio resource metric estimation (RME) that can be aware of the multiple layer
230
Next Generation Mobile Access Technologies
c Fig. 9.1. Resource metric estimator in RRA algorithm at the BS. 2004 J. Wiley
capabilities and states. Based on the knowledge of the desired loads and channel and radio resources, the RRM in cooperation with RME can manage both up and down the protocol stack. Thus, it can decide and control the parameters and functions required to optimise the desired features such as QoS, throughput, power utilisation and overall system capacity. RME is a crucial part of the radio resource allocation (RRA) algorithm that performs call admission control (CAC), resource scheduling and power/rate scheduling tasks, which provide the following control tasks as shown in Figure 9.1†. • the radio channel characteristics and session quality requirements are used for optimal power and rate allocation; • the current channel load, characteristics and quality requirements are used for controlling the resource scheduler (Jorguseski et al., 2001). With built-in capacity models, the RME assists the CAC in accepting or rejecting new sessions. Hence, the RME is the crucial part of the RRA algorithm because of its important role in providing the measurements and determining the impact on other parts of the RRA algorithm. The question of how to combine the interference measurements with the current load situation and QoS requirements of the existing traffic classes to control CAC, or channel allocation is a very interesting issue. The most appropriate resource metrics should permit efficient inter-working between the physical layer and higher layers in the protocol stack and thus, it is essential to optimise the overall system performance. From the physical layer point of view, the most accurate method to access the actual measurements of the link quality † Figures 9.1 - 9.20, and 9.24 and Table 9.1 are reproduced with permission from: (Lee and McLaughlin, 2004).
Radio resource metric estimation
231
estimation should be taken into account rather than just considering the resource management point of view. The current generic method of assessing resource metric is to use link quality information in a long-term averaged manner. Current resource management use this information in 2G and 3G systems and work from the network layer down to the data link and physical layer without any intermediate layers (H¨am¨al¨ ainen et al., 1997); (Oloffson et al., 1997). Typically, the radio network simulations can be divided into two parts: link level and system level simulations. The link level simulation covers the physical layer including accurate receiver performance evaluation at a chip level or symbol level. In the system level simulation the traffic models and multi-cell network are simulated. The system level simulation can be used to evaluate the performance of the CAC or RRA algorithm as applied to the higher layer as shown in Figure 9.1. A single simulator approach considering both layers would be preferred but the complexity of such a simulator is far too high due to the required simulation resolutions and differing simulation timescales. Therefore, separate link level and system level simulators are needed. In such case, it is essential to define the interface between the link level and the system level simulations. As a solution of how the RME can provide fast and reliable resource information to a higher layer in an efficient manner, we propose a radio resource metric mapping function (RMMF) associated with the RRA. This function deals with the issue of how to combine the interference measurements relevant to current load and traffic condition to control call admission control. In our study we consider the RMMF applied to a TDD-CDMA system supporting non-real time packet service, e.g. Web browsing service, in terms of two multirate transmission strategies, multislot and multicode.
9.2 Radio resource metric mapping function A conventional system level simulator with a large number of mobile terminals and several base stations is capable of calculating the average signalto-interference ratio (SIR) per simulation time step as an input parameter. By doing so the effects from multipath fading and interference variation are taken into account in terms of their average characteristics. Usually the measured BER is the average value over the whole link level simulation run. However, the average SIR sample does not fully reflect the perceived link quality for users causes fast fading and time-varying interference characteristics; this means that a certain average SIR may result in very different user link quality. Hence, in order to reflect the actual link quality perceived
232
Next Generation Mobile Access Technologies
by the user and map the SIR sample onto actual link quality in terms of the block error rate (BLER) or average raw bit error rate (BER), an accurate and fast interface between link level and system level is needed. In the system level simulation, the SIR sample must be mapped on to the link quality via the correct look-up table, taking a radio mobile channel’s characteristics and interference into account. The interface between the link level and the system level simulations as applied to a GSM system was studied in (Oloffson et al., 1997). This study proposed similar methodology to here, calculating instantaneous raw BER on the basis of burst-by-burst SIR samples. In this study, the main concept in (Oloffson et al., 1997) is applied to establish the RMMF. The performance of the RMMF is assessed for a TDD-CDMA system supporting bursty and asymmetric traffic such as WWW traffic. This concept enables efficient utilisation and monitoring of resource usage using the interface between link level and system level simulations approach which permits dynamic resource metric estimation, that is the current channel load conditions and resource pool condition can be achieved in a non-averaged manner. The procedure of this mapping function can be summarised in the following stages.
9.2.1 Building up the procedure for the radio resource mapping function Channel Characterisation For each user link, the radio channel characteristics can be estimated at the channel estimator on a burst-by-burst basis, but a number of these measurements are usually averaged out for several time slots or bursts in the system level simulation. Due to fast user mobility, the channel measurement period is an important factor in deciding the fidelity of estimates at each burst. The interference has to be calculated and predefined time basis while the corresponding interference situation changes due to the time-varying fading process. Since the SIR can be calculated by the ratio of received power to total interference plus thermal noise, with K active interferers and thermal noise power N the burst SIR can be expressed as i SIRburst
L
= l=1 K
i α2 β 2 Ptx l
k=1 Ik
+N
,
(9.1)
i is the transmitted power of the ith burst, and α and β are the where Ptx l path loss and slow fading gain and the magnitude of the lth path of fast
Radio resource metric estimation
233
Fig. 9.2. Block diagram of resource metric mapping function procedure; ⌈x⌉ denotes c the largest integer near x. 2004 J. Wiley
fading, respectively, and L is the number of multipath components. The interfering power Ik can be calculated as Ll=1 Ptx α2 βl2 . Extracting Burst Information from the Link Level For a TDD-CDMA system, the channel can often be assumed to be invariant during the transmission of one time slot, e.g. one data traffic burst. Under this assumption, the interface between link level and system level could be carried out on burst-by-burst basis. Thus, the link level performance needs to be derived as a function of instantaneously measured burst SIR. Link level performance based on burst-by-burst link quality such as BLER and BER can be easily extracted from a standard link level simulation rather than relying on average values as a function of averaged SIR with several i ) for data bits in instantaneous SIR measurements. The raw BER (Pburst the ith burst as a function of the actual SIR burst measurements during the burst can be extracted. The SIR is estimated and mapped to raw BER
234
Next Generation Mobile Access Technologies
i function obtained from the link level by using a raw BER versus SIRburst simulation results, which can be expressed as i i = f (SIRburst ) rawBER = Pburst
(9.2)
i is a random variable depending on fast fading characteristics. where SIRburst Then, the average raw BER (P bk ) of the kth set of bursts is extracted as a function of the mean of the BER (μkBER ) or the mean of the SIR (μkSI R ) in the subsequent time slots (i.e. bursts) belonging to a frame, and the k k ) or (σSI standard deviation (σBER R ) of those BER or SIR values. With similar reasoning to (Oloffson et al., 1997) and some modification, these parameters can be derived as follows:
μ ˆkSI R =
k σ ˆSI R
and k σ ˆBER
n
n
i=1
i=1
1 1 i i SIRburst andˆ μkBER = Pburst , n n
1 2 2 =3
(9.3)
n
1 i E[(SIRburst −μ ˆkSI R )2 ], (n − 1)
(9.4)
i=1
1 2 n n 2 1 1 i i i 3 = Pburst (1 − Pburst ) + (Pburst −μ ˆiburst )2 , (nlb ) (n − 1) i=1
i=1
(9.5) respectively, where n is the number of bursts in a burst-set and lb is the number of bits in a burst. This relationship can be illustrated in a threedimensional diagram with the very reasonable assumption that the average raw BER depends on the instantaneous raw BER in the bursts (see Figure 9.13). This representation of the RMMF can be used to replace a conventional SIR look-up table. Estimating Link Quality The task, after taking SIR estimates for each burst, is to map a sequence of such SIR values onto a perceived link quality by means of the previously obtained link level results. As shown in Figure 9.2, the parameter n is the size of a sequence set of the measured SIR bursts. The parameter m is the total number of bursts; a typical length of a SIR sequence may be 125, which corresponds approximately to a packet call in a Web-browsing service (P´erez-Romero et al., 2000). The number of bursts in a frame will be
Radio resource metric estimation
235
changed and determined according to the time slot allocation scheme and traffic asymmetry. The mapping method comprises the following steps as shown in Figure 9.2: i for the ith • Measure a sequence of burst SIR values and determine Pburst burst by means of the mapping function: rawBER = f unction(SIRburst ). • Estimate the average raw BER (ˆ μkBER ) and the average SIR (ˆ μkSIR ) for the k ) kth set of bursts and each standard deviation of the BER values (ˆ σBER k and the SIR values (ˆ σSIR ). k k • Map each parameter pair (ˆ μkBER , σ ) on the average ˆSIR ) or (ˆ μkSIR , σ ˆBER BER P bk by using the link level results and estimate link quality of each segment by using obtained sequence of P bk .
9.2.2 Block error rate (BLER) mapping function and user data throughput On top of the average BER, the block error rate (BLER) and link utility function can be used as link quality parameters. If we use a block error correcting code able to correct up to t errors, we can evaluate the probability μkSIR ), as: of having a correctable block, PC (ˆ μkSIR ) = PC (ˆ
t
n=0
(Lb )P b (ˆ μkSIR )n (1 − P b (ˆ μkSIR ))L b −n ,
(9.6)
where Lb is the length of block, and the average block error probability BLER is given by μkSIR ). BLER = 1 − PC (ˆ
(9.7)
This analysis could be extended for other coding schemes and error correctability such as convolutional or turbo codes, etc. (P´erez-Romero et al., 2000). With these equations, the estimated average BER P bk can be mapped to BLER and thus we can build up the resource metric mapping function of SIR burst to BER and BLER. Consider a 1/3 rate (r = 1/3) convolutional correcting code that is able to correct up to 7 (t = 7) for Lb (= 244) bits length information block (see Figure 9.3). Now we can define the user data throughput as the number of correctly received bits per second (Goodman and Mandayan, 2001); (Sung and Wong, 2001):
236
Next Generation Mobile Access Technologies BLER versus Raw BER mapping functio n 1
0.9
0.8
0.7
BLER
0.6
0.5
0.4
0.3
0.2
0.1
0.01
0.02
0.03
0.04 0.05 0.06 Average BER = f(SIR)
0.07
0.08
0.09
0.1
Fig. 9.3. BLER mapping function as a function of average BER and SIR burst (1/3 c rate convolutional code for Lb = 244 bits). 2004 J. Wiley
S=
μkSI R ) Lb PC (ˆ = Rb PC (ˆ μkSI R ) = Rb (1 − BLER), Lb /Rb
(9.8)
μkSI R ) indicates the expected number of successfully transmitwhere Lb PC (ˆ ted information bits and the denominator Lb /Rb is the total consumed time for transmitting Lb information bits. 9.3 Multirate transmission operations for a TDD-CDMA system 9.3.1 WWW Traffic Modelling The non-real-time WWW traffic service model proposed in (ETSI 30.03, V3.2.0 (1998-04), 1998) is used here. The user activates a WWW session with Poisson arrival rate λ. Each session Npc will contain a number of packet calls, which is a geometric random variable with a mean of 5. The reading time will also be geometric with a mean of 412 s. The packets are Pareto-distributed with a mean of 25. The IP packets have a fixed size of 480 bytes. The inter-arrival time between two packets inside a packet call is geometrically distributed with a mean of 10.4 ms. In a TDD-CDMA system
Radio resource metric estimation
237
model, transport can be performed within each time slot of 666 μs duration. Therefore one packet must be split into several time slots for transmission over the radio link. The typical behaviour of the packet data traffic is illustrated in Figure 9.4. The properties that are typical for non-real-time packet services from the air interface point of view are bursty, tolerant to longer delay, and re-transmittable. Although it is possible to transmit realtime service over packet network such as voice over IP, the transmission of real-time services is not considered in this study. The non-real-time WWW traffic service model proposed in (ETSI 30.03, V3.2.0 (1998-04), 1998) is used in this paper. The user activates a WWW session with Poisson arrival rate. Each session will contain a number of packet calls, which is a geometric random variable with a mean of 5. The reading time will also be geometric with a mean of 412 s. The packets are Pareto-distributed with a mean of 25. The IP packets have a fixed size of 480 bytes. The inter-arrival time between two packets inside a packet call is geometrically distributed with a mean of 10.4 ms. In a TDD-CDMA system model, transport can be performed within each time slot of 666 μs duration. Therefore one packet must be split into several time slots for transmission over the radio link. The typical behaviour of the packet data traffic is illustrated in Figure 9.4. The properties that are typical for non-real-time packet services from the air interface point of view are bursty, tolerant to longer delay, and re-transmittable. Although it is possible to transmit real-time service over packet networks such as voice over IP, the transmission of real-time services is not considered in this study.
9.3.2 Multirate transmission operations In contrast to voice traffic, packet data traffic is normally bursty, and its source information is segmented into packets of equal or variable length depending on its statistical distributions as described in the previous section. For a TDD-CDMA system, in each time slot a simultaneous transmission of up to 16 traffic bursts by means of different code sequence is possible. Thus, each resource unit (RU) can be specified by code and time slot. In this configuration, 240 RUs (16 codes and 15 time slots) are available per BS. Multirate transmission services are achieved by the pooling of multiple codes within one time slot (i.e. multicode allocation) or by the pooling of multiple time slots (i.e. multislot allocation). Additionally, any combination of both technologies is possible. In the multislot approach a large number of users can have a low bit rate channel available simultaneously, while the instantaneous bit rate per packet user is low. When users require a capacity increase, then the allocated
238
Next Generation Mobile Access Technologies
c Fig. 9.4. Characteristics of a packet data service session. 2004 J. Wiley
number of time slots per user increases. The simple example of multislot allocation configuration with fixed transmission rate per time slot is shown in Figure 9.5. The number of allocated time slots depends on the packet size in a user buffer since the allocated code and bit rate per time slot is kept fixed. In this study a fixed single time slot switching point is used so that the DL utilise the maximum available time slots for WWW service, i.e. maximum DL asymmetry of 14:1. In the multicode approach the average delay will be shorter than with the multislot approach since the instantaneous bit rate is higher. From the allocation algorithm point of view, the shorter delay is an advantage with this approach, but one disadvantage is the short transmission time. Furthermore, we note that the multicode approach uses higher bit rates and generates more bursty traffic, which leads to higher variations in the interference levels than in the multislot approach. An example of multicode allocation configuration with multiple transmission rates per time slot is shown in Figure 9.5. Here, the number of allocated codes depends on the packet size in a user buffer since the allocated time slot and bit rate per each code/time slot is kept fixed.
Radio resource metric estimation
239
Fig. 9.5. Multirate transmission operation example in a TDD-CDMA system frame. c 2004 J. Wiley
9.3.3 TDD-CDMA system model and operation for multirate transmission The basic UTRA TDD-CDMA system parameters such as chip rate, bandwidth, and modulation are used. Each frame is of length 10 ms, divided into 15 time slots each of 2560 chips, i.e., the TS duration is 666 μs. Each time slot may be allocated to either the uplink (UL) or the downlink (DL) (3GPP, TSG, RAN, 2000b)(Haardt et al., 2000) (Holma and Toskala, 2000). Within a time slot users are separated in the code domain, since several users can transmit information in the same time slot by means of different channelisation codes. In (3GPP, TSG, RAN, 2000b), the DL users either have a spreading factor of 16 with the possibility of multicode transmission or a spreading factor of 1 for high bit rate applications, here the burst type I which have a shorter midamble of length 256 chips and shorter data fields are considered. This has a length of 2560 chips and consists of two data fields, a midamble and guard period (GP), with the midamble being used for both channel equalisation and coherent detection at the receiver. As described above, WWW browsing sessions are generated according to a Poisson distribution. Figure 9.6 illustrates a DL scenario model that describes how the data packets arrive to a user-specific buffer before the air interface. The downlink transmitter model under consideration is depicted in Figure 9.7. Although only the DL model is presented here, it should be noted that this model could be easily generalised for the uplink remembering that in the UL, file size and idle period distributions are different. A
240
Next Generation Mobile Access Technologies
c Fig. 9.6. The downlink packet flow model. 2004 J. Wiley
Fig. 9.7. Transmitter model for a user having different number of codes and TS. c 2004 J. Wiley
base station with a large number of independent users sharing a cell-specific scrambling code is considered. For each user the generated packets should be put into a user-specific buffer before the radio link. The generated packet data from the buffer is channelised depending on the number of codes (KC ) or the number of TSs (KT S ), which are decided by the code and TS allocation algorithm. As mentioned before, the amount of transmitted data for the multislot or the multicode approach in one TS duration (TT S ) is different. Assuming that the bit rate per code slot (one channelisation code with one time slot) for one time slot is fixed, users using the multicode method require
Radio resource metric estimation
241
Table 9.1. Mapping the required data rate onto the resource unit and c RMMF2004 J. Wiley Data rate
Number of TS KT S
Number of codes Kc
Total number of resource units
RMMF lookup reference
24.4 kbps 48.8 kbps
1 1 2 1 2 4 1 2 4 8
1 2 2 4 2 1 8 4 2 1
1 2 2 4 4 4 8 8 8 8
Kc Kc Kc Kc Kc Kc Kc Kc Kc Kc
97.6 kbps 195.2 kbps
=1 =2 =1 =4 =2 =1 =8 =4 =2 =1
transmission with a larger KC than in the multislot operation. In Figure 9.7 the transmitter model is devised for our purpose of supporting different kinds of code and TS allocation schemes, which can be used in the uplink as well as in the downlink. The conventional general multicode transmitter has a serial-to-parallel block so that it could transmit serial data at higher bit rate with several parallel channels at lower bit rate by reducing the bit rate per code slot. It can be assumed without loss of generality that the basic transmit bit rate per code slot is fixed corresponding to its spreading gain of 16 as defined in the TDD-CDMA downlink model. Thus, the multicode transmitter shown in Figure 9.7 uses a fixed spreading gain. This structure model will be used throughout this section. For the TDD-CDMA system model in our study, since we consider the TDD traffic burst type I with a data field length of 244 (Lb = 244 bits) per time slot with one code, the expected data rate during one TDD frame (10ms) is 24.4 kbps and then, the data rate per one resource unit (herein, one code per one time slot unit) is 24.4 kbps. With a similar manner, we can calculate the required resource units according to required data rates in the context of KC and KT S , which is tabulated in Table 9.1. Thus, the achievable transmission data rate can be given by 2.44kbps × KC × KTS . When considering a resource allocation algorithm for the multirate transmission, this table can be used to judge the current resource metrics such as the average BER (P b ), BLER, and the second-order statistics of these according to the measured SIR. With this information, the resource
242
Next Generation Mobile Access Technologies
allocation algorithm can collect the resource status and then assign the available codes or time slots to new users, depending on their fidelity and stability, that are provided by the RMMF. 9.3.4 Example of radio resource metric function Interference characteristics of WWW service Since the downlink could be used as the main transmission link for a Webbrowsing service, the focus is mainly on the downlink interference characteristics. The major link level and the system level simulation parameters assumed in this evaluation are shown in Table 9.2 in which the link level and the system level simulation are performed simultaneously. As listed in Table 9.2, it is assumed that the available multicode set for each multirate implementation scheme is defined. For example, the multislot option can use only one code for each code slot, whereas the multicode option can use from 3 to a maximum of 16 codes per slot. Obviously the allocated number of codes is decided by the code and time slot allocation algorithm according to the size of the generated packet and available codes in the current code pool. The aim of this section is to investigate not only the received interference perceived by a user, but also the interference characteristics depending on code and/or time slot allocation strategies. A comparative study for multirate transmission schemes is performed. As to investigation of a more dynamic signal variation, e.g. for a much higher loaded system, we generated new users’ packets and inserted them into the inter-arrival TS between packet calls as shown in Figures 9.8 and 9.10, which are depicting the interference perceived by a certain user from intra-cell and inter-cell for each RU allocation algorithm. In Figures 9.8 and 9.10, the upper red solid line in each figure shows the variation of the sum of all active users’ allocated codes within a desired cell. Moreover, as to a comparison of the variation of the received interference, it can be observed that the variation level of interference or sum of allocated codes in one time slot is dependent on its RU allocation algorithm. The fewer codes are that assigned, the lower the level of interference that is expected. For a multislot case, the fluctuation of the sum of allocated codes seems to be minimal, but the perceived interference still has a large amount of variation. On the other hand, the multicode method shows a larger variation, since each user has a different number of codes at every time slot. It should be noticed that the standard deviation and the mean of the interference samples for each case are mainly influenced by the RU allocation
Radio resource metric estimation
243
Table 9.2. Radio link simulation parameters. Parameter
Explanation
Link level simulation parameters
(section 9.3.4.1, 9.3.4.2, and 9.3.4.3)
System model Chip rate Processing gain Data rate Doppler frequency Number of TS per DL Standard dev. of lognormal shadowing Channel model
TDD-CDMA system 3.84 Mcps 16 24.4 kbps (constant per code slot) 5 Hz 15 7 dB
System level simulation parameters
(section 9.3.4.3)
Cell radius Number of cell Total number of users per cell Power control Max. BS TX power Max. MS TX power Noise figure Multirate transmission
Rayleigh fading
100m 7 8 Path loss and shadowing compensated 10 dBm 4 dBm 5 dB Multislot, Multicode operations (simple code/TS allocation used based on a first-come first-served rule) Available multicode set Multislot = [0, 1] Multicode = [0, 3, 6, 9, 16]
algorithm. As shown in Figures 9.9 and 9.11, the standard deviation of empirical samples for the multicode scheme is larger than that of multislot option. Furthermore, the empirical CDF for each algorithm matches well to a Gaussian CDF. This result leads to the general conclusion that the distribution of interference characteristics follows a Gaussian distribution irrespective of allocation method in a heavily loaded system. The feature of standard deviation can be regarded as useful information that can tell us the current status of the RU pool, since it indicates what kind of RU allocation algorithms are being used at the current load condition. This will be described in association with the RMMF in the following section.
244
Next Generation Mobile Access Technologies Received interference
50
variation of sum of allocated multicodes within a desired cell 0
dB
50
variation of received interference
100
150
MultiSlot K=8/cell, DL only 200
0
100
200
Doppler=50Hz Shadowing=7dB 300 400 time slot
500
600
700
Fig. 9.8. Interference for multislot scheme excluding inter-arrival TS: Received inc terference and variation of sum of allocated codes. 2004 J. Wiley
Resource metric mapping function example A. Simulation platform for the RRMF In order to build up the RMMF, some parameters listed in Table 9.2 should be changed and modified, since the procedure is based mainly on the link-level simulation. Here we consider the single-cell scenario, since the interference perceived by user is mainly dependent on intra-cell interference. In order to focus on the SIR characteristic corresponding to the allocated codes and to build up the RMMF, we set the number of codes (KC ) used for a multirate transmission fixed. Moreover, it is assumed that all active users use a fixed average number of allocated codes K C such as K C = 1 , 2, 4 and 8 during build up of the RMMF. Thus, we can generate the simulation environment, such that the case of using K C = 1 is the same method as the multislot implementation option and the interference behaviour of K C > 1 is very similar to that of the multicode option. For a simulation scenario involving multiple users, the transmitted power signal level of the desired user or the interference power level must vary at each simulation setup, in such a way that we can simulate the expected SIR values at each SIR point and, thus, get the link-level results throughout the wide range of SIR. So, the number of total users (herein we set up K = 3)
Radio resource metric estimation
245
Empirical CDF 1 0.9 0.8 0.7
F(x)
0.6 0.5
MultiSlot Ku=8/cell, DL only Doppler=50Hz Shadowing=7dB
Empirical Gaussian
empirical mean= 92.7 dB empirical std. = 2.2 dB Gaussian ~( 92.7, 2.2)
0.4 0.3 0.2 0.1 0 110
105
100
95 90 x: Interference (dB)
85
80
Fig. 9.9. Interference for multislot scheme excluding inter-arrival TS: CDF of inc terference. 2004 J. Wiley
in (9.1) is not of concern since the SIR is controlled by the desired user’s power over the total sum of interfering users’ powers, which means that the actual number of interfering users is not a controlling parameter. Here the number of burst SIR sequences is considered as 125, which corresponds to the approximate length of a packet call in a Web-browsing service. Under these simulation assumptions, the SIR for the ith user can be rewritten as: i KCi Prx , SIRi = K j i )+N ( i=j=1 KCj Prx − KC Prx
(9.9)
i is the received where KCi is the number of codes for the ith user and Prx power from the ith user including fading and multipath effects. This simulation set-up is used throughout the section 9.3.4.2.
B. Mean and standard deviation of burst SIR versus average raw BER The SIR for each burst is first measured in the simulator, and then this is mapped to a raw BER using a relationship of rawBER = f unction (SIRburst ). This mapping procedure was described in section 9.3.2.1. Each mean of burst SIR values results in a certain average raw BER P b that
246
Next Generation Mobile Access Technologies Received interference
50
variation of sum of allocated multicodes within a desired cell 0
variation of received interference
dB
50
100
150
MultiCode Ku=8/cell, DL only 200
0
100
200
Doppler=50Hz Shadowing=7dB 300 400 time slot
500
600
700
Fig. 9.10. Interference for multicode scheme excluding inter-arrival TS: Received c interference and variation of sum of allocated codes. 2004 J. Wiley
can be extracted by averaging several link-level simulation results for instantaneous raw BER versus burst SIR value. Multipath fading effects are relatively small since it is assumed that the fading environment is slow, such as in a pedestrian channel model. For this reason, the interesting parameter here is the average number of allocated codes. For multirate transmission, the BS allocates the required codes and slots to active users according to their requests and channel availability. In this study, however, the BS simply assigns codes and slots based on a first-come-first-served basis. If the required number codes in the same time slot are over the maximum available number of code, this request is rejected and delayed to the next usable resource. Figure 9.12 shows the average raw BER P b curve as a function of the mean of burst SIR and of the average number of allocated codes. It is an obvious point in Figure 9.12 that the average number of allocated codes has a very large impact on the result. However, we should note that the K C = 8 case has eight times more interferers than the K C = 1 case for a fixed number of active users, since in the multicode transmission, the actual number of simultaneous transmission is K C times larger than single-code
Radio resource metric estimation
247
Empirical CDF 1 0.9 0.8 0.7
F(x)
0.6
MultiCode Ku=8/cell, DL only Doppler=50Hz Shadowing=7dB empirical mean= 89.7 dB empirical std. = 10.7 dB
0.5
Gaussian ~( 89.7, 10.7) 0.4 0.3
Gaussian ~( 90, 12)
0.2 0.1 0 130
120
110
100 90 80 x: Interference (dB)
70
60
50
Fig. 9.11. Interference for multicode scheme excluding inter-arrival TS: CDF of c interference. 2004 J. Wiley
transmission. This relationship will be discussed in the following subsection in terms of throughput comparisons. The impact of the allocated number of codes on the standard deviation of SIR is quite serious as shown in Figure 9.12. It indicates that the standard deviation of burst SIR values is likely to strongly depend on the average number of allocated codes. In other words, hidden characteristics such as the standard deviation of measured burst SIR as well as the mean of SIR can be useful information for higher-layer resource mapping to implicitly determine what kind of interference exists at the current channel load and to select a dynamic call admission control algorithm, reflecting actual link quality variation. Note that the standard deviations of multirate transmission strategies are mainly governed by the allocated number of codes. The standard deviation of K C = 1 outperforms the others, which resembles the behaviour of the circuit switched mode, i.e. the conventional voice oriented transmission. Moreover, the average raw BER increases abruptly and becomes saturated as both the mean of the burst SIR value and K C increase. This result indicates that the standard deviation has a straightforward effect on the BER performance. The average raw BER as a function of mean
248
Next Generation Mobile Access Technologies 0
10
1
Average raw BER
10
10
2
KC (std. dev. of SIR [dB]) KC =1 (0.44 dB) KC =2 (2.88 dB) KC =4 (3.12 dB) KC =8 (3.09 dB)
3
10 20
15
10
5
0
5
10
15
Mean of burst SIR [dB]
Fig. 9.12. Average raw BER performance for the TDD-CDMA system according c to its average number of allocated codes, Kc . 2004 J. Wiley
and standard deviation of burst SIR is shown in Figure 9.13. It should be noticed that the average raw BER is influenced not only by the mean burst quality but also the standard deviation. This 3-D figure presents a RMMF, in that it demonstrates the actual link quality according to SIR values representing channel load and standard deviation implying code availability at a current resource pool. Since usually the conventionally measured average SIR sample does not fully reflect the perceived link quality for users with such dynamic interference characteristics, a certain average SIR may result in very different user link quality. However, by using this kind of resource metric mapping function, an accurate link quality in terms of average raw BER observed by a user can be guaranteed and, thus, leads to an accurate interface between link level and system level. C. Standard deviation of BER and SIR as a function of allocated codes Compared to the GSM interface results reported in (Oloffson et al., 1997), the study here obtains a different result as a function of mean and standard deviation of raw BER. Although we do not include a comprehen-
Radio resource metric estimation
249
0
Average raw BER
10
10
10
1
2
4 10 30
3
3 20
2 10
0 Mean of burst SIR (dB)
10
20
0
1 Standard deviation of burst SIR (dB)
c Fig. 9.13. RMMF as a function of mean and standard deviation of burst SIR.2004 J. Wiley
sive comparison, it can be clearly observed that the effect of the standard deviation of the raw BER on the average BER is not as serious as in (Oloffson et al., 1997). This can be illustrated in Figure 9.14 where the standard deviation of BER is shown as a function of average number of codes for desired user (KC,des ) and for interferers (KC,I ). Moreover, the standard deviation of SIR as a function of average number of codes for desired user and for interferers is illustrated in Figure 9.15, which demonstrates that the standard deviation of SIR could be influenced by KC,I , especially in a lightly loaded system. This leads to the conclusion that the main decision parameter is not KC,I but KC,des . However, since the standard deviation graph of the BER for the number of codes shown in Figure 9.14 is not so discriminatory compared to that of the SIR, it is possible to draw the conclusion that the standard deviation of the SIR graph is more favourable in building up a resource mapping function rather than that of the BER plot. This implies a similar conclusion for the multi-cell scenario could be derived, since the interference from other cells is likely to be of a Gaussian distribution with a
250
Next Generation Mobile Access Technologies
Standard deviation of BER
0.12 0.1 0.08 0.06 0.04 0.02 8 8 4 Average no. of codes for desired user, Kcdes
2
1
1
2
4 Average no. of codes for interferers, KcI
Fig. 9.14. Standard deviation of BER as a function of the average number of codes c for desired user (KC ,des ) and for interferers (KI ). 2004 J. Wiley
different standard deviation associated with different multirate transmission schemes, i.e. different number of codes allocated for users.
D. BLER mapping function results The BLER mapping function based on the estimated average, BER P bk , and (9.6) and (9.7) can be depicted as in Figure 9.16, which eventually build up the resource metric mapping function of BLER from the average SIR bursts and the raw average BER. With this BLER mapping function along with the RMMF function in Figure 9.13, the base station can now utilise more link quality information since this can deliver the effect of an error-correcting code. Although the standard deviation impact of SIR on the performance of K C transmission is not shown, the hidden characteristics are contained in the RMMF and, thus, are still useful for delivering actual link quality to a resource allocation algorithm. Therefore, by using these two resource metric mapping functions, the link quality in terms of average raw BER and BLER can be determined very accurately and an accurate interface between link level and system level can
Radio resource metric estimation
251
Standard deviation of SIR (dB)
4
3
2
1
0 8 8
4 Average no. of codes for desired user, Kcdes
4 2
1
1
2
Average no. of codes for interferers, KcI
Fig. 9.15. Standard deviation of SIR as a function of the average number of codes c for desired user (KC ,des ) and for interferers (KI ). 2004 J. Wiley
be achieved. In a higher layer, these mapping functions can be used as a resource look-up table having actual radio mobile channel and interference characteristics. User data throughput and system throughput comparisons of multirate transmission A. User data throughput comparisons versus average SIR In our study, the basic code branch rate is set to 24.4 kbps since the data amount of one time code slot is fixed by using a TDD type II data traffic burst. So, if the data rate of users Rb in Eq.9.8 can be rewritten as KC R0 , where R0 is the basic code branch rate of multiple code branches, we can rewrite the user data throughput as: S = KC R0 PC (ˆ μkSIR ) = KC R0 (1 − BLER),
(9.10)
where k is the expected user data throughput of successfully received bits per second. Note that the achievable transmission data rate can be given by
252
Next Generation Mobile Access Technologies 1
0.9
0.8
0.7
BLER
0.6
0.5
0.4
0.3
0.2
0.1
5
KC =1 KC =2 KC =4 K =8 C
0
5 Average SIR [dB]
10
15
Fig. 9.16. BLER versus SIR as a function of number of codes: a 1/3 rate convoluc tional error-correcting code for Lb = 244 bits. 2004 J. Wiley
KC R0 KT S , but this reduces to KC R0 since we consider the signal behaviour within one time slot in our simulation model. The user data throughput as a function of average SIR is depicted at Figure 9.17, which can be obtained by using the BLER results in Figure 9.16 and (9.10) according to the number of allocated codes within a time slot. Thus, this can lead to a different insight into the impact of multirate transmission operations on the data throughput of users who are using a different number of codes. Note that this graph can be used as a reference of expected throughput performance for each user by utilising the link-level results such as measured SIR burst, mean and standard deviation of SIR, average raw BER and BLER into the system-level throughput performance. As shown in Figure 9.17, it is observed that the user data throughput is influenced by K C and the average SIR. It is shown that the user data throughput of K C = 1 outperform at a low SIR region, while both of K C = 4 and K C = 8 can achieve much higher user data throughput in a higher SIR region. This result indicates that a larger number of codes can deliver more
Radio resource metric estimation
253
Fig. 9.17. User data throughput as a function of average SIR with a 1/3 convoluc tional error-correcting code for 244 bits. 2004 J. Wiley
data information bits within the same transmission time, providing a high enough SIR to combat the large number of interferers which arises when using multiple codes. With this user data throughput graph along with the SIR-to-BER RMMF and the SIR-to-BLER mapping function reflecting actual link quality information, the system can take the utilisation efficiency of RU into consideration. B. System data throughput comparisons Besides the user data throughput performance, the total system throughput is another interesting performance metric in the context of multirate transmission operations, since it is an important feature to check the system capability of transmission depending on its traffic load condition (herein the number of arriving request code slots). When we take into account the user arrival model, the signal behaviour within a time slot of a TDD-CDMA system where multiple
254
Next Generation Mobile Access Technologies
users’ signals conglomerated within the same time slot with different allocated codes is quite similar to that of a multicode CDMA slotted-ALOHA system (Sandouk et al., 1999) since every arriving users of both systems are trying to use the limited code slots (or packets in (Sandouk et al., 1999)) with a Poisson arrival distribution. Note that a CDMA slotted-ALOHA system features CDMA techniques combined with slotted-ALOHA properties (Makrakis and Murthy, 1992). Thus, we can assume that the user signal behaviour within a specific time slot resembles that of a CDMA slottedALOHA system, and thus we can use this model for a throughput performance comparison of different multirate transmission options on a time slot basis from the aspect of code utilisation. Based on this assumption, the equations introduced to analyse the multicode CDMA model using slottedALOHA in (Sandouk et al., 1999) can be used in our study. To show this, Figure 9.18 illustrates an example of simultaneous transmission for single (or multi)-code operation in both systems. As shown in Figure 9.18, the predetermined KC code slots are supposed to be assigned to the requesting user, each of which has Lb = 244 length-bits within one time slot duration (TT S ). If the number of simultaneously generated packets is assumed as the steady-state probability of a Poisson process, the probability that total code time slots are generated within one time slot can be given by (Sandouk et al., 1999): (G/KC )k exp(−G/KC ), (9.11) k! where k is the arrived user within a time slot and G is the offered load defined as the average number of transmitted code slots. So, if k users attempt to transmit their time slots, a total of KC k code slots are generated to use resource units within the same time slot. With some modification of the multicode throughput equation given in (Sandouk et al., 1999), we can define the average system data throughput of successfully received code time slots per time slot as follows: Pt (k, G) =
μkSI R ) = KC R0 (1 − BLER), S = KC R0 PC (ˆ
(9.12)
where K is the maximum number of users that the system can handle and PC (k) is the probability of receiving correctable code slots with error-control coding as in (9.7). The simulation model we set up is to generate the arriving code slots model with the steady-state probability Pt (k, G) as shown in Figure 9.18 and calculate the throughput at every generated code slot G and every SIR level. The system data throughput can be obtained by
Radio resource metric estimation
255
applying the average BER and RMMF from link-level simulation under a fading channel environment as listed in Table 9.2 to be used for calculating PC (k). Figure 9.19 shows the system throughput comparison curves as a function of the mean SIR, the offered average code slots, and the average number of allocated codes K C for a processing gain (pg) of 16. From these four figures, it is noticeable that the total achievable data throughput is obviously influenced by the mean of SIR and the choice of multicode option (or multirate transmission option). In particular, the system throughput performance is dependent on the offered code slots, i.e. the arrived traffic load condition. For the cases of K C = 1 and K C = 2 , the system throughput can achieve relatively good performance in a lightly loaded condition, on the other hand, in a heavy offered load the throughput performance of K C = 4 and K C = 8 outperforms counterparts. Thus, we can draw the conclusion that the multi code transmission operation is preferable in a heavily loaded system because of its capability for simultaneous multiple transmissions using multiple parallel code channels. Moreover, in a lightly loaded system, the multislot operation is preferable by virtue of its agile transmission capability. When we select a higher number of multicodes K C , a greater number of code slots can be sent at a time in one time slot duration, so that the number of succeeded code slots increases which, in turn, results in throughput improvement. Note that the conventional call admission control is likely to use the average link quality characteristics such as average BER and average throughput irrespective of RU allocation options. For this reason the resource metric mapping function corresponding to its user and system throughput could provide very useful information for a higher layer, since it can deliver the current channel load condition using the mean SIR, the RU availability, or the efficiency calculated from the standard deviation of SIR and throughput associated with the RU allocation algorithm. For example, we can inform a higher layer of the measured mean and standard deviation of burst SIR, the mapped average raw BER and the RU (in particular code) availability based on these measurements. The exact RU measurement and estimation algorithm combined with call admission control and predicted throughput is an area of further investigation.
9.4 Radio resource metric region The question of how to combine the interference measurements with the current load situation and QoS requirements of the existing traffic classes to control CAC, or channel allocation is a very interesting issue (Jorguseski
256
Next Generation Mobile Access Technologies
Fig. 9.18. Illustration of simultaneous transmission of multicode slots with Poisson process for both a TDD-CDMA system and a CDMA slotted-ALOHA system. c 2004 J. Wiley
et al., 2001). Thus, the most appropriate resource metrics permit efficient inter-working between the physical layer and higher layers in the protocol stack and thus, it is essential to optimise the overall system performance. In this section, as a solution for efficient cross-layer inter-working, the radio resource metric region (RMR) concept has been introduced, which can provide the acceptable resource region where QoS and acceptable link quality can be guaranteed with an achievable resource margin to be utilised in terms of the resource metric mapping function (RMMF).
9.4.1 Power controlled RRA for multimedia CDMA system RRA for W-CDMA CAC criterion The maximum achievable capacity can be achieved by following the call admission control (CAC) criterion in (Sampath et al., 1995), and (Gurbuz and Owen, 2002), assuming the existence of the optimal power level for each user:
f (ε, G, h) =
K i=1
εi η0 W + < 1, i εi + Gi min1≤i≤K [Pmax,i hi εi +G εi ]
(9.13)
Radio resource metric estimation
257
Fig. 9.19. System throughput as functions of mean of SIR and generated average code slots (G) according to average number of allocated codes KC ; KC = 1 (left upper), KC = 2 (right upper), KC = 4 (left bottom) and KC = 8 (right bottom); the red solid line indicates the maximum capacity that the system can handle. c 2004 J. Wiley
where W is the system bandwidth, Gi is the processing gain, hi is the channel coefficient, and η0 is the noise spectral density. K is the total number of mobile user admitted in the system, which is the total sum of each users supporting corresponding data rate classes K = N i=1 Ki , where N is the total data rate services classes. In our study, we consider three kinds of service classes, voice, video, and data, of which number of users are denoted as Kv , Kvid , Kd , respectively. εi is the required Eb /I0 for the ith user, which is defined by
εi = K
j=i
Gi hi Pi hj Pj + η0 W
,
where Pi is the transmitted power for the ith user.
(9.14)
258
Next Generation Mobile Access Technologies
c Fig. 9.20. Block structure of (MC)2 -CDMA system.2004 J. Wiley
RRA for multi carrier CDMA A. Single-code multi carrier CDMA system The single-code MCCDMA system is simply a special case of the multicode multicarrier ((MC)2 )CDMA system in that it uses only one code branch, i.e. one code serial-toparallel (S/P) branch. Thus, the data coming in is of a rate N R bps using N rate S/P channels rather than M N R bps using code S/P branches as in the (MC)2 -CDMA system. A single-code MC-CDMA system block consists of rate S/P channels of rate R, each of these is spread by a spreading code of length and passed to a N G-point IDFT. In a SC MC-CDMA system, the Eb /I0 for the ith the desired user and the nth rate S/P channel is given as follows:
εi,n = K
Gi,n hi,n Pi,n
j=i
hj,n Pj,n + η0 W
.
(9.15)
Assuming that Eb /I0 in a MC-CDMA system is the averaged value cross all sub-carriers, this equation can be reduced to εi similar to that for a conventional CDMA system as in (9.14). Therefore, this system has the same CAC criterion as a W-CDMA system.
Radio resource metric estimation
259
B. Multi-code multi carrier ((MC)2 )-CDMA system The processing gain G for each code remains the same, and the supported data rate for each user is decided by the number of multi-codes and rate channels, i.e. a rate of M N R bps. Assuming that for a multicode system, the received power for each of the M codes (of the same multicode user) is the same and the multiuser interference experienced by each of the M codes is also the same (3GPP, TSG, Services and System Aspects, 2004), the hi,m,n and Pi,m,n for each code channel are the same and thus, reduced to hi,n and Pi,n , respectively. The Eb /I0 for the nth rate S/P channel of the ith user in a (MC)2 -CDMA system can be given by εi,n = K
j=i
Gi,n hi,n Pi,n hj,n Pj,n + (Mi − 1)Pi,n + η0 W
,
(9.16)
where Mj is the number of codes assigned to the jth user. The denominator includes the multicode interference by other users and the self-interference caused by imperfect orthogonality. Thus, the overall Eb /I0 for the ith user can be obtained by Ni 1 εi = εi,n , Ni
(9.17)
n=1
where Ni is the number of rate S/P channels assigned to the ith user. Now, the CAC criterion for (MC)2 -CDMA system with a maximum of K users is as follows: K M i εi η0 W < 1. + i +G i εi + Gi min1≤i≤K [Pmax,i hi εM ] i=1 i εi
(9.18)
The maximum number of supportable user K is determined by the predefined outage probability Pout , which is set to as 0.01 in our study. 9.4.2 Capacity plane and ADR with systematic error As shown Figure 9.21, the theoretical maximum available number of user pairs (Kv , Kd ) with different data rates and QoS requirements for a given number of video users is significantly dependent on the variance of Eb /I0 , which is caused by systematic errors such as inaccurate power control, traffic variation and fading. Figures 9.21 and 9.22 show the maximum aggregated data rate (ADR) defined by the total sum of the number of users multiplied by the corresponding data rate, i.e. ADR = (Kv Rv + Kvid Rvid + Kd Rd ).
260
Next Generation Mobile Access Technologies 7
εσ = 0 dB
W = 3.84 MHz [Rvoice, Rdata, Rvideo] = [8, 128, 64] Kbps
εσ = 1 dB
εmean = [5, 7, 5] dB, Kvideo = 0
6
εσ = 2 dB εσ = 3 dB εσ= 4 dB
Number of data users with Rdata
5
C: Non-feasible region
4
3 B: Negotiable region 2
1 A: Feasible region 0
0
20
40
60
80
100
120
140
Number of voice users with Rvoice
Fig. 9.21. Maximum capacity plane for a fixed number of video user with different systematic error.
If the systematic error is increased (here, the standard deviation of Eb /I0 ), with the admission control criterion as in (9.13), the maximum available number of users and the ADR are dramatically reduced, as shown in Figures 9.21 and 9.22. Thus, it is necessary to consider the variance of the received Eb /I0 along with the possible maximum capacity margin according to its situation. When the number of users satisfying the QoS requirement is determined based on the measured and estimated load, this can be categorised as: (A) feasible region, (B) negotiable region, and (C) non-feasible region, depending on current load status and the amount of capacity margin. The capacity margin in region (B) shown in Figures 9.21 and 9.22 is too stringent to admit or reject a call in a conventional system, although it has still available resources. Thus, in the following section we describe the proposed predicted RMR-based RME, which is dependent on the standard deviation of capacity and the degree of confidence (DCL) of capacity and user status that can be driven from the second-order statistics of Eb /I0 . Note that the higher the
Radio resource metric estimation 105
18 16
σ εv,d,vid = 0 dB
W = 3.84 MHz, [Rv, Rd, Rvid] = [8, 128, 64] Kbps
σ εv,d,vid = 1 dB
, εdreq, εreq = [5, 7, 5] dB εreq v vid
σ εv,d,vid = 2 dB
14 Maximum Aggregated Data Rate (ADR)
261
C: Non-feasible region
σ εv,d,vid = 3 dB σ εv,d,vid = 4 dB
12
10 8 6
B: Regotiable region
4 A: Feasible region 2 0 0
50
100
150
Number of voice users, Kv
Fig. 9.22. Maximum ADR as a function of number of active voice users for a given averaged standard deviation of required Eb /I0 targets.
variance of the Eb /I0 , the lower the maximum capacity, capacity margin and resource availability. For an estimation of the maximum available number of users and available resource margin, both the received Eb /I0 and the variance of Eb /I0 have to be taken into account. 9.4.3 Predictive load-based radio resource metric region In an aggregated traffic stream the estimation of available resource can be either optimistic or conservative due to the inaccurate link quality information and the coarse estimation of overflow traffic, and thus all the resource units cannot be exploited. If we know the resource availability, i.e. the required total average resource plus the excessive resource that cannot be utilised because of stringent call admission criterion, then this information allows the acceptable resource metric region to be established on a call, a packet, or a time slot basis, thus enabling the resource allocation algorithm to maximise the resource utilisation. Our approach to this concept is depicted in Figure 9.24, where the measurement of SIR and traffic followed by a Kalman filter-based prediction is
262
Next Generation Mobile Access Technologies W=3.84MHz R1, 2,3 = [8, 128, 64] Kbps ,ε =[5 7 5]dB Eb/Io target
target
6
Maximum Aggregated Data Rate (bps )
x 10 1.3 1.2 1.1 1 0.9
0
0.8 0
1 1 2
2 3
3 Std. dev. of (Eb/Io )
[dB], ε
video
3
4
4
Std. dev. of (Eb/Io ) voice[dB], ε1
Fig. 9.23. Maximum ADR as a function of standard deviations of required Eb /I0 targets for different data rate services, εσv and εσvid .
Fig. 9.24. Block diagram of the predicted RMR-based resource metric estimation c concept 2004 J. Wiley.
Radio resource metric estimation
263
performed. This provides measured or predicted resource parameters for the resource scheduler to search their current position on the surface of the RMMF and to deliver the actual status of the resource usage. All of these are integrated by establishing RMR in which the calculation and estimation of resource availability are performed based on the resource region decision criteria and help the call admission decision on a call, packet, or time slot basis (Lee et al., 2003; Lee and McLaughlin, 2004). Degree of confidence level As mentioned before, the capacity margin in the negotiable region is too stringent to admit or reject a call in a conventional system, even though it still has available resources. Therefore, we have proposed a predicted RMRbased RME with the DCL of capacity and user status that can deliver the available capacity or resource margin, considering the systematic error. The procedure of the proposed predictive RME method is briefly summarised as follows: (i) BS measures the mean and standard deviation of received Eb /I0 . (ii) BS estimates the current load condition, interference level and satisfied users. (iii) Then BS decides the current capacity region. (iv) With the Kalman predictor, BS predicts capacity margin. The predicted capacity margin or variation (ΔK) can be rewritten as ˆ G, ˆ c) from (9.13) with some modifications, which takes Δf (ˆ ε, εˆσ , h, the current and predicted standard deviation of Eb /I0 into account. ˆ is given Therefore, the final predicted available number of users K, by the sum of the estimated number, K, and its variance, ΔK, i.e. ˆ =K ˆ + ΔK. ˆ K (v) DCL calculates the weighting factor considering the current and predicted capacity margin and status of users belonging to one of all possible regions and delivers these to the RRA algorithm to accept or reject an incoming call. DCL is the main feature of proposed RME. It is defined as the capacity status by comparing the distance from the optimal capacity limit (OCL) and the pessimistic capacity limit (PCL) to estimated current user status. The amount of the predicted variation of available number of users is either increased or decreased by the Kalman predictor (ΔK). This means that it is mainly dependent on the current and predicted capacity margin and status of users belonging to one of all possible states, e.g. current state → predicted state ((A, B, or C) → (A, B, or C)). Depending on the status of the
264
Next Generation Mobile Access Technologies Degree of Confidence Level (DCL)
Region C
25
20
0.2 0.1 0.3 0.4
0.2 0.1 0.3 0.4
0.20.1 0.3 0.4
0.20.1 0.3 0.4
0.5
0.6
0 0.8 .7
0.3
0.5
0.4
Region B
10
15
0.3
0.5
0.7 5
0.9 0.8
Region A 0
0.6
7
1.0
0.6
Region A
5
0
0.2 0.4
9
0.
0.
0.9 0.8
0.9
0.1
0.2 0.3
0.8
0.8
0.6
0.8
0.7 0.9
0.5
0.7
0.6 10
0.6
0.4
Region B
0.5
0.6 7 0.
Predicted Capacity (Kpr)
0.5
0.1 0.2 0.3
0.5 15
0.20.1 0.3 0.4
Region C
20
25
Measured Capacity (K ) m
Fig. 9.25. Degree of confidence level example.
current user’s capacity region, the corresponding DCL can be calculated in quite a different manner, as shown in Figure 9.25. This DCL example applicable to wireless CDMA systems is dependent on system assumptions such as the maximum available negotiable region, the maximum capacity limit corresponding to system parameters, and the fidelity of estimator and predictor, as well as service traffic characteristics. From the QoS point of view, the DCL can provide a decision parameter with which adaptive QoS-based admission control can perform taking capacity and resource availability into account in a predictive manner. 9.4.4 Multimedia traffic simulation model A single micro-cell configuration is considered next, in which a set of powercontrolled mobile terminals transmit packets to a BS in a CDMA system. With uplink channel, RTT (voice call and video-phone calls) and NRTT mixed traffic are considered. We generated the video-phone traffic model as in (3GPP, TSG, Services and System Aspects, 2004) for an H.263 video sequence, in which distribution is assumed to be Gamma distribution. The
Radio resource metric estimation
265
call arriving from each service is modelled as Poisson random variables with corresponding exponential distributed holding time. For data traffic, packetswitched Web-browsing traffic is considered. The system level multimedia traffic parameters are shown in Table 9.3. In order to fully utilise available resource (or residual capacity), the monitoring RTT load and controlling NRTT packet data traffic is crucial in multimedia traffic call admission control. Since RTT calls are always given the highest priority and are allowed to transmit without delay, in our study NRTT calls are allowed to transmit according to the available residual capacity (here the residual aggregated data rate (ADR)) obtained by subtracting the RTT contribution from the total system capacity. Therefore, the resource availability for data service is determined by estimating and predicting the RTT load contribution for the next time state by Kalman filtering and the predictive load-based RME (Lee et al., 2003). This predictive RME method can deliver the fidelity information of resource availability of the current and the next time-stage to the resource allocation algorithm, taking the variance of Eb /I0 into account. 9.4.5 Examples of radio resource metric region Simulated and admission controlled multimedia traffic for both W-CDMA and (MC)2 -CDMA systems is shown in Figure 9.26, where the systematic errors, the standard deviations of the received Eb /I0 targets (εσv = εσd = εσvid ), are all assumed to be 2 dB. This shows the achievable ADRs for each system controlled by the CAC criterion as in (9.13) and (9.18), respectively, with parameters given in Table 9.3. In Figure 9.27, the DCL and admissible residual ADR with/without DCL for each system are shown in terms of frame index. It is demonstrated that the DCL of (MC)2 -CDMA is relatively higher than that of W-CDMA, which means that for the (MC)2 -CDMA system, the fidelity of information of the capacity status is more reliable than W-CDMA since long-term fading is assumed in our study. Thus, compared to the optimistically estimated residual ADR (Opt. resADR in Figure 9.27), the residual ADR with DCL weighting given by the predictive RME gives almost the same ADR, unlike W-CDMA. With the same parameters, Figures 9.28 and 9.29 show the admissible residual capacity plane with the predictive RME for W-CDMA and (MC)2 CDMA, respectively, for a multimedia traffic scenario. Here the residual ADR for NRTT is estimated and predicted by Kalman filtering after measuring the current ADR for RTT, and this is converted into available data users versus voice or video users. As shown in Figures 9.28 and 9.29, the admissible residual capacity region is quite dependent on the selection of
266
Next Generation Mobile Access Technologies
Table 9.3. Multimedia Traffic Simulation Parameters. Common parameters
Value
System bandwidth Noise spectral density, η0 Max. MS power limit (voice, data, video) Standard dev. of shadowing Mean of required Eb /I0 for voice, data, video: εrv eq , εrdeq , εrv eq id Data rate for voice, data, video: Rv , Rd , Rv id Outage probability constraint, Pou t Parameters CDMA
for
(MC)2 -
System bandwidth No. of rate S/P channels, N No. of code S/P channels for voice, data, video (M ) Spreading gain (G) Total number of sub-carriers (IDFT point) Basic data rate of rate S/P channel (R)
3.84 MHz -174 dBm 5, 17.1, 14.1 dBm 10 dB 5, 7, 5 dB 8, 128, 64 kbps 0.01 Value 4.096 MHz 2 1, 16, 8 512 dB 1024 4 kbps
system, i.e. the choice of resource dimension, which in turn delivers the exact capacity status to CAC algorithm and resource allocation algorithm to control the packet service data rate in a dynamic fashion. The adaptive rate control mechanism incorporated with the admissible residual capacity will be our future work. Compared to Figure 9.28, it is noticeable that Fig. 9.29 can achieve a larger admissible residual capacity area for NRTT users, which of course depending on the fidelity of between the estimated and predicted residual capacity and the selection of system with different resource dimension. For example a (MC)2 -CDMA system has a three-dimensional resource plane such as code, frequency and spreading gain, even though this could be constrained by the channel fading and the receiver structure. With these results, the resource allocation with the predictive RME can deliver the exact status of residual ADR and admissible capacity region, which in turn can decide appropriate allowable rates for each data user. Thus, the full utilisation of the remaining resource, i.e. the residual ADR, can be
Radio resource metric estimation (a) Arriving multimedia traffic
6
ADR [bps]
3
267
x 10
2 1 Ttoal traffic Voice traffic Data traffic Video traffic
0 5
ADR [bps]
15
(b) Admission controlled multimedia traffic (WCDMA)
x 10
10 5 0 (c) Admission controlled multimedia traffic ((MC)2 CDMA)
5
ADR [bps]
15
x 10
10 5 0
0
50
100
150
200
250
300
Frame index
Fig. 9.26. Simulated and admission controlled multimedia traffic for W-CDMA and (MC)2 -CDMA system; λv , λd , λv id = [3, 2, 1] call/frame with 2 dB standard deviation of required Eb /I0 target.
achieved in a dynamic and predictive manner. Furthermore, if combined with adaptive modulation, or rate control, the proposed resource allocation method is able to provide the exact resource availability. 9.5 Conclusions In the first part of this chapter, an efficient approach of how to combine the interference measurements relevant to current load and traffic condition in a call admission control was proposed as a resource metric mapping function (RMMF) applicable to a TDD-CDMA system supporting WWW traffic. In this function, the actual measurements of link quality estimation are taken into account in accordance with multirate transmission methods, since the actual interference level during the given time slot is mainly dependent on the resource unit (RU) allocation algorithm. Furthermore, it was demonstrated that the CDF of the interference follows a Gaussian distribution for all cases and the standard deviation of interference is mainly dependent on its allocation choice. These results allowed the resource metric mapping function to be defined and established by comprehensive link-level
268
Next Generation Mobile Access Technologies
Fig. 9.27. DCL and admissible residual ADR for W-CDMA and (MC)2 -CDMA systems.
simulations. Information from this function can be used for a higher layer to implicitly determine the current channel load by monitoring the mean of SIR bursts and to select a dynamic call admission control scheme by estimating the standard deviation of SIR bursts. From the BER point of view, the multislot scheme (K C = 1) outperforms the multicode scheme (K C > 1). However, from the throughput performance point of view the multislot cannot always be guaranteed. From this result, the throughput associated with allocated codes should be taken into consideration along with the RMMF, since the throughput indicates the code resource availability and efficiency. By using this kind of mapping function, a higher layer’s work such as RRA and resource management can reflect current interference and actual radio mobile channel characteristics. In future work, the objective function to optimise resource unit (code and slot) availability should be built up incorporating the proposed RMMF. In order to achieve this purpose, the proposed RMMF should be considered in more dynamic and on-line manners (Lee et al., 2003)(Lee et al., 2003). In the second part of this chapter, the resource allocation with an RME
Radio resource metric estimation
269
Fig. 9.28. Admissible residual capacity plane with RME for residual ADR in a W-CDMA aggregated multimedia traffic.
method based on predicted load was investigated for multimedia CDMA systems such as W-CDMA and (MC)2 -CDMA. With aid of this RME method, the acceptable residual resource region can be achieved with QoS and acceptable link quality for NRTT data users guaranteed with an achievable resource margin in terms of capacity margin, the DCL of the system, and the second-order statistics of Eb /I0 . With the DCL-weighted admissible residual capacity plane, the (MC)2 -CDMA system with more flexibility of resource dimension delivered more reliable and exact admissible residual ADR in our study, compared to the W-CDMA system. This result is reversed if a more severe frequency-selective fading is used. However, with an adaptive control of spreading gain corresponding to its channel characteristics, such as coherence bandwidth, the severe frequency-selective fading would not be a major drawback.
270
Next Generation Mobile Access Technologies
Fig. 9.29. Admissible residual capacity plane with RME for residual ADR in a (MC)2 -CDMA aggregated multimedia traffic.
10 Interference-based cancellation techniques for TDD Stamatis Georgoulis Stephen McLaughlin David Cruickshank
10.1 Introduction The objective of this chapter is to consider techniques that remove multiuser detection algorithms from the mobile terminal while maintaining satisfactory performance. The motivation for this is the significantly reduced complexity of the receiver required in the mobile terminal. The chapter will initially discuss the motivation for this and the special conditions that make this idea realistic in a W-CDMA-TDD system. Section 10.3 presents the precoding concept and provides an analytical framework for the calculation of its bit error rate performance. Section 10.4 then classifies precoding techniques into two categories and some power issues are introduced in section 10.5.
10.2 Motivation The linearity of a conventional CDMA system, presented simplified in Figure 10.1, suggests that it might be possible to circumvent the complexity implications of using a multiuser detection technique. The aims therefore
d
Spreading
s Channel 1
Channel K
Linear MUD 1 c 1...K , h 1
d1
Linear MUD K c 1...K , h K
dK
Fig. 10.1. Linearity of CDMA downlink.
271
272
Next Generation Mobile Access Technologies
are to transfer the complexity and computational load to the transmitter (BS) where weight and size are of less concern and resources (such as power and space for hardware) are readily available while maintaining at least the same performance that receiver-based MUD techniques would provide. The MS is restricted to the knowledge of its own signature waveform. Therefore, the k-subscriber’s receiver shall consist of a simple correlator or a filter matched to the spreading sequence, ck , which is ideal for a single-user in an additive white Gaussian noise (AWGN) channel where no channel estimation, adaptive equalisation or feedback from the BS is required. In addition, the overall capacity of the downlink in system is increased as no training sequences need be transmitted. h1
d
hK
precoding & spreading
c1
cK
sp Channel 1
c1 (Qn)
d1
Channel K
cK(Qn)
dK
Fig. 10.2. Linear precoding.
The manner in which this problem is initially addressed is to apply a linear transformation matrix, T , to the transmitted vector, d, (the data spreading is included in the T matrix), in order to eliminate intersymbol interference (ISI) and multiple access interference (MAI) before transmission. Under this scheme, the actual transmitted signal is distorted before transmission, in such a way that after passing through the radio channel and mobile receiver it will yield the desired data. Each MS needs knowledge of the downlink channel impulse response, hk (n), and the spreading codes of every user to apply a joint multiuser detection. If we take into account all mobile receivers, then the total information required is all K downlink channels and signature waveforms (hk , ck , k = 1, . . . , K). The transformation matrix T should be a function of the same knowledge in linear precoding. In other words, to achieve good performance the transmitted signals in the downlink must be jointly optimised based on the spreading code and the downlink channel impulse response of every user. The BS station obviously knows the spreading codes of all the active users in a cell and all it needs is the downlink channel impulse responses. The uplink channels
Interference-based cancellation techniques for TDD
273
can be estimated by the BS and exploit the fact that it is reasonable to assume channel reciprocity for 3GPP W-CDMA-TDD, provided that the time taken to switch between uplink and downlink is smaller than the coherence time of the channel.
10.3 Performance analysis of linear precoding After applying the transformation matrix T on the data block d the resulted transmitted signal will be sp . Subscript p distinguishes the transmitted signal between the conventional CDMA system and the one with precoding: sp = T d.
(10.1)
sp is still of length N Q, like the original s and thereby matrix T is of size N Q × KN . The received vector ek at each mobile site will now be: e k = H k T d + vk
(10.2)
It is desirable for the receiver of the mobile terminal k to be a filter matched to the spreading code ck . However, a RAKE receiver is often adopted in CDMA receivers, so one precoding algorithm that requires a RAKE receiver will be examined. Thus, in (10.3) the output of the receiver is given for both cases: & BTk Hk T d + BTk vk , matched filter, ˆ (10.3) dk = T T T T RAKE receiver. Ck Hk Hk T d + Ck Hk vk , From (10.3) we can conclude that the detected signal at the output of the receiver has the general form ˆ k = A k d + A v vk , d
k = 1, . . . , K,
(10.4)
where Ak is an N ×KN matrix with elements {[Ak ]i+M +1,j }, i = −M, . . . , M , j = 1, . . . , KN and Av is also of dimensions N × KN . Based on (10.4) we follow a general approach to calculate the theoretical BER performance of the precoding systems. The technique is similar to the one followed in (Klein, 1996) for receiver-based linear multiuser detection ˆ techniques. Each component dˆm k , m = −M, . . . , M , in dk consists of three contributions, the first determined by the desired symbol dm k , the second determined by ISI and MAI and the third determined by noise: ˆ k = diag(Ak )d + diag(Ak )d + AT vk . d v desired symbol
ISI and MAI
noise
(10.5)
274
Next Generation Mobile Access Technologies
In (10.5), diag(Ak ) represents a diagonal matrix containing only the diagonal elements of the matrix Ak and diag(Ak ) = Ak − diag(Ak ) represents a matrix with zero diagonal elements containing all but the diagonal elements of Ak . The covariance matrix of the noise term, in (10.5), is equal to ATv Rv Av .
(10.6)
A performance measure is the signal-to-noise and interference ratio (SNIR) ˆm φ(dˆm k ) of dk at the receiver’s output. The SNIR is given by: 4 5 2 E symbol energy in dˆm k 5. 5 4 4 (10.7) φ(dˆm k )= 2 2 + E noise in dˆm E ISI and MAI in dˆm k
k
The SNIR defined in (10.7) takes into account the perturbation due to noise, ISI and MAI. We can substitute the components from (10.5) to obtain: desired symbol
) * m 2 E |[diag(A )d] | k *. * ) ) φ(dˆm k )= E [diag(Ak )d]m 2 + E |[Av v]m |2
(10.8)
noise
ISI and MAI
Proceeding further to determine the desired and perturbation coefficients we can write:
φ(dˆm k )=
desired symbol
m ′ ,j ′
[Ak ]
2
4 ′ 52 j E |dk |
′
′
ISI & MAI + [Av Rv ATv ]m ,j
,
(10.9)
noise
where
′
′
(10.10) ISI & MAI = [Ak Rd ATk ]m ,j 5 4 ′ 5 4 ′ ′ ′ ′ ′ ′ 2 . − 2Re [Ak Rd ]m ,j [ATk ]m ,j + |[Ak ]m ,j |2 E |dm k |
In (10.9) and (10.11) m′ = m + M + 1 and j ′ = M + 1 + N (k − 1). For a matched filter receiver, and taking into account the assumptions previously made that Rd = I and Rv = σ 2 I, the expression in (10.8) can be simplified to: ′ ′ 2 [Ak ]m ,j φ(dˆm . (10.11) k ) = K N m ′ ,j )2 + σ 2 j =1 ([Ak ] j =j ′
Interference-based cancellation techniques for TDD
275
The SNIR as given by (10.11) can be used for calculating the error probability Pe for the detected BPSK bit dˆ0k . For K → ∞ we can consider that ISI and MAI follow a Gaussian distribution as dictated by the central limit theorem. Therefore, the error probability Pe (dˆ0k ) is given by applying the Q function:
m m ˆ ˆ P e (d k ) = Q φ(dk ) . (10.12) For a system of K users the average theoretical error probability is: K Pe (dˆm m k ) ˆ ¯ . (10.13) Pe (d ) = k=1 K A critical issue in CDMA systems is to provide all users with equivalent performance assuming that signals of equal power are addressed to them. That means that the SNIR φk must be approximately the same for every user. By defining the SNIR spread Φxs as: Φxs =
max{φi } min{φj }
i, j ∈ {1, . . . , K} for Eb /No = x [dB]
(10.14)
we have a measure of the variety among the K individual performances for Eb /No = x dB. Ideal values for Φxs are close to unity. SNIR spread behaviour varies according to the precoding algorithm used and it will be considered in the following chapters.
10.4 Precoding techniques in classification As discussed above the precoded transmitted signal can be described in general as a linear transformation matrix T applied on the data. In this section we classify the realisation of this linear algorithm in either blockwise or bitwise fashion. Although they can both be described by an appropriate matrix T there are important practical differences. 10.4.1 Blockwise techniques In blockwise algorithms the data is precoded and transmitted in blocks of N data bits for each of the K users in a cell. This means that the transformation matrix T is applied to a block of N K data bits. Furthermore, it is desirable for the resulting signal sp to retain the same length N Q as produced by the conventional CDMA system. Thus, dimensions of the T matrix shall be N K × N Q. The block diagram of a blockwise precoding technique is shown in Figure 10.3.
276
Next Generation Mobile Access Technologies M
d1 0
d1 M
d=
d1
T d
M
dK
sp NQ 1
0
dK T : NQ
M
dK
KN
N=2M+1 KN
1
Fig. 10.3. Blockwise precoding diagram.
Matrix T can be chosen according to different optimisation criteria and the complexity for determining it varies from algorithm to algorithm. Ignoring this complexity, the multiplications required per bit per user for transmission of any blockwise technique results from sp = T d, is denoted as Ωblockwise and given by:
Ωblockwise =
(KN )(N Q)(KN ) = N 2 QK. KN
(10.15)
Equation. (10.15) shows that the multiplications required for one symbol transmission are proportional to the square of a block of length N 2 , which for a practical WCDMA-TDD would be excessively high. Another not so apparent problem with blockwise algorithms is the fact that in a multipath environment the end bits of each block are not correctly precoded and the overall performance will degrade. Each block is treated independently and the ISI, between the symbols of the block itself, will be cancelled. However, under the scenario of continual block transmission each block is extended in duration due to multipath and therefore overlaps with the next one. The degree of damage (the number of symbols affected) depends on the channel’s delay spread L. One solution to mitigate this problem is to increase the block length, but this results in increased complexity. A second solution is to neglect the number of affected symbols resulting in loss of capacity.
Interference-based cancellation techniques for TDD
277
10.4.2 Bitwise techniques In bitwise techniques, the precoding realisation does not include blocks of data to be processed. The data of user k after being spread is pre-filtered by a finite impulse response (FIR) filter of length P with a discrete time impulse response pk (n), as shown in Figure 10.4. The resulting modified signals are summed to form the final transmitted signal sp . The filter is applied to the transmitted waveform, rather than the data symbols as in the blockwise techniques. Thus the multipath resolution afforded by the CDMA spreading is exploited. Moreover, since the filters are applied at the chip level and the derivation of the algorithms are restricted to consider only a few symbols (even one), the higher computation required for data-block solutions is eliminated. Pre-filters have also the advantage of not modifying the original CDMA structure directly as it is a simple addition of an array of FIR filters to the existing BS transmission system. Furthermore, the bitwise nature of the algorithms removes the undesirable effect of the end block-bits as described in the previous section. The taps of the pre-filters are determined by the adopted technique. Ignoring the complexity of the algorithm used to determine the taps, the multiplications required now per transmitted symbol per user, Ωbitwise , is given by: Ωbitwise = P Q.
(10.16)
c 1 (n)
y (n) p 1 (n)
m d1
1
Σ
sp
c K (n) p K (n)
dm K
y (n) K
c (0)
c(Q)
Q
p(0)
p(P)
P
Fig. 10.4. Bitwise precoding diagram.
Comparing (10.16) and (10.15) we calculate the relative complexity of
278
Next Generation Mobile Access Technologies
blockwise and bitwise techniques: Ωblockwise N 2K . = Ωbitwise P
(10.17)
In a CDMA system with linear MUD the users that can be accommodated do not outnumber the spreading gain: K ≤ Q . If we select as a reasonable pre-filter length P twice the spreading gain Q and K = Q/2 then: Ωblockwise N2 N 2 Q/2 = . = Ωbitwise 2Q 4
(10.18)
Apparently, for N ≥ 2 (in a realistic system it is N >> 2) the multiplications required with a blockwise implementation of precoding outnumber the corresponding bitwise implementation by a large number. It is essential for the derivation of theoretical BER to describe the bitwise techniques in the matrix form of (10.1). This description is straightforward for the blockwise approach. Vector pk represents the tap-coefficients of the k th user-specific FIR pre-filter: pk = [pk (0) · · · pk (P − 1)]T .
(10.19)
gk = [gk (0) · · · gk (G − 1)]T .
(10.20)
m The transmitted symbol is now expanded to dm k ck (n) ∗ pk (n) = dk gk (n), where ∗ denotes the convolution effect. The length of the pre-filter output symbol gk in vector form is of length G = Q + P − 1 chips:
The pre-filters impose intersymbol interference on the system. This interference is caused by the symbols transmitted before and after the symbol under consideration. Without affecting the generality, for the bitwise techniques let us assume that the transmitted symbol under consideration is d0k . Let yk (n) be the signal at the output of the kth pre-filter as shown in Figure 10.4. In order to describe this signal in matrix form we must take into account M transmitted bits before and after. The right choice for M is 6 7 M = (G − Q)/Q , where x denotes that x is rounded up to the nearest integer. Thus, the G-length transmitted vector for user k, yk , is written as: yk = Gk dk , where the G × N matrix Gk is:
(10.21)
Gk = {[Gk ]i,(j+M +1) }; i = 1, . . . , G, j = −M, . . . , M, & gk (i − jQ − 1), for 0 ≤ i − jQ − 1 ≤ G − 1, i,(j+M +1) = [Gk ] 0, otherwise. (10.22)
Interference-based cancellation techniques for TDD
279
The summation of yk for all users gives the transmitted signal sp , which can be written as the G-length column vector: sp =
K =1
yk = Gd,
(10.23)
k=1
where the G × N K matrix G is defined as: ( ' G = G 1 · · · Gk · · · GK .
(10.24)
From (10.1) and (10.23), it is obvious that matrix G corresponds to T for the bitwise algorithm matrix description. Matrix G is associated with the pre-filters taps and it is based on the optimisation method chosen to define the taps. Now we can apply G directly in the analysis of section 10.3 to calculate the theoretical BER of a bitwise precoding system.
10.5 Power scaling factor All the precoding techniques examined aim to mitigate the ISI and MAI impairments of a CDMA system before the transmission. In Figure 10.2, ignoring noise and assuming a unit gain channel, it is apparent that the m objective is dˆm k = dk . Depending on the method employed in the ‘precoding and spreading procedure’ box , a certain transmit power is required for interference elimination which is usually greater than the one required for conventional spreading. In real systems, however, power is limited. Therefore, the transmitted signal sp must be scaled to maintain the total (for K users) average transmitted energy per bit interval to the desired level. The transmitted power may vary from user to user individually, according to the precoding algorithm employed, but the overall average energy per bit interval is constant. The reference power used is the one for the corresponding conventional CDMA system. The unscaled total transmission power per symbol produced by any algorithm is denoted as Eg . For a precoding system that handles blocks of data the transmitted energy per symbol (Vojˇci´c and Jang, 1998) is given (10.25): trace{T T T } . (10.25) N The average total transmitted energy Eg per symbol for a system using pre-filters is: Eg =
Eg =
K k=1
ck ∗ pk 2 .
(10.26)
280
Next Generation Mobile Access Technologies
For a conventional CDMA system the energy per transmitted symbol per user is denoted as Eg and for W = I it is equivalent to: Eg = trace{CT C}.
(10.27)
The matrix CT C is the cross-correlation matrix of the spreading codes and in the case of signature waveforms normalised to ck = 1, Eg is equal to the number of users active in the system: Eg = K.
(10.28)
At this point the vectors pk , k = 1, . . . , K, (10.19), will be arranged in a KP -length vector p as: p = [pT1 · · · pTk · · · pTK ]T .
(10.29)
The energy Eg of the transmitted signal sp , for a bitwise algorithm, will be expressed as a function of p: Eg (p) = pT UT Up,
(10.30)
where U is a matrix of size K(Q + P − 1) × KP : U = blockdiag U1 , . . . , Uk , . . . UK ,
and Uk is a matrix of dimension (Q + P − 1) × P , given in (10.31) and illustrated in (10.32): Uk = {[Uk ]i,j }; [Uk ]
i,j
=
⎡
&
i = 1, . . . , Q + P − 1,
ck (i − j), 0,
ck (0) .. .
for
j = 1, . . . , P,
0 ≤ i − j ≤ Q − 1, otherwise,
⎢ ⎢ ck (0) ⎢ ⎢ .. .. ⎢ ck (Q − 1) . . ⎢ Uk = ⎢ (Q − 1) c k ⎢ ⎢ .. ⎢ . ⎢ ⎢ ⎣ 0
0
ck (0) .. . ck (Q − 1)
(10.31)
⎤
⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥. ⎥ ⎥ ⎥ ⎥ ⎥ ⎦
(10.32)
The energy level of transmitted signal sp cannot exceed Eg . To control the energy of the output signal sp of the precoding algorithms one of the following methods can be used:
Interference-based cancellation techniques for TDD
• Scale the signal sp before transmission with a scaling factor F=
Eg . Eg
281
√
F: (10.33)
This is called a global power normalisation technique and can be applied in both the bitwise and blockwise techniques. If this normalisation factor F < 1.0, the scaling results in a decrease in the SNR at the receiver decision point. With this in mind, the optimum precoding technique is one that minimises the MAI and ISI and maximises F. • In (Brandt-Pearce and Dharap, 2000) a different power scaling technique was applied for the bitwise algorithm, called decorrelating pre-filters. Each user is individually normalised to achieve a specified received SNR in contrast with the global normalisation derived in (10.33). The data bit 1 dm k is scaled before spreading with the factor ck ∗pk . Thus, after being filtered by the pk pre-filter its power is normalised to 1. • Power constraint optimisation criteria should be imposed on the algorithms that give the precoding solution. In this case the resulting power may not be exactly the intended Eg and a extra scaling step should take place, but the scaling step would be a small adjustment. 10.6 Joint transmission In joint transmission ( (Meurer et al., 2000; Papathanasiou et al., 2000)) the algorithm seeks for a common signal sp , of length N Q, which gives the desired data at the output of every user’s receiver. A noiseless version of the received signal ek at user site k is written as: ek = Hk sp .
(10.34)
Arranging all the K matrices Hk into the K(N Q + W − 1) × N Q matrix gives H: (T ' (10.35) H = HT1 · · · HTk · · · HTK
and arranging all the ek , k = 1, . . . , K, vectors to a total received vector e gives: (T ' (10.36) eT = eT1 · · · eTk · · · eTK , which results in:
e = Hsp .
(10.37)
The total received signal e of (10.37) cannot be observed by a single MS. ˆ k , k = 1, . . . , K, at the decision variable of the K users are The outputs d
282
Next Generation Mobile Access Technologies
ˆ combined to form the vector d: ( ' T ˆ ···d ˆT · · · d ˆT T . ˆT = d d 1 K k
(10.38)
Recalling that the k-user’s receiver is a filter matched to the user’s spreading ˆ can be written as: code, d ˆ = BT e = BT Hsp d
(10.39)
In (10.39) the K(N Q + W − 1) × KN block diagonal matrix B is defined as: (10.40) B = blockdiag B1 , . . . , Bk , . . . , BK .
The objective of any precoding algorithm is that the output of the mobile receiver yields the transmitted data: ˆ = d. d
(10.41)
By substitution of (10.41) into (10.39) the problem of JT takes the form: BT Hsp = d.
(10.42)
Matrices B, H and vector d in (10.42) are known at the BS and sp is an unknown vector with length N Q. The condition KN ≤ N Q
(10.43)
is a basic assumption for TD-CDMA (Klein, 1996), and hence (10.42) constitutes an optimisation problem (Walsh, 1975), where the number of restrictions contained in the vector d of length KN is smaller than the number of degrees of freedom N Q, expressed by the vector sp . Therefore, the validity of (10.43) implies that (10.42) has infinitely many solutions for sp . Consequently it is necessary to impose a constraint criterion on the system. It is desirable for the transmitted vector sp to be of minimum energy this is achieved by following standard Lagrange techniques (Scharf, 1991). The minimisation of the power sTp sp using the constraints as expressed in (10.42) is described with the following Lagrange function: F (sp ) = sTp sp − λT (BT Hsp − d),
(10.44)
where λ is the KN × 1 vector of Lagrange multipliers. The minimisation of (10.44) can be achieved by taking the gradient of F (sp ): ϑF (sp ) = 2sp − HT Bλ = 0. ϑsp
(10.45)
Interference-based cancellation techniques for TDD
283
The solution is straightforward: sp = HT Bλ.
(10.46)
The factor 2 in (10.45) has been absorbed by λ. By replacing sp in (10.42) with the right-hand part of (10.46) the Lagrange multipliers can be calculated: −1 (10.47) λ = BT H HT B d. Hence, by manipulating (10.46) the final solution for signal sp is proved to be: −1 (10.48) sp = HT B BT H HT B d.
Recalling that the general linear precoding is described in (10.1), it is obvious that the joint precoding matrix T for JT can be written as: −1 T J T = HT B B T H HT B . (10.49)
Equation (10.46) shows that joint transmission can be considered as a set of independent pre-RAKEs (HT B) applied on the modified information signal given by λ, which is a linear transformation of the data vector d (Barreto and Fettweis, 2001). The same equation for linear transformation of the transmitted signal has been derived through different methods in (Joham and Utschick, 2000) and (Kowalewski, 2000). The conclusions reached for a system with a conventional receiver apply also when a RAKE receiver is employed, in which case the channel matrix H in (10.48) should be substituted by the matrix H′ that represents the cascaded channel and RAKE receiver H′ = HT H. In case the K channels are AWGN channels H = I and B is equal to C. Consequently T J T becomes: −1 T J T = C CT C . (10.50) 10.7 Transmitter precoding Transmitter precoding was presented in (Vojˇci´c and Jang, 1998). It represents a linear transformation of the transmitted data, such that the mean squared errors at all receivers are minimised. The authors describe both a power unconstrained and constrained optimisation. However, the case of power constrained precoding is given only for a flat fading channel. In the unconstrained algorithm the transmitted signal is scaled with an appropriate factor in order to maintain the same transmit power as in the case without precoding. (Vojˇci´c and Jang, 1998) consider transmitter precoding with either a conventional matched filter receiver or a RAKE receiver.
284
Next Generation Mobile Access Technologies
10.7.1 Unconstrained optimisation In a conventional CDMA system the matched filter outputs of all mobile receivers can be jointly expressed by defining the following N K ×N K matrix U: ⎤ ⎡ T ⎤ ⎡ B1 BT1 H1 C ⎥ ⎢ .. ⎥ ⎢ .. ⎥ ⎢ . ⎥ ⎢ ⎥ ⎢ T ⎥ ⎢ T . ˆ ⎥ ⎥ ⎢ (10.51) d = ⎢ B k Hk C ⎥ d + ⎢ ⎢ Bk ⎥ v, ⎥ ⎢ .. ⎥ ⎢ .. ⎦ ⎣ . ⎦ ⎣ . T B HK C BTK K U
` v
` is a zero-mean Gaussian noise vector with covariance matrix equal where v to diag{σ 2 }, if c2k = 1. To reduce the effect of MAI and ISI, (Vojˇci´c and Jang, 1998) apply a linear transformation matrix T of dimension N K ×N K, on the block of data before the spreading process. Therefore, after precoding and spreading, the vector at the outputs of the MS receivers will be: ˆ = UTd + v `. d
(10.52)
In order to calculate T the Minimum Mean Square Error (MMSE) criterion is employed. The cost function J that has to be minimised is defined as: ' ( ˆ 2 J = Ed,`v d − d ' ( ` )2 . (10.53) = Ed,`v d − (UTd + v Ed,`v [·] represents the expectation with respect to the data block vector d ` . The matrix T that minimises J is given in (10.54) (the and noise vector v derivation is given in (Vojˇci´c and Jang, 1998)): T = U−1 .
(10.54)
ˆ =d+v `. d
(10.55)
Hence, (10.52) becomes:
The multiuser detection problem that normally should have taken place in the receiver has been decoupled into N K separate single-user detection problems, by means of precoding. An additional advantage is that the noise is not enhanced at the receiver, due to the lack of a receiver-based decorrelating process. The cost is an increase in the transmitted energy, which may neutralise the benefit described before. (Vojˇci´c and Jang, 1998) give the precoding matrix solution when a RAKE
Interference-based cancellation techniques for TDD
285
receiver is employed. Under this scenario T is still equal to U−1 but now U is defined as: ⎡ ⎤ CT1 HT1 H1 C ⎢ ⎥ .. ⎢ ⎥ . ⎢ ⎥ T T ⎢ U = ⎢ Ck Hk Hk C ⎥ (10.56) ⎥. ⎢ ⎥ .. ⎣ ⎦ . T T CK HK HK C
Hereafter, we will denote the TP with matched filter receiver as TP-MAT to distinguish it from the TP with RAKE receiver, which in turn will be denoted as TP-RAKE. From both definitions of U in (10.51) and (10.56) C can be taken out to write U = U′ C. C spreads the data after precoding and U′ corresponds to the cascaded channel and receiver. Observing (10.52) and replacing U with U′ C, it is easy to write sp as: sp = CTd.
(10.57)
From (10.57) and (10.54) it is obvious now that the transformation matrix T T P for TP is: T T P = CU−1 . In the case of a single-path channel, matrix U is simplified as: −1 T = CT C
(10.58)
(10.59)
which is the N K × N K cross-correlation matrix of the spreading codes. The same result was produced in (Tang and Cheng, 1994b,a) and described as pre-decorrelating. For the scenario −1 of a flat fading channel the precoding matrix for TP is T T P = C CT C , which is the same with T J T in (10.50). The later shows that for a single-path channel the JT and TP transformation matrices are identical. 10.7.2 Constrained optimisation The transmitter precoding technique was also extended in (Vojˇci´c and Jang, 1998) to include a power constraint. However, this was done only for the scenario of a single-path AWGN channel. Under this scheme (10.52) is ˆ = CT CTd + CT v. A direct approach to find the optimum written as d linear transformation matrix T, of dimensions N K ×N K, under the average power constraint can be formulated as: ' ( ˆ 2 (10.60) minEd,`v d − d T
286
Next Generation Mobile Access Technologies
under the constraint: Eg (T) = trace{TT CT CT} = Eg ,
(10.61)
where Eg (T) is the total average transmit energy with precoding and Eg is the corresponding power of a non-precoding equivalent CDMA system. For the AWGN channel there is no intersymbol interference and the analysis is sufficient for a block length of one (N = 1). In that case, the total average transmit energy per bit interval Eg (T) should be equal to Eg = K, assuming that ck 2 = 1. This problem can be solved by means of the Lagrange multiplier. The precoding matrix T should minimise the constrained cost function Jc : * ) ˆ 2 + λEg (T) Jc = E d − d (10.62) The solution is given in (Vojˇci´c and Jang, 1998): T = (CT C + λI)−1 ,
(10.63)
where I is the identity matrix and λ can be found from (10.62) and (10.63). It is shown in (Vojˇci´c and Jang, 1998) that a unique λ ≥ 0 exists as the desired solution where U, where U is a positive define matrix (a property of the cross-correlation matrix (Haykin, 1991)). The constrained transmitter precoding optimisation results in a worse performance than the unconstrained one with power scaling, as demonstrated in (Vojˇci´c and Jang, 1998). This fact, in conjunction with the lack of an extension for the multipath environment, means that we will not examine this algorithm any further.
10.8 Decorrelating prefilters jointly optimised sequences In (Brandt-Pearce and Dharap, 2000) channel and multiple access effects are mitigated by jointly designing user transmit waveforms. This is an alternative approach to the one given in (Vojˇci´c and Jang, 1998), having different implementation. Rather than relying on transmit data filtering and receiver RAKE or matched filter processing, the transmit waveforms themselves are modified. The methods described in (Brandt-Pearce and Dharap, 2000) are named as decorrelating pre-filters (DPF) and jointly optimised sequences (JOS). They attempt to precompensate for the channel multipath and minimise the cross-correlation of other users at the individual receivers simultaneously. The system is analogous to a system using a multiuser detector, specifically the decorrelator (Klein, 1996), at the receiver except that the processing is now at the transmitter.
Interference-based cancellation techniques for TDD
287
10.8.1 Decorrelating prefilters The block diagram of the DPF follows the bitwise precoding one as illustrated in Figure 10.4. The algorithm is similar to TP-MAT, which calculates the best pre-filter to apply to the data symbols, filtering the data by blocks. DPF are applied to the transmit signal (instead of the data symbols directly, as in (Vojˇci´c and Jang, 1998)), so that the filter can be applied externally and the multipath component resolution available from spread-spectrum coding can be exploited. Under the assumption of a synchronous CDMA system the received signal at mobile user j is written as: ej (n) =
K k=1
dk ck (n) ∗ pk (n) ∗ hj (n) + vj (n).
(10.64)
The pre-filters pk (n), k = 1, . . . , K, are designed such that the signal meant for interfering users is orthogonal to the matched filter response for the desired user, irrespective of the received signals powers. Each user’s prefilter must be designed jointly taking into account the channel information and the transmit codes of all users. The approach taken in (Brandt-Pearce and Dharap, 2000) is to design the pre-filters that maximise the signal to noise ratio at the receiver while reducing the MAI interference. This method is not a strict error probability minimising scheme, but rather a zero-forcing simultaneous multiuser interference rejecting and channel pre-equalisation method. The problem formulation is stated as follows. The matched filter output at the receiver of user j is: dˆj =
K k=1
dk ck (n) ∗ pk (n) ∗ hj (n) ∗ cj (Q − n) + v˜j (n),
(10.65)
where v˜j (n) is the noise contribution to sample n of the receiver’s output. To simplify the notation, let: γij (n) = ci (n) ∗ hj (n) ∗ cj (Q − n); i, j = 1, , , K.
(10.66)
Then (10.65) becomes: dˆj =
K k=1
dk pk (n) ∗ γkj (n) + v˜j (n).
(10.67)
Now the pre-filter optimisation problem is written mathematically as: max pj (n) ∗ γjj (n) |n=Ts s , pj
288
Next Generation Mobile Access Technologies
subject to constraints: pk (n) ∗ γkj |n=Ts s
= 0 ∀ j = k,
pk ∗ ck = 1,
(10.68)
where Tss is the decision sample. The second constraint in (10.68) allows for proper normalisation purposes. Each user is individually normalised to achieve a specified received SNR in contrast with the global normalisation in (10.33). Therefore, the signal addressed to user k must be scaled by 1/ck ∗ pk . The solution is found by solving for the minimum two-norm vector pk satisfying: ⎤ ⎡ γk1 (TSS ) γk1 (TSS − 1) · · · γk1 (TSS − P + 1) ⎢ γk2 (TSS ) γk2 (TSS − 1) · · · γk2 (TSS − P + 1) ⎥ ⎥ ⎢ ⎥ ×pk = D k , ⎢ .. ⎦ ⎣ .
γkK (TSS ) γkK (TSS − 1) · · ·
γkK (TSS − P + 1)
˜k A
(10.69) where D k is an all-zero vector except for a unity entry at position k. This ˜ k and a solution is guaranequation is solved using the pseudo-inverse of A ˜ k is full teed to exist as long as the length of the pre-filter P ≥ K (if matrix A rank, an exact zero-forcing solution is obtained; otherwise, a least-squares solution results). DPF carries all the advantages of the bitwise approach in terms of complexity and can be applied externally to an existing design, for the preprocessing matrix is applied to the output of the spread spectrum encoder. On the contrary, TP applies the linear transformation first and then the spreading. 10.8.2 Jointly optimised sequences This method, also described in (Brandt-Pearce and Dharap, 2000), is suitable for systems that can update the spreading codes adaptively. According to this algorithm the transmitted sequence is designed for each channel such that each desired received signal at the matched filter is orthogonal to the signals meant for all other users. The formulation of this problem is similar to the one for the decorrelating pre-filters except that the concatenated ˜ k is different. channel filter matrix previously defined as A The optimally designed signature sequence ζ k , of length Ls , used for the signal intended for receiver k must now replace what was represented above as ck ∗ pk . The user-to-user concatenated channel, denoted by γij (n) above, is defined now as: zj (n) = hj (n) ∗ cj (Q − n).
(10.70)
Interference-based cancellation techniques for TDD
The channel filter matrix equation becomes: ⎡ z1 (TSS ) z1 (TSS − 1) · · · z1 (TSS − Ls + 1) ⎢ z2 (TSS ) z2 (TSS − 1) · · · z2 (TSS − Ls + 1) ⎢ ⎢ .. ⎣ . zK (TSS ) zK (TSS − 1) · · · zK (TSS − Ls + 1) ˜ B
289
⎤
⎥ ⎥ ⎥ ×ζ k = D k . (10.71) ⎦
˜ must The length Ls of the signature sequence must be at least K and B be a full rank matrix, or the system will not have a direct solution. Again ˜ and the minimum norm solution ζ k is found using the pseudoinverse of B the unit-norm signature sequence for user k is ζ k /ζ k . The normalisation guarantees that the transmit power will be equal to a conventional CDMA system. Jointly optimised sequences could also be derived for a block of symbols in a manner similar to the derivation of TP in (Vojˇci´c and Jang, 1998). The solution for the best jointly optimised sequence would be found by modifying (10.71) to include the transmitted symbols for the block data. Block processing has the advantage that the residual ISI is eliminated between symbols of the same block. However, it has the disadvantages that the channel must remain stationary for the duration of the block and the end bits of the block are not correctly precoded.
10.9 Pre-RAKE diversity This technique was initially described in (Esmailzadeh and Nakagawa, 1993c) for DS-spread spectrum systems and then extended in (Esmailzadeh et al., 1999) for CDMA-TDD mobile communications. The algorithm for preRAKE comes from the diversity theory in frequency selective channels. In a mobile environment the combination of the received signals from diverse independent paths or mediums can improve the system performance. According to the central limit theorem, as the number of independent paths increase, the combination of their signals will have less of a Rayleigh fading characteristic and more of a Gaussian one (Haykin, 1991). Hence in the radio mobile communications it is desirable to receive a signal from diverse independent paths and then combine their powers. This is what a RAKE receiver does as described in section 10.3. The pre-RAKE technique is straightforward and takes advantage of the fact that in TDD the channel impulse response can be assumed the same for the uplink and the downlink. The idea is to transmit a number of signals that when received after the multipath channel merge to a signal with the
290
Next Generation Mobile Access Technologies
characteristics of a RAKE diversity signal. Each one of these transmissions should be then delayed according to the estimated relative path delay, and amplified according to the estimated path complex coefficient. In other words, in the downlink the BS multiplies the signal to be sent to user k by the time-inverted complex conjugate of the uplink channel impulse response of the same user. When the signal is transmitted it is convolved with the channel impulse response of user k. This produces a strong peak at the output of the channel which is equivalent to the RAKE receiver’s output. Therefore, the receiver of the MS does not need to estimate the channel impulse response and only uses a matched filter tuned to this peak. Figure 10.5 displays how the pre-RAKE combination affects an input signal (Dirac function) and the delayed peak produced at the output of the channel.
Fig. 10.5. Pre-RAKE combination process
The analogy of the RAKE and pre-RAKE concepts for a single-user scheme is illustrated in Figures 10.6 and 10.7, while it is mathematically explained in the following pages. We begin with the SNR analysis at the output of the RAKE receiver. In Figure 10.6 the input signal is x(n) = wk dm k ck (n) 2 with a transmission power wk . The signal at point 2, y2 , after passing from the multipath channel and being perturbated by AWGN is: y2 =
L−1 l=0
h(l)x(n − lTc ) + v(n).
(10.72)
The filter immediately after point 2 is matched to the spreading code. The RAKE combination follows and at point 3 we have: y3 =
L−1 l=0
h∗ (L−1−nTc )v(n−lTc )+
L−1 L−1
h(l)h∗ (L−1−m)x(n−(l+m)Tc ).
l=0 m=0
L−1
(10.73)
∗ l=0 h(l)h (l)x(n − (L − 1)Tc )
The desired output of the RAKE system is and occurs at time n = (L − 1)Tc (l + m = L − 1).
Interference-based cancellation techniques for TDD
291
x(n) 1 spreading
+
h0
2 Tc
matched filter
3
h L* 1
h1
Tc
h L* 2
2 Tc
h L* 3
decision output
noise v(n) 2 Tc
(L1) Tc
h2
(L1) Tc
h L1
h 0*
RAKE combiner
multipath channel
Fig. 10.6. RAKE combiner.
x(n) 5
4
* h L1
spreading
h0
Tc
* h L2
Tc
h1
2 Tc
* h L3
2 Tc
h2
+
matched filter
6
noise v(n)
(L1) Tc
h 0*
(L1) Tc
preRAKE Combiner
h L1
multipath channel
Fig. 10.7. Pre-RAKE combiner
This has a signal strength of: L−1 l=0
h(l)h∗ (l)
2
wk2 .
Noise adds up incoherently and its power is: L−1 h(l)h∗ (l) σ 2 .
(10.74)
(10.75)
l=0
From (10.74) and (10.75) the SNR at the output of the RAKE receiver is equal to: L−1 wk2 SN RRAK E = 2 h(l)h∗ (l) (10.76) σ l=0
In Figure 10.7 the block diagram of a pre-RAKE combiner is illustrated. In order to make sure that the power of the signal at the output of the
292
Next Generation Mobile Access Technologies
pre-RAKE transmitter is equal to wk2 , a power scaling factor should be chosen to compensate for the gains produced by h(l)∗ . In (Esmailzadeh and Nakagawa, 1993c) the scaling factor is: 1 . F = L−1 ∗ l=0 h(l)h (l)
(10.77)
In Figure 10.7 after pre-RAKE combining the signal at point 4, y4 , is: y4 = F
L−1 l=0
h∗ (n − lTc )x(n − lTc )
(10.78)
After transmission, convolution with the multipath channel and addition of AWGN the signal at point 6 of Figure 10.7 becomes: L−1 L−1 ∗ 2 y6 = F h (l)h(L − 1 − m)x(n − (l + m)Tc ) + v(n). (10.79) l=0 m=0
The desired signal occurs again at time n = (L − 1)Tc and has a power equal to: 2 L−1 ∗ 2 (10.80) h(l)h (l) . wk F l=0
The noise at the output of the matched filter has a power equal to σ 2 . We replace F with the right-hand part of (10.77) and thus the SNR is written as: L−1 L−1 2 2 wk2 2 wk ∗ ∗ = 2 h(l)h (l) h(l)h (l) . (10.81) SN Rpre-RAKE = F 2 σ σ l=0
l=0
From (10.76) and (10.81) the SNR for the pre-RAKE system is equal to that of the RAKE system, for a single-user. This analysis holds only under the assumption that the channel parameters are estimated ideally on the transmitter (Esmailzadeh and Nakagawa, 1993c). The two systems have similar performance. In Figure 10.8 the noiseless RAKE receiver’s output is illustrated and this corresponds to point 3 of Figure 10.6. The graph in Figure 10.9 shows the matched filter’s output and corresponds to the noiseless pre-RAKE block diagram graph of Figure 10.7. In both Figures we indicate the delay after which the peak is detected. The multipath channel has a length of L = 11 chips. The data has been spread with a signature waveform of Q = 16 chips. The similarities between single-user pre-RAKE
Interference-based cancellation techniques for TDD
293
1.5 data receiver output 1
0.5
0
0.5
1
Delay ∆ in detection 1.5
0
20
40
60
80
100
chips
Fig. 10.8. Single-user noiseless output of a RAKE receiver.
and RAKE systems are obvious and will be verified in the next section by means of a BER performance comparison. In (Esmailzadeh et al., 1999) the concept of pre-RAKE was extended to a multiuser CDMA-TDD system. The signal for user k after spreading is filtered by the time-inverse complex conjugate impulse response of the k-user’s channel and this process is repeated for every user before transmission. The produced signals are appropriately scaled to maintain the power at the desired levels and then superimposed at the antenna element. The block diagram of the pre-RAKE technique applied in a CDMA-TDD system is identical with the general description of bitwise precoding technique illustrated in Figure 10.4. Therefore, pre-RAKE has advantages of a bitwise approach in terms of complexity and simplicity in implementation. Furthermore, the calculation of the pre-filter’s tap coefficients don’t require any complicated algorithms. Each pre-filter’s impulse response is the timeinverse conjugate of the corresponding uplink channel. The transmitted signal sp (n) under the multiuser scheme is: sp (n) =
K k=1
Fk
L−1 l=0
h∗ (n − lTc )xk (n − lTc ).
(10.82)
294
Next Generation Mobile Access Technologies
Fig. 10.9. Single-user noiseless matched filter output after pre-RAKE processing.
The principle applied is that in the downlink the desired user’s signal is maximal ratio combined by the channel itself while other user signals are not. In the analysis presented in (Esmailzadeh et al., 1999) at the sampling time the matched filter output consists of the desired part of the transmitted bit, the intersymbol and multiple access interference and the noise. The mean and the variance of each one of these terms is calculated. Under the multiuser scenario the pre-RAKE is proved to be an insufficient technique unable to compensate for the multipath interference and MAI, as shown in (Esmailzadeh et al., 1999) and (Georgoulis and Cruickshank, 2001).
10.10 Complexity The precoding techniques described in this chapter require excessive computational cost compared to the usual conventional spreading. The complexity is similar or exceeds the one demonstrated by the receiver based multiuser techniques as described in (Klein and Baier, 1993), (Zvoran and Brady, 1995) and (Klein, 1996). However, it is shifted to the BS, where the resources are more readily available and thus it is less critical. Table 10.1
Interference-based cancellation techniques for TDD
295
summarises the computational cost for JT, TP-MAT, DPF and Pre-RAKE. The criteria used are the dimension of the matrix that needs to be inverted and the number of multiplications per transmitted symbol. The number of multiplications are calculated after the final transformation matrix T , for the blockwise techniques, as defined in section 10.4, has been formulated. As expected the blockwise techniques JT and TP-MAT are more demanding in terms of complexity compared to the bitwise DPF and pre-RAKE. Table 10.1. Computational complexity of precoding algorithms. Precoding algorithm
Dimension of matrix to be inverted
Number of multiplications per symbol
JT
KN × KN
N 2 QK
TP-MAT
KN × KN
N 2 QK
DPF
K(P × P )
PQ
Pre-RAKE
none
LQ
It will be shown in the next section that JT outperforms in terms of performance the other techniques. Therefore, it is useful to emphasise in this technique. The JT algorithm can be optimised by taking advantage of T T two properties of matrix B H H B (Kowalewski, 2000): (i) BT H HT B is Hermitian (ii) It consists of non-vanishing blocks of coefficients located near the diagonal which are repeated along the diagonal. Property 1 allows the triangular decomposition of the matrix (Cholesky −1 decomposition), which in turn allows us to calculate BT H HT B d efficiently by back-substitution. Property 2 allows us to decompose T T B H H B more efficiently than by the usual full Cholesky decomposition. The complexity of JT can be compared with the basic equation of receiver-based ZF-JD: ˆ = (Hk C)H (Hk C) −1 (Hk C)H ek . d
(10.83)
The matrices to be inverted for both receiver- and transmitter-based techniques are of dimension KN × KN . Both have exactly the same structure with exactly the same non-vanishing elements. Both matrices can be inverted efficiently by taking into account properties 1 and 2. Therefore inverting the matrices for ZF-JD and JT requires the same computational
296
Next Generation Mobile Access Technologies
power. The only complexity difference between JT and ZF-JT arises from the matrix multiplications HT B and (Hk C)H respectively. The complexity of the joint predistortion techniques can be further reduced by exploiting the structure of the matrices as explained in (Wathan et al., 2002). In a realistic system, however, the multiplications required per symbol after the matrix inversion outnumber the ones required by a bitwise technique.
10.11 Other techniques The precoding techniques described in this chapter are only representative and most commonly used as a reference. In the recent bibliography there are alternative techniques for reducing the mobile complexity that may utilise smart antennas (discussed in the next chapter), or even the frequency domain of a system. A comparison and review among different linear precoding techniques can be found in (Barreto and G.Fettweis, 2001) and (Georgoulis and Cruickshank, 2001). The usage of antenna arrays can result in spacetime (ST) algorithms that, when applied at the downlink, minimise the MAI and ISI. Such a ST precoding downlink transmission is described in (Wang et al., 2001; Ringel et al., 2002). In (Gameiro et al., 2001), (Morgado et al., 2001) and (Morgado et al., 2002), develop a space-time zero-forcing preequalisation technique for interference cancellation in the TDD downlink, by transferring the analysis to the frequency domain. Moreover, an alternative blockwise precoding technique that incorporates a power constraint (JT with power constraint) is described in (Barreto and Fettweis, 2001), named joint signal precoding. An extension of the pre-RAKE named post-RAKE is presented in (Barreto and Fettweis, 2000). The idea of post-RAKE is to maximise the signal to noise ratio at the MS by replacing the matched filter in the pre-RAKE block diagram with a receiver which is matched to the combination of the pre-RAKE filter and the channel. Finally, joint transmitter-receiver optimisation is also investigated in (Jang et al., 1998). The impairment of the post-RAKE and the transmitter-receiver optimisation algorithms is that the idea of the simplest possible receiver is not valid any more. The precoding techniques described so far induce distortion in the signal’s spectral properties. A power constrained algorithm that acts on the amplitude of the transmitted signals by a simple scaling is demonstrated in Hons et al. (2002). This method is called optimising precoder and maintains the spectral properties of the initial designed signals. However, this method is developed only for a single-path Gaussian channel and, on top of that, the determination of each user’s power need to be readjusted at each bit interval, which in turn results in excessive computational cost.
Interference-based cancellation techniques for TDD
297
10.12 Simulation results One of the drawbacks of the blockwise techniques is that they do not allow for correct precoding of the end data, when a multipath channel occurs. Therefore, the overlapping of the signal’s replicas affect the first and last transmitted bit(s). This results in a performance degradation which is more intense when the length of the transmitted block is small. This is illustrated in Figure 10.10. The performance of JT versus the number of users is shown for block lengths of 1, 3 and 20 data. Spreading gain is set to a default of Q = 16 chips. The simulation corresponds to a system where the channels follow a severe multipath profile (L = 11 chips) and Eb /No = 10 dB. Each curve represents an average over 60 different sets of random codes to smooth out the effect of codes with low or high crosscorrelation. As shown the performance is better for N = 20. Figure 10.11 illustrates how the block
Fig. 10.10. Effect on block length on JT performance.
length affects the scaling factor needed to maintain the total transmitted energy per symbol Eg equal to Eg . The blockwise algorithms displayed are TP-MAT, TP-RAKE and JT. The same system is simulated for each curve and thus a conclusion of how the different methods affect the transmitted energy can be drawn. The system adopted is one of K = 5 users √ in a multipath environment. For each algorithm the scaling factor F is almost
298
Next Generation Mobile Access Technologies 0.75 0.7 JT TPMAT TPRAKE
0.65
√F
0.6 0.55 0.5 0.45 0.4 0
5
10 Block LengthN
15
20
Fig. 10.11. Effect of block length on transmitted energy Eg .
flat versus N . For small N there is some differentiation √ but it can be assumed negligible. Ideally the scaling factor should be F ≥ 1.0, otherwise it deteriorates the SNR at the receiver. Comparing the three algorithms it is obvious that TP-MAT results in the √ highest increase of transmitted energy (it corresponds to the smallest F), while JT is the most modest and displays the least energy increase. TP-RAKE lies in between. The excess energy is the penalty paid for the cancellation of MAI and ISI before transmission. It is expected from these observations that JT will demonstrate the best BER performance and TP-MAT the worst. This will be verified by the simulations that are following. In Figure 10.12 the BER performance is shown versus Eb /No . The system is one of K = 5 users in a severe multipath environment. The best performance is displayed by JT and the worst by the pre-RAKE. In the blockwise techniques the end bits had been discarded from the calculation of BER to give more optimistic results. TP-MAT is worse than TP-RAKE. This is because TP-MAT over-increases the required transmitted energy Eg , compared with TP-RAKE. The RAKE receiver utilised by TP-RAKE exploits the diversity principles and increases the SNR at the receiver output. In the same figure the performance of the MMSE-JD receiver-based
Interference-based cancellation techniques for TDD
299
0
10
10
2
BER
10
1
10
10
3
Pre RAKE DPF TP MAT TP RAKE JT MMSE JD
4
5
10
0
5
10 Eb/No(dB)
15
20
Fig. 10.12. Transmitter-based precoding techniques and MMSE-JT vs Eb /No .
multiuser detection is illustrated. We observe that MMSE-JD outperforms all the transmitter based precoding techniques except JT. JT completely eliminates the ISI and MAI within a block of data and selects the transmission vector with the minimum energy. This, in conjunction with the lack of any receiver-based decorrelation that may increase the noise, justifies the superiority of JT.
11 Smart Antennas for TDD-CDMA Systems John Thompson Ali Dakdouki
11.1 Introduction 11.1.1 Overview of topic The last two decades have been witness to the rapid growth and widespread success of wireless communications. This is largely due to breakthroughs in communication theory and progress in the design of low-cost power-efficient mobile devices. Third generation (3G) and other future systems (Nakajima and Yamao, 2001) must provide transmission at a significantly higher data rates as well as a lower tariff per transmitted bit than current second generation (2G) systems. It should be noted that wireless technology is becoming more familiar and common for society. They will also need to address the issue of how many mobiles can be simultaneously served with the infrastructure that the provider has in place. It is clear that we will witness also newer modes and improved services of communication due to progress in radio technology. The requirements on data rate, link reliability, spectral efficiency and mobility in a fading environment cannot be met with conventional single antenna systems. Therefore, antenna arrays may be used at least on the base station (BS) side to improve performance. Indeed, the new Chinese time division duplex (TDD) system time-division synchronous CDMA (Yang et al., 2002) has been specifically designed for use with smart antenna arrays. Antenna arrays can be used to provide an increased antenna gain and/or an increased diversity gain for each user in the system. At the same time, less interference is received from the other directions (on the uplink) or transmitted to other mobiles (on the downlink). Hence, more users can be accommodated by the system and a corresponding capacity increase is achieved (Thompson et al., 1996; Grant et al., 1998). Dual antenna arrays, one at each end of the wireless link, introduce the multiple input-multiple 300
Smart Antennas for TDD-CDMA Systems
301
output (MIMO) channel. MIMO channels can permit spatial multiplexing which provides significant improvements in link capacity for one wireless link, compared to an equivalent single transmitter and receiver antenna system. This chapter provides an introduction to the topic of antenna array processing for use in the BSs of TDD code-division multiple access (CDMA) systems. Advantages and drawbacks are explained next. Section 11.2 introduces the essential and important channel modelling concepts of smart antennas. The channel capacity of a wireless system is then investigated in section 11.4 to introduce the concepts of antenna gain, diversity gain and interference suppression for smart antennas. Antenna array receiver algorithms for uplink reception in CDMA BSs are introduced and their performance compared in Section 5. Downlink algorithms for TDD CDMA systems are then described and simulation results presented in section 11.5. The final part of the chapter will look at the use of smart antennas in future wireless systems, describing the use of orthogonal frequency division multiplexing (OFDM) and spatial multiplexing techniques. Then a brief discussion and conclusions are presented to finish the chapter.
11.1.2 General comments on advantages and drawbacks of smart antenna systems Since the early days of wireless communications, simple dipole antennas have been used in mobile handsets, which radiate and receive in all directions. Omnidirectional strategies directly and adversely impact spectral efficiency, limiting frequency reuse. Base station antennas have up till now been omnidirectional or sectored. This can be regarded as a ’waste’ of power as most of it will be radiated in other directions than toward the user. In addition, the power radiated in other directions will be experienced as interference by other users. A smart antenna system combines an array of multiple antenna elements with a signal-processing capability to optimise its antenna pattern automatically in response to changes in its radio frequency (RF) signal environment. It changes its pattern dynamically to adjust to the noise, interference in the channel and mitigate multipath fading effects on the signal of interest. This can dramatically increase the performance characteristics (such as capacity) of a wireless system. Today, smart antennas are vital to high-capacity wireless systems. More specifically, the main features and advantages (Grant et al., 1998; Thompson et al., 1996; Winters, 1998; Winters et al., 1994; Naguib and Paulraj, 1995;
302
Next Generation Mobile Access Technologies
Veen and Buckley, 1988; Liu and Zoltowksi, 1997; Barrett and Arnott, 1994; Tsoulos et al., 1997; Paulraj and Papadias, 1997; Liberti and Rappaport, 1999) derived from using smart antennas in wireless communications are listed below: • Increase the number of active users for a given bit error rate (BER) quality threshold and thus capacity (spectral efficiency). • Improve the BER performance for a given number of users within cell. • Permit a less stringent form of receiver link power control while maintaining acceptable BER performance. • Due to high directivity the uplink range and coverage is increased. Lower power requirements at the handset also enable a longer battery life and smaller/lighter handset size. • Smart antennas can give wireless networks access to spatial information about the users and thus possibility for new services like position location (Liberti and Rappaport, 1999). Although there are several benefits of using smart antennas, there are also drawbacks (Grant et al., 1998; Thompson et al., 1996) and cost factors, which are listed below: • The hardware and software requirements increase as M demodulators are required for each user and as M increases. • The M receivers must be accurately synchronised in time to provide effective performance. • The computational complexity of array processing algorithms can be very large. • Physical limitations on the size of wireless smart antennas present a fundamental limit on performance. • Practical antenna arrays performance may be adversely affected by channel modelling errors, calibration errors, phase drift and noise which is correlated between antennas. Thus, smart antennas extract more capacity from current CDMA network resources and their implementation at BSs results in a more efficient network. Combined CDMA and smart antenna technologies are an advantageous option for future wireless communications. 11.2 Channel modelling issues Increased services and lower costs have resulted in an increased air time usage and number of subscribers. Since the radio (spectral) resources are
Smart Antennas for TDD-CDMA Systems
303
naturally limited, scarce and very expensive, the main issue for wireless networks is the complexity of communication protocols that support full user mobility. Mobile wireless communication is burdened with a very hostile propagation environment compared to fixed radio systems. The antenna height at a mobile handset is usually close to the ground. Hence, obstacles and reflecting surfaces in the vicinity of the antenna have a substantial influence on the characteristics of the propagation path. Moreover, the radio propagation characteristics (Jakes, 1974; Lee, 1989a; Feher, 1995; Proakis, 1995) and (Naguib, 1996) change from place to place and, if the user moves, from time to time.
11.2.1 Large-scale propagation effects There are two major propagation effects which vary slowly as the mobile moves through an environment: • Path loss: This is the received power falloff due to signal attenuation, which occurs as the propagation distance r increases between transmitter and receiver antennas. The path loss models simplify Maxwell’s equations and are given by the classical Friis formula (Feher, 1995), where the received power is proportional to r−η . The exact value of the path loss exponent η depends on the particular wireless environment and range from 2 to 5. • Shadowing: The random power fluctuation due to fixed and large-scale obstructions in the propagation path of the radio signal. It is usually modelled as a log-normal distribution. The standard deviation of the shadowing can range from 4 to 12dB for particular environments. Shadowing will adversely affect the cell coverage area and, to compensate its effects in CDMA wireless systems, transmitter power control may be useful.
11.2.2 Small-scale propagation effects Over small distances, multipath effects can give rise to fading which leads to significant variations in received signal power. Multipath is a condition where the transmitted radio signal is reflected by physical features/structures, creating multiple signal paths between the base station and the user terminal. The phase of each multipath is different and they can add constructively or destructively at the receiver. Indeed, moving the transmitter or receiver over a distance as small as half a wavelength (7.5cm for a carrier frequency
304
Next Generation Mobile Access Technologies
(a)
(b) Power(dB)
10 Profile 8 Power Density (dB)
0
6
4
2
0 -100
Time
(c)
-50
0 50 Doppler Frequency (Hz)
100
(d) 0
Delay Profile
Power Density
Power Density (dB)
-5
-10
2∆
-15
-20
θo
-25
Angle of Arrival
-30 0
1
2
3 4 Time Delay (microsec)
5
6
7
Fig. 11.1. (a) A typical plot of the variation of received power on a Rayleigh fading channel with time. (b) The power density spectrum plotted against Doppler frequency for the classical Doppler model with a maximum Doppler of 100Hz. (c) A typical power density delay profile, which decays exponentially with time delay. (d) The power density spectrum plotted against angle of arrival for the uniform scattering model.
of 2GHz) can completely change the phase values, resulting in dramatic changes to the received signal strength. This phenomenon is known as fading. These variations significantly degrade signal quality, leading to much higher error probabilities in data communication. In the absence of a line of sight propagation condition, the channel is usually modelled as a zero-mean complex-valued Gaussian process, the envelope of the signal at any instant is Rayleigh-distributed (Proakis, 1995). This leads to large variations in received power of as much as 30-40dB over time if there is movement of the transmitter or receiver. A typical plot of the power variation is shown in Figure 11.1(a). Multipath propagation gives rise to a number of dispersive phenomena, which must be properly accommodated in any receiver design. These are: • Doppler spread: When the mobile is in motion with speed v, individual propagation paths and radio signal at the receiver will experience a shift
Smart Antennas for TDD-CDMA Systems
305
in the frequency domain (also called Doppler shift). The magnitude of the shift depends on the angle θq between the direction of motion v and the that of the qth multipath. The Doppler shift is Δfq = λ cos θq , where λ is the RF carrier wavelength. It is often assumed that many reflected waves arrive at the mobile each with its own random angle of arrival (AOA) which is uniformly distributed within [0, 2π], independently of other waves. This leads to the classical U-shaped Doppler power spectrum of the received signal (Jakes, 1974; Lee, 1989a; Feher, 1995; Thompson et al., 1996; Paulraj and Papadias, 1997; Liberti and Rappaport, 1999; van Rooyen et al., 2000) which is shown in Figure 11.1(b) for a maximum Doppler frequency of 100 Hz. The Doppler spread causes time selective fading, since the channel characteristics varies with time. Any coherent communications system must track these variations effectively in order to provide a reliable system. A useful measure of channel variation is the coherence time of the channel which can be defined as 1/(10 ∗ Δfm ), where Δfm = v/λ is the maximum Doppler frequency. The coherence time represents the time period for which the channel behaviour is essentially unchanged. • Delay spread: Each multipath component associated propagation delay. The impulse response of a wireless channel is not a single impulse but the sum of many, each with a different amplitude, phase and delay. The amount of delay dispersion is usually measured by the root-mean-square (rms) delay spread Trms , which is the standard deviation of all the different multipath delays. Wireless signals of different bandwidths B will experience different degrees of fading. This can be related to the channel’s coherence bandwidth Δ(f )c , which is the frequency range over which the fading amplitude is unchanged. It is proportional to the inverse of Trms . Frequency selective fading, which is modelled as tapped delay line, occurs when B > Δ(f )c and manifests itself in a received CDMA signal as interchip interference (ICI). An exponentially decaying power profile with time delay is widely used to model frequency selective fading – one example is shown in Figure 11.1(c). • Angle spread: Each multipath is also associated with a unique angle-ofarrival at the receiver. These angles will generally be different, giving rise to angle spread effects. The angle spread (AS) Δ is typically measured as the standard deviation of angle around the mean angle θ0 . It is a function of BS or terminal location, distance and environment. The AS gives rise to space selective fading. The range of space for which the fading remains invariant is called the coherence distance, denoted by Δ(d)c , and is inversely related to the AS. As a result, two antennas spaced by
306
Next Generation Mobile Access Technologies
more than the Δ(d)c tend to experience uncorrelated fading. The angle spread can be large when measured for indoor communications or at a mobile terminal in a cluttered outdoor environment. For such cases, the coherence distance is only a fraction of a wavelength. However, for base station antennas located above rooftops in an uncluttered site, the coherence distance is several wavelengths. The distribution of multipath with angle is sometimes modelled as a uniform distribution and this is shown in Figure. 11.1(d). In order to simulate angle spread effects properly with antenna arrays, a model of radio propagation at an antenna array is required. This will now be introduced.
11.2.3 The antenna array steering vector Consider an M -element uniformly linear array (ULA) of antennas, shown in Figure 11.2. The elements are equally spaced by a distance d, and a plane wave arrives at the array from a direction θ off the array broadside. The angle θ is the AOA of a given multipath ray. The plane wavefront of the ray at the first element should propagate through a distance d sin θ to arrive at the second element and so on. For a system whose data bandwidth is much smaller than the carrier frequency, the effect of the time delay on the signal can be represented by a phase shift. A plane wave of wavelength λ arriving at an angle θ from broadside experiences a phase difference of (2πd sin θ) /λ. For a ULA, the steering vector is given by: 8 & 8%T $ & √ 2π √ 2π , a(θ) = 1, exp − −1 d sin θ , ..., exp − −1 (M − 1)d sin θ λ λ (11.1) where T denotes the vector transpose operation. Equation (11.1) describes the spatial response of the array to a waveform impinging from direction θ. The steering vector depends on the AOA but does not depend on time unless the transmitter or receiver is moving. In general, however, the steering vector is also a function of the individual element response, the array geometry and signal frequency. The collection of steering vectors for all angles and frequencies is referred to as the array manifold.
11.2.4 Complete uplink channel model The typical low-pass model for the RF channel between a single antenna mobile and the M -element array at a base station consists of a L tap delay
Smart Antennas for TDD-CDMA Systems
307
Uniform Linear Antenna Array o
0
θq d .... Antenna 1 Antenna 2 Antenna M
Fig. 11.2. A diagram of a uniform linear array of spacing d, showing a plane wave impinging from angle θ.
line (Proakis, 1995; Thompson et al., 1996; Grant et al., 1998), where L is the number of resolved multipath components. The M × 1 channel vector for the kth user may be written in the following form (Thompson et al., 1996): hk (t) =
L−1 l=0
hk,l (t − τk − lTc )δ(t − τk − lTc ).
(11.2)
The scalar τk denotes the propagation delay of the RF channel for the kth user and δ(t) is the Dirac delta function. The quantity Tc is the resolvable delay between different multipaths, which equals the chip period of the CDMA signal. The mth entry of the vector hk,l (t) is the complex scalar, representing the amplitude and phase of the channel between the mobile and the mth antenna element of the receiver for the l th channel tap. The multipath tap vectors hk,l (t) are typically modelled as the sum of a large number of multipath components that cannot be separately resolved. A reasonable model for the Rayleigh fading channel that results is Raleigh’s model (Grant et al., 1998):
hk,l (t) =
Q √ (rk−η ζk ) αq exp { −1(φq + 2πνq t)}a(θq ),
(11.3)
q=1
where rk−η represents the pathloss due to the mobile-base station distance rk and ζ denotes a log-normal variable modelling shadowing. The received waveform is made up of Q multipath rays with amplitudes αq , phase φq , Doppler frequencies νq and AOA θq . For a given angle spread Δ, the uniform AS model means that all signals from the mobile arrive at the BS antenna array uniformly within [θ − Δ/2, θ + Δ/2]. Suppose the BS uses an ULA antenna array of M elements to receive
308
Next Generation Mobile Access Technologies
signals from K users. The M × 1 antenna array vector output at time t is: x(t) =
K L−1 Pk hk,l bk (t − τk − lTc )ck (t − τk − lTc ) + n(t),
(11.4)
k=1 l=0
where bk (t) ∈ {1, −1} is the kth mobile information bit sequence with a bit period of Tb and ck (t) ∈ {1, −1} is the pseudo noise (PN) code of the k th user with a chip period of Tc . The processing gain of this system is thus G = Tb /Tc . The scalar Pk is the transmitter power for the kth user. The vector n(t) contains spatially and temporally independent and identically distributed (i.i.d.) additive white Gaussian noise of zero mean and variance σ2.
11.3 Channel capacity issues: information-theoretical background of smart antennas Smart antennas at the receiver or/and the transmitter are currently one of the hottest topics in wireless communications, as they are a key technology for increasing network capacity. This section describes the informationtheoretical background of a wireless communication system using smart antennas and shows different ways in which the potential capacity of a network can be increased.
11.3.1 Shannon’s capacity equation The Shannon capacity C (Shannon, 1949; Wozencraft and Jacobs, 1993) is the maximum amount of information which could be transmitted error free in unit time over a specific channel. For a single input–single output (SISO) flat fading channel we can write the Shannon capacity as:
P |h1 |2 [bits/sec], C1 = B log2 1 + σ2 where h1 is the channel amplitude, B is the system bandwidth, and P is the transmitter power. In a fading environment, h1 is a random variable and hence so is the channel capacity C1 . One useful measure of system performance in this case is outage capacity. This is the capacity that can be attained in a given system for a certain percentage of the channel ensemble, e.g. 10% outage capacity represents the capacity that is attained or exceeded for 90% of channel realisations.
Smart Antennas for TDD-CDMA Systems
309
11.3.2 Capacity increase with antenna arrays The single input–multiple output (SIMO) channel capacity with an M antenna receiver is given by: M 2 P |h | m m=1 . C2 = B log2 1 + σ2 In this case, hm is the channel from the transmitter to the mth receiver antenna. Two extreme channel conditions will be discussed. • No angular spread – If all the multipath components that comprise the channel arrive from one direction (Δ = 0), the channel vector h = [h1 h2 · · · hM ] is proportional to a steering vector a(θ), where θ is the angle of arrival. Thus, |hm |2 = |1 |2 , so C2 can be written as:
P M |h1 |2 . C2 = B log2 1 + σ2 By increasing the number of antennas from 1 to M , it can be seen that the signal-to-noise ratio (SNR) (P |h1 |2 /σ 2 ) has been multiplied by M , which is often called the array gain. This factor represents the gain over noise or omnidirectional interference through the use of multiple antennas. • Large angular spread: If the multipath components arrive from a range of different angles and Δ is large, the channel vector h equals the sum of lots of different steering vectors, as seen in equation (11.2). In this case, the channel coefficients hm will become statistically independent of each another. This means that if one channel value h1 is small, it is unlikely that another coefficient, e.g. h2 , will be small in amplitude. This means that the likelihood that all M channel coefficients are small is much less than for the case of no angular spread (where all the channel coefficients had the same amplitude value). This improvement in channel reliability is often called diversity gain. The idea of receive diversity was introduced by the Dutch engineer A. (Haas, 1927), with the first implementation being reported in 1927 for short wave radio communication. The mean or expected value of the power for each antenna’s channel, denoted E[|hm |2 ], is the same value regardless of m. Therefore, the mean SNR of the M antenna array is again M times that of one antenna, due to the array gain. Figure 11.3(a) shows capacity results comparing the 10% outage channel capacity of the no angle spread (NAS) and large angle spread (LAS) cases for bandwidth B = 1 Hz. The channel capacity for the NAS case increases by
310
Next Generation Mobile Access Technologies Channel Capacity (in bits/sec) for 10% Outage
(a) 14
1 TX 1 RX NAS: 1 TX 2 RX 1 TX 4 RX 1 TX 8 RX LAS: 1 TX 2 RX 1 TX 4 RX 1 TX 8 RX
12 10 8 6 4 2 0 0
5
10 15 20 Signal-to-Noise Ratio (dB)
25
30
10 15 20 Signal-to-Noise Ratio (dB)
25
30
Channel Capacity (in bits/sec) for 10% Outage
(b) 90
1 TX 1 RX 2 TX 2 RX 4 TX 4 RX 8 TX 8 RX
80 70 60 50 40 30 20 10 0 0
5
Fig. 11.3. The 10% outage channel capacity for (a) a single transmit multiple receive antenna scenario with no angle spread (NAS) and large angle spread (LAS); (b) a multiple transmit multiple receive antenna scenario with K = M .
approximately 1 bit/sec for each doubling of the receiver array size, because of the base 2 logarithm. The results for LAS show significant increases in outage capacity due to the improved diversity gain in this case. However, it can be seen that diversity gain starts to saturate beyond M = 4 receiver antennas and the capacity for M = 8 is only about 1 bit/sec higher than for M = 4.
11.3.3 Multi-user capacity increase These results may be extended to consider the sum capacity of K users transmitting to an M antenna receiver. This is a MIMO system with K inputs and M outputs. The MIMO channel that is formed for a flat fading
Smart Antennas for TDD-CDMA Systems
311
scenario can be written as a M × K matrix H, whose mth column and kth row entry Hmk represents the channel amplitude and phase from transmitter k to receiver antenna m. If the channel is known to the receiver, the capacity of a flat fading MIMO channel is given by (Foschini, 1996; Foschini and Gans, 1998) and (Telatar, 1999): &
8 P , C3 = B log2 det IM + 2 HH∗ σ where ‘det’ is the matrix determinant, IM is the M × M identity matrix and H ∗ is the complex conjugate transpose. The value of C3 can be upperbounded as follows: K 2 P M |H | mk m=1 . log2 1 + C3 ≤ B σ2 k=1
This can be interpreted as K parallel channels being formed between the K transmitters and the M receivers. Equality in the above equation holds when the K channels do not interfere, so that they are orthogonal. Unfortunately, the maximum possible number of orthogonal spatial channels is M , so that if K > M , there will be interference between users which will limit the achievable capacity C3 . The level of interference that arises in practice will be determined by the spatial locations of users. Significant interference may be caused from one user to another, particularly if their angles of arrival are similar. One way to support more simultaneous users is to use DS-CDMA spreading, though this requires bandwidth expansion proportional to the processing gain G. Figure 11.3(b) shows the capacity increases available when the number of transmitters K = M . A LAS scenario is assumed here. It can be seen that large increases in outage capacity can be attained due to the use of both multiple transmit and multiple receive antennas. In practice, these gains will be offset by combining losses in the receiver algorithm and the clustering of users within one part of the cell (Thompson et al., 1999b). 11.4 Uplink processing algorithms We next turn to the details of how the required combating of multipath and multi-user interference can be achieved. In order to understand these techniques it is best to consider time domain processing first, which is usually implemented in standard receivers by means of a RAKE receiver. We will then discuss the different receiver structures and algorithms for uplink processing, distinguishing between fixed beam antennas/antenna selection,
312
Next Generation Mobile Access Technologies
space-time 2D RAKE receiver (space-time processing), MMSE combining and combined array processing/multi-user detection (MUD).
11.4.1 Time domain RAKE receivers A spread spectrum signal can recover most of the energy from multipath, reduces narrowband interference and mitigate ICI. A popular single-user CDMA receiver is the RAKE combiner or receiver (Proakis, 1995). Consider a single antenna (M = 1) RAKE receiver which has L branches that synchronise to L different multipath components. This receiver uses multiple correlators, one for each path, whose lth output for the ith data bit can be calculated as: ! iTb 1 ck (t)x1 (t + (l − 1)Tc )dt. (11.5) y(l, i) = Tb (i−1)Tb The quantity x1 (t) denotes the scalar received signal, as defined in (11.4). The outputs y(l, i) of the correlators (called fingers) are weighted and then combined into a single output to minimise the SNR. The nth transmitted data symbol bk (i) can thus be estimated as: bˆk (i) =
L
y(l, i)w(l).
(11.6)
l=1
If the lth correlator weight w(l) is set to the complex conjugate of the corresponding channel response hk,l (t), we obtain a combining filter which is a matched filter to the spreading code plus multipath channel. A diagram showing the structure of the RAKE filter receiver is shown in Figure 11.4(a). Provided that the multipath signals do not fade simultaneously, multipath signals can be utilised as an important form of diversity against Rayleigh fading. Since only the temporal structure of the multipath received signals is exploited efficiently by this RAKE receiver, it is referred to as a time-domain or one-dimensional 1D-RAKE receiver.
11.4.2 Space-time processing The signals from multiple receiver antennas can be processed in a similar general manner to the single antenna RAKE receiver. However, signal combining must now be performed in time delay and across receiver antennas. If the receiver now contains M antenna elements, the correlator output for the mth antenna and lth delay component can be written in a similar way
Smart Antennas for TDD-CDMA Systems
313
Fig. 11.4. (a) A RAKE combining receiver. (b) A space-time combiner. (c) One arrangement of fixed beam patterns to cover a 120◦ sector.
to (11.5): 1 ym (l, i) = Tb
!
iT b
(i−1)T b
ck (t)xm (t + (l − 1)Tc )dt.
The subscript notation ym denotes the fact that this signal is obtained from the mth receive antenna. The notation xm (t) denotes the mth entry of the vector x(t) in (11.4). An estimate of the nth transmitted symbol is then
314
Next Generation Mobile Access Technologies
obtained as: bˆk (i) =
L M
ym (l, i)w(l, m).
m=1 l=1
A diagram of this space-time combiner is shown in Figure 11.4(b). A number of different algorithms for choosing the weights w(l, m) will now be described. 11.4.2.1 Space domain fixed beam/antenna selection The fixed-beam (FB) method can be used with antenna array to form a set of overlapping beams patterns to provide coverage in a 120◦ sector. This is a simple but effective approach which is an extension of the conventional sectorisation method. The receiver contains a set of Z fixed steering vectors a(θz ) which can be set to cover the cell sector (Grant et al., 1998). A diagram showing one possible arrangement of four steering vector beam patterns (numbered 1-4) to cover a 120◦ sector is shown in Figure 11.4(c). The receiver operates by measuring the power from each steering vector separately for each multipath xk (l, i), perhaps averaged over multiple symbols. At the output, we have available Z RF-signals, one for each possible beam or direction. The simplest processing approach is to select the largest beam for combining along with the equivalent beam outputs from the other (L − 1) multipath delays for further processing by the standard 1D-RAKE receiver, (11.6). The receiver might employ a RAKE filter to combine the outputs from the selected fixed beams. Results for FB receivers suggest that the available diversity gain may be inferior to that obtained through using space diversity with widely spaced antennas. Therefore, a FB receiver will perform best when additional diversity, e.g. multipath diversity, is present (Grant et al., 1998). The system losses arise from a number of factors, including cusping loss, mismatch loss, beam-selection loss and path diversity loss. Moreover, the signal strength varies as the user moves through the sector. 11.4.2.2 Space-time 2D–RAKE receiver A more effective approach to dealing with these signals received at the individual array element outputs is to apply a separate spatial filter in turn to each set of samples collected across the array output, i.e. to the observed signal vectors. The receiver that efficiently resolves different multipath components in both space and time and combines them is referred to as the 2DRAKE receiver (Thompson et al., 1996; Grant et al., 1998). The 2D-RAKE
Smart Antennas for TDD-CDMA Systems
315
receiver exploits the spatial structure in the received multipath signal in addition to the time diversity in order to provide improved performance over the 1D-RAKE filter (Grant et al., 1998; Thompson et al., 1999a, 1998). A 2D-RAKE receiver utilises a total of M L correlators, one for each path and for each receiver antenna. As with the 1D-RAKE receiver, the correlator outputs are weighted and summed to form a single scalar decision variable. The weight w(l, m) for the mth antenna and lth multipath is proportional to the complex conjugate of the mth entry of hk,l (t). This 2D-RAKE filter is therefore a matched filter to the spreading code plus multipath channel as seen at each receiver antenna. The advantage of this receiver is that it can exploit both multipath and space diversity when present in the received signal. Unlike the fixed beams configuration, the antenna element spacing is not limited to half wavelengths and can be increased to exploit the fading decorrelation that arises due to AS in the environment. 11.4.2.3 MMSE combining One disadvantage of the 2D-RAKE receiver is that its weights optimise performance only in the presence of white noise interference. However, this receiver can degrade in the presence of ICI and co-channel interference (CCI), which arise due to the spread spectrum waveform. One improved approach is to design the combiner weights in the 2D-RAKE receiver using the linear minimum mean squared error (MMSE) linear criterion (Paulraj and Papadias, 1997), where the weighting coefficients are chosen during a training phase. It is a space-time linear filter whose weights are chosen to minimise the mean squared error between the symbol decision and the transmitted data of user and is hence superior to MRC in the presence of coloured interference. The optimum solution for the MMSE combiner can be found by defining the M × L vector y(i) as follows: y(i) = [y1 (1, i)y1 (2, i) · · · y1 (L, i)y2 (1, i) · · · yM (L, i)]T . This vector contains all the correlator outputs for all M antennas. The combiner weights can also be written as a vector: w = [w(1, 1)w(1, 2) . . . w(1, L)w(2, 1) . . . w(M, L)]T . Then the complex conjugate of the optimum weights are given by the Wiener Hopf equation: w∗ = R−1 yy ryd . The size M L × M L matrix Ryy is the mean covariance matrix of the vector y(i) and ryd is the mean cross-correlation vector of y(i) with bk (i). The
316
Next Generation Mobile Access Technologies
notation w∗ denotes the complex conjugate of w and R−1 yy denotes the matrix inverse of Ryy . 11.4.2.4 Combined array processing and multi-user detection A typical CDMA system will suffer from multiple access interference (MAI) arising at the base station receiver, limiting system capacity. One way to improve performance is to cancel MAI by MUD techniques (Moshavi, 1996). The optimum receiver for CDMA antenna arrays consists a receiver that combines multipath energy in space, followed by a Viterbi decoder, containing enough states to compensate for all the multipath time lags of all the CDMA users present on the same channel. In practice, this detector has exponentially increasing complexity with the number of users, so that simpler suboptimal detectors are likely to be preferred (Moshavi, 1996; HassellSweatman et al., 2000). Spatial beamforming at each time delay may thus be followed by one of the following MUD detectors:
• Decorrelating detector (DD): This detector takes all the correlator outputs for each user and applies a matrix transformation which sets all the MAI terms between each pair of users to zero. The drawback of this scheme is that background noise levels may be enhanced. • Successive interference cancellation (SIC): This detector orders users by their received power levels. Using this order of users, each user is decoded in turn and their transmitted data is estimated. Their received signal components are then reconstructed and subtracted from the received signal vector x(t). This should reduce the interference to other users, although if the received signal is detected wrongly the interference cancellation process will actually enhance the level of interference to other users. • Decision feedback detector (DFD): This detector combines ideas from the decorrelating detector and successive interference cancellation. Again users are ordered by their received power level. For each user, the decorrelating detector is used to minimise interference from users yet to be detected. Successive interference cancellation is used to remove interference from users whose signals have already been detected. Implementation details for these detectors and other algorithms can be found in (Moshavi, 1996) and (Hassell-Sweatman et al., 2000).
Smart Antennas for TDD-CDMA Systems 11
Coherent Combining Fixed Beam (Best) Fixed Beam (Worst) Fixed Beam (Ave)
10
SINR Gain (dB)
317
9
8
7
6
5 0
10
20 30 40 Angular Spread of Scatterers (deg)
50
60
Fig. 11.5. Performance comparison of fixed beams and coherent combining (used in 2D-RAKE receiver) in a single user scenario with no delay spread. The array size M = 8 and the maximum achievable SNR is 9dB.
11.4.3 Uplink algorithm comparisons In this section, simulation results will be presented to highlight some of the key advantages and drawbacks of the different algorithms. To begin, Figure 11.5 considers the SNR performance of the fixed beams and coherent combining, as used in the 2D-RAKE receiver. A single-user scenario with no multipath is considered and the maximum achievable SNR is 9dB. In the case of fixed beams, results are shown for the best case (user located in middle of one beam), the worst case (user located in the cusp between two beams) and the average performance over all positions. An M = 8 element ULA is considered with element spacing 0.5λ, so that the beamwidth of the array is approximately 120◦ /8 = 15◦ . The SNR performance is plotted as a function of the AS Δ, which varies from 0◦ to 60◦ . The results clearly show that, for small angular spread, fixed beams will usually perform well, except for the worst-case situation. On average the performance loss of fixed beams over coherent combining is only 1dB for small AS values. However, when the AS becomes larger than the array beamwidth, fixed beams begins to degrade significantly compared to coherent combining. For an AS = 60◦ , the performance loss of fixed beams is more than 3dB. The second set of results compares the SNR performance of the 2D-RAKE receiver and MMSE combining. In this case, a three-tap, Rayleigh-fading multipath channel is considered. The mean power of each tap is assumed to be the same and the AS of each tap is 0◦ . The AOAs of the second and third paths are randomly distributed over a range of ±5◦ relative to that of the first one. In the simulation, K users are randomly distributed in AOA
318
Next Generation Mobile Access Technologies
over a single cell sector of 120◦ , and the average SNR is measured for 1000 different sets of user positions. The results plotting SNR vs number of users K are shown in Figure 11.6. Part (a) shows results for M = 2, 4 and 8 antennas with P G = 8 and part (b) the same antenna configurations for P G = 16. It is assumed that each user requires an SNR of 7dB in order to obtain adequate performance. The results in 11.6(a) show that antenna arrays can significantly improve performance with PG=8. Increasing the array size from M = 2 to M = 8 allows the number of supported users at an SNR of 7 dB to increase from 1 to 4 with the 2D-RAKE receiver. Using the MMSE receiver with M = 8 can double capacity to 8 users. In 11.6(b), for P G = 16, the results again show significant improvements in capacity as M increases. The number of supported users for an SNR of 7dB increases from 3 with M = 2 to 6 for M = 4 and 11 for M = 8. This time the improvement with MMSE combining is not so large: using MMSE combining increases the number of supported users to 14, an increase of about 25% over 2D-RAKE combining. The reason for this is that with the larger PG value of 16, the interference contribution of one user reduces and suppressing the effect of one interferer does not improve the SNR as much as in the case when P G = 8. The final set of results considers the performance of antenna array processing combined with MUD. The results again consider a multiuser scenario with a single cell sector of 120◦ , but this time perfect power control of each user’s transmission is assumed so that Rayleigh fading effects can be neglected. One channel tap L = 1 is simulated and the PG value is 31 in this case. The background noise level was set to be 7dB less than the power of each user’s received signal. The results are plotted as average BER vs number of users in Figure 11.7, with part (a) assuming one receiver antenna (M = 1) and part (b) assuming M = 4. In each case, results are shown for coherent combining (CC), and the three MUD techniques described above, namely, DD, SIC and DFD. The results in Figure 11.7(a) demonstrate that considerable improvements in BER performance are achievable through the use of MUD techniques. The best performance in both part (a) and (b) is achieved through the use of the SIC technique. In Figure 11.7(b), it can be seen that for M = 4 receiver antennas, the BER performance is virtually independent of the number of users for the SIC algorithm.
Smart Antennas for TDD-CDMA Systems
319
(a) 14
2D-RAKE MMSE
12
Output SINR (dB)
10 M=4
8 6
M=8
4 M=2
2 0 0
4
8
12 16 20 Number of Active Users P
24
28
32
28
32
(b) 14
2D-RAKE MMSE
12
Output SINR (dB)
10 8 6 M=4
4
M=8
M=2 2 0 0
4
8
12 16 20 Number of Active Users P
24
Fig. 11.6. Performance comparison of 2D-RAKE receiver and MMSE combining for M = 2, 4 or 8 receiver antennas, with an L = 3 tap multipath channel for each user. The results in part (a) are for a progressing gain P G = 8 and part (b) are for P G = 16.
11.5 Downlink processing algorithms In order for antenna arrays to improve the capacity of a wireless system, it is important that performance gains can be provided on both the uplink and downlink. Downlink transmission is another very important issue for smart antenna operation. It is particularly crucial for the next generation of wireless systems in which wireless Internet and multimedia services are required. These services are likely to be asymmetrical in data requirements, with most of the data being transmitted on the downlink to mobile terminals. This section will investigate the performance benefits of using a smart antenna at the base station to transmit to mobile terminals. This section will begin by considering how TDD transmissions are beneficial for the use of
320
Next Generation Mobile Access Technologies (a) 0.1 CC DD DFD SIC Bit error ratio
0.01
0.001
0.0001
0
5
10
15 20 Number of users
25
30
35
10
15 20 Number of users
25
30
35
(b) 0.1 CC DD DFD SIC Bit error ratio
0.01
0.001
0.0001
0
5
Fig. 11.7. Performance of multi-user detection for a CDMA system with perfect power control and P G = 31. The results show bit error ratio (BER) vs number of users for (a) M = 1 receiver antenna and (b) M = 4 receiver antennas.
downlink antenna array operation. Then a number of candidate algorithms are described and their performance compared.
11.5.1 Exploiting uplink channel information for the downlink in TDD The major challenge in transmit antenna array processing is the estimation of the downlink channel coefficients. In frequency-division duplex (FDD) systems, the uplink and downlink transmissions are conducted at different
Smart Antennas for TDD-CDMA Systems
321
carrier frequencies. This means that the instantaneous channel behaviour on these two carrier frequencies are different. Channel measurements obtained on the uplink are unlikely to be useful for determining current downlink channel coefficients. However, in TDD mode the uplink and downlink channels can be considered reciprocal (because the frequency used is the same in both directions). The only requirements are that the same antenna array is used at the BS for reception and transmission and that the channel characteristics have not changed considerably between the receive (Rx ) and transmit (Tx ) slots. This will hold provided that the duplexing time, which is the time delay between Rx and Tx slots is small compared to the coherence time of the channel. This means that channel estimates of the L uplink channel taps hk,l (t) also apply to the L downlink channel taps from the M transmitter antennas to the mobile terminal antenna of the ith user. This property can be exploited to design simple, convenient downlink transmission algorithms for TDD systems. A number of candidate techniques for transmitting to a single user are now described, which are classified in transmit diversity and transmit beamforming techniques. Multi-user downlink transmission techniques are considered in detail in Chapter 10 of this book. 11.5.1.1 Downlink transmit diversity As for the uplink, downlink performance can be improved significantly through the use of transmit diversity techniques. Finding effective transmit diversity techniques is a more challenging problem than receive diversity, since there is now only one receive antenna. Despite this, a number of effective transmit diversity techniques have been found, see (Narula et al., 1999; Thompson et al., 2000) and the references therein. A comprehensive performance comparison of different transmit diversity schemes is provided in (Thompson et al., 2000). In this section, attention will be restricted to space-time transmit diversity and antenna selection diversity. Space-time transmit diversity (STTD) – (Alamouti, 1998) proposed a simple and elegant method to achieve diversity gain for systems with two transmit antennas. This is a space-time block codes (STBC) scheme which provides a performance gain similar to that obtained by using two receive diversity antennas. This is an open-loop scheme, since it does not require knowledge of the downlink channel coefficients to operate properly. In Alamouti’s scheme, two data symbols are sent by the two antennas in two symbol periods, as follows: d(i) = [b(i)b(i + 1)]T d(i + 1) = [b∗ (i + 1) − b∗ (i)]T ,
322
Next Generation Mobile Access Technologies
where d(i) denotes the vector of the ith data symbols for transmission. It turns out that the matrix [d(i)d(i + 1)] is an orthogonal design (Tarokh et al., 1999), so that the data symbols b(i) and b(i + 1) are linearly separable at the receiver. The same user specific spreading code may be applied at both transmitters, without affecting the STBC scheme. Since each symbol is sent from both transmit antennas, the diversity gain is equivalent to that obtained from using two receiver antennas. This scheme can be extended to arbitrary numbers of transmit antennas M , provided the modulation used is real, e.g. BPSK. For complex modulation schemes, such as Quadrature Phase Shift Keying (QPSK), there are no known STBC schemes which can use more than M = 2 antennas without a reduction in data rate or a reduction in diversity gain (Alamouti, 1998; Tarokh et al., 1999). This technique has been incorporated in the European 3G standard (3GPP, TSG, RAN, 2001). Antenna selection diversity (ASD) – This is a simple scheme where 1 out of the M possible transmit antennas is selected for transmitting to the mobile terminal. The antenna selection can be based on channel measurements from the last uplink transmission slot. Whichever receiver antenna achieved the highest SNR for the uplink is selected to transmit to the terminal on the downlink. This exploits the principle of reciprocity for the channel in a TDD system. 11.5.1.2 Downlink beamforming techniques The transmit diversity techniques described above do not fully exploit knowledge of the downlink channel coefficients which can be obtained from uplink channel measurements. Two techniques are now considered which use this information to exploit array gain as well as diversity gain. Maximum SNR Beamforming – This approach tries to design a size M weight vector v to be used at the transmitter, in order to maximise the overall SNR at the receiver. The baseband signal for transmission at the mth antenna at time t for user k is: sm (t) = v(m)bk (t)ck (t). where v(m) denotes the mth transmit antenna weight, bk (t) is the data sequence for transmission and ck (t) denotes the user-specific spreading code. In a flat fading channel, with L = 1, the optimum solution for v(m) is to choose weights matched to the channel coefficients. So v is set to the complex conjugate of the corresponding channel vector hk,1 . In the case of a frequency selective channel, there are L different paths to consider in the choice of v. Assuming that the mobile terminal contains a RAKE receiver
Smart Antennas for TDD-CDMA Systems
323
to combine the L different delay components, the optimum solution for v is now: v = arg max{vH (hk,1 hk,1 + hk,2 hk,2 + · · · + hk,L hk,L )v/(vH v)} = arg max{vH Gv/(vH v)},
where G is the M × M covariance matrix of the L channel vectors hk,L . The notation vH denotes the complex conjugate transpose of v. Solving the equation yields an eigenvalue problem, where v is the eigenvector corresponding to the largest eigenvalue of G. This approach is not dissimilar to closed-loop beamforming techniques for FDD systems. However, in FDD mode, a closed-loop approach is required where the mobile measures the channel coefficients, quantises them and then transmits them to the base station (Gerlach and Paulraj, 1994). PreRAKE beamformer – This method is a modified version of the maximum SNR beamformer for dealing with frequency selective channels (Esmailzadeh and Nakagawa, 1993d). Since downlink channel estimates are available to the transmitter, it can artificially pre-distort the signals to match the channel state. The transmitted signal consists of the sum of L delayed versions of the data signal, multiplied by the complex conjugate of the L channel vectors {hk,l }. The radio channel itself acts in effect as a RAKE receiver, so that the signal observed at the receiver has already been coherently combined. The baseband signal for transmission at the mth antenna at time t for user i is thus of the form: sm (t) =
L−1 l=0
v(l, m)bk (t − lTc )ck (t − lTc ).
The transmitter now uses of a set of M L weights, where the scalar v(l, m) is the weight for the lth delayed tap and mth transmit antenna. Define the vector v(l) = [v(l, 1), v(l, 2), . . . , v(l, M )]T . The value of v(l) is proportional to the complex conjugate of the vector hk,L−1−l (t). The scalar received signal at the mobile is the convolution of sm (t) and hk (t) from (11.3): r(t) =
L−1 L−1 p=0 l=0
v(l)hk,p (t − τk − pTc )
×bk (t − τk − (l + p)Tc )ck (t − τk − (l + p)Tc ) + n(t). where n(t) represents additive white Gaussian noise. If we restrict the terms in r(t) to only those with overall delay τk + (L − 1)Tc , so p = L − 1 − l, we
324
Next Generation Mobile Access Technologies
have:
r(t, τk + (L − 1)Tc ) =
L−1 l=0
v(l)hk,L−1−l (t − τk − (L − 1 − l)Tc )
×bk (t − τk − (L − 1)Tc )ck (t − τk − (L − 1)Tc ).
Since v(l) = hk,L−1−l (t), this delayed component consists of the coherently combined sum of all L multipath components. Thus the mobile terminal only requires a single correlator, rather than a RAKE receiver, which is locked to the delay τk + (L − 1)Tc .
11.5.2 Simulation results and comparisons The performance of the four downlink transmission techniques described above will be investigated and compared with that for a baseline single antenna transmitter case. The simulations presented here consider an uncoded transmitter with M transmitter antennas, which uses BPSK modulation. The results are plotted as the required SNR at the receiver to achieve a BER of 0.01, with the total transmit power normalised to be M . The simulations assume ideal CDMA code properties, so that there is no ICI. The first simulations consider the performance of the different algorithms for an L = 2 tap Rayleigh-fading multipath channel, with both taps having equal average power. The required SNR is plotted against number of transmit antennas in Figure 11.8. The results show that the pre-RAKE and maximum SNR approaches both perform well. However, as M increases, the performance of antenna selection and particularly STTD degrades in comparison to these two techniques. The second set of results in Figure 11.9 show required SNR as a function of the number of equal power Rayleigh-fading multipath channel taps L. This time the number of antennas M is set to be 3. The results again show that the pre-RAKE technique performs the best among the algorithms. The maximum SNR technique degrades slightly as the number of paths L increases, because it has to try to match all the channel vectors hk,l using only one transmit vector v. The performance of the STTD and antenna selection schemes rapidly flattens out, as the diversity gain saturates. Indeed, for L = 6, the performance of the single antenna scheme is only about 0.75 dB worse than STTD.
Required Transmit Power for 1% BER
Smart Antennas for TDD-CDMA Systems One Antenna STTD Antenna Selection
10
325
Max SNR Beamformer Pre-RAKE
5
0
-5
2
4
6 8 10 12 Number of Transmit Antennas
14
16
Fig. 11.8. Results plotting required SNR for a BER of 0.01 vs the number of transmit antennas M . The number of channel taps L = 2.
Required Transmit Power for 1% BER
14 One Antenna STTD Antenna Selection Max SNR Beamformer Pre-RAKE
12 10 8 6 4 2 0 -2 1
2
3 4 Number of Channel Taps
5
6
Fig. 11.9. Results plotting required SNR for a BER of 0.01 vs the number of channel taps L. The number of antennas M = 3.
11.6 Future TDD wireless systems There is currently great interest in the use of wireless local area networks (WLAN), which can provide high data rate services over short distances. Examples of current systems are the European HIPERLAN 2 system and the American IEEE 802.11a standard. These systems use higher order modulation schemes to provide data rates up to 54Mbits/s in a 20MHz radio channel. In order to minimise the effects of multipath fading, these systems use orthogonal frequency division multiplexing (OFDM). This is a multiple carrier transmission technique, where different data bits are transmitted in parallel using different narrowband channels located at different frequen-
326
Next Generation Mobile Access Technologies
cies. In order to further improve spectral efficiency, OFDM signalling may be combined with a MIMO channel configuration. Unlike the MIMO configuration in section 11.3.3, the transmit antennas belong to a single user. This permits that user to exploit multiple spatial channels to increase the achievable data rate with a fixed bandwidth radio channel. This section will discuss both OFDM and single-user MIMO systems and will describe the operation of future wireless TDD systems employing both technologies.
11.6.1 Introduction to OFDM Since the 1950s OFDM (Chang, 1966), which is a multicarrier (MC) technique, has been a well known approach for data transmission avoiding intersymbol-interference (ISI). The idea of OFDM is to divide the available spectrum into J equally spaced narrow band subchannels. The frequency response of the subchannels are overlapping and mutually orthogonal (independent of each other), hence the name OFDM. The data bits for transmission are sent in parallel over frequency non-selective channels using many SCs where each subcarrier operates at low data rate. The main advantage of OFDM is that it permits data transmission over a frequency selective fading channel, with only simple equalisation required at the receiver. The main disadvantages of OFDM signalling is the high peak-to-average power ratio means that linear amplifiers must be used at the transmitter. Second, there is a requirement for precise time and frequency synchronisation at the receiver to maintain orthogonality of the OFDM carriers and avoid intercarrier interference. However, these effects can be minimised with proper transceiver design. Due to the high data rates of modern multimedia applications and the availability of efficient fast Fourier transform (FFT) algorithms, OFDM has become a very important wireless transmission technique. A diagram of an OFDM transmitter is shown in Figure 11.10(a). The data symbols for transmission are split between the OFDM carriers, using a serial-to-parallel device. In order to determine the equivalent time domain signal for transmission, the J carriers’ signals are passed through an inverse FFT (IFFT). This signal is then transmitted over a radio channel, which may potentially be frequency selective. A typical OFDM receiver structure is shown in Figure 11.10(b). The first job of the receiver is to remove a prefix from the received signal. As shown in Figure 11.10(c), each OFDM signal is transmitted with a cyclic extension, which is a copy of the start of the preceding OFDM symbol. At the receiver, the start of the OFDM symbol is removed. As shown in Figure 11.10(c),
Smart Antennas for TDD-CDMA Systems
327
this will remove ISI from the remaining signal, provided the cyclic extension was long enough. The remaining signal is a delayed version of the original OFDM symbol. However, using the properties of the Fourier transform, a delay in the time domain only causes a phase shift to be applied to the different frequency carriers. Thus, the procedure of processing the carriers is essentially unaffected. Because each carrier observes a flat fading channel, a simple one tap equaliser {wj }, as shown in Figure 11.10(c), is sufficient to remove the amplitude and phase shift caused by each carrier’s channel. 11.6.2 Combining OFDM with spatial multiplexing architectures An important approach to improving the capacity of an OFDM system is to use it in a multiple input–multiple output configuration. In this case, the multiple antennas belong to one user, so that the spectral efficiency of their transmission may be increased by spatial multiplexing. The fact that each OFDM carrier observes a flat fading channel means that each carrier can be separately processed at the receiver, leading to simpler algorithmic processing than for an equivalent CDMA system, which may be corrupted by ICI and ISI. The remainder of this section will describe candidate transceiver configurations and compare their channel capacity performance. A similar model for the operation of this spatial multiplexing system may be used to that employed in section 11.3.3. This time, as shown in Figure 11.11 there are N transmit antennas used at one transmitter site to transmit to a receiver array of M antennas, with M ≥ N . The use of OFDM at the transmitter ensures that the receiver observes J flat fading channels. The received signal for carrier j may be written as: x(t, j) = H(j)b(t, j) + n(t, j).
(11.7)
The vector x(t, j) is the received signal at time t on carrier j. The M × N matrix H(j) represents all N M channels from each transmit antenna to each receive antenna. Its mth column and nth row entry H(m, n) represents the channel amplitude and phase from transmitter n to receiver antenna m. The size N vector b(t, j) contains the N baseband signals for transmission by the N transmit antennas. Finally, the size M vector contains spatially and temporally independent and identically distributed (i.i.d.) additive white Gaussian noise. Usually this type of MIMO system must obey a total transmit power constraint, namely that the value of b(t, j)H b(t, j) must be no more than a fixed transmit power P . This leads to slightly different capacity equations than were obtained in section 11.3.3. The performance of a spatial multiplexing system depends on the struc-
328
Next Generation Mobile Access Technologies
Fig. 11.10. Diagram of (a) OFDM transmitter and (b) OFDM receiver. (c) Illustration of the removal of the cyclic prefix from an OFDM signal.
ture of the matrix H(j). Consider the singular value decomposition (SVD) of the channel matrix H(j). Using the SVD, H(j) may be decomposed as: H(j) = U(j)D(j)V(j)H . The matrix M × M matrix U(j) contains the left column eigenvectors of
Spatial Processing
Smart Antennas for TDD-CDMA Systems
Data Coding/ Multiplexing
N Transmit Antennas
Multipath Channel
329
Decoding Data
M Receive Antennas
Fig. 11.11. Diagram of a typical spatial multiplexing system with N transmit antennas and M receive antennas.
H(j), which are the eigenvectors of the Hermitian matrix H(j)H(j)H . Similarly, the N × N matrix V(j) contains the right column eigenvectors, obtained from the matrix H(j)H H(j). The M × N matrix D(j) is a diagonal matrix, whose diagonal entries are the square roots of the n eigenvalues {λn (j)} of H(j)H H(j) and of H(j)HH (j). The eigenvalues indicate the power transfer between different eigenmodes of the channel and the spatial multiplexing system will perform best when all eigenvalues {λn (j)} are large. The best performance is obtained when the entries of H(j) are all uncorrelated Rayleigh-fading channels. If there is correlation between the different channels, the performance degrades as the different transmit antennas become harder to separate spatially. The performance will be poorest when all the entries of H(j) are completely correlated. Then the apparent AOA of all antennas is the same and only one eigenvalue is non-zero. In this case, no spatial separation of the transmit antennas is possible. Channel correlation issues will not be considered further here, the ideal case of no fading correlation will be assumed for the simulation results presented later in this section. The action of the channel matrix H(j) in equation (11.7) will ensure that each receive antenna observes the superposition of all the transmit antenna signals b(j). Some candidate algorithms to separate the received signals are described below: • Eigenmode transmit/receive (ETR): The channel matrix H is assumed to be known at both transmitter and receiver (Raleigh and Cioffi, 1998; Telatar, 1999). In a TDD system, we have seen that this is a realistic assumption for the base station transmitting on the downlink, with uplink channel measurements available. Define the following transformations: ˜ j) = V(j)b(t, j) and n ˜ (t, j) = U(j)H n(t, j). ˜ (t, j) = U(j)H x(t, j), b(t, x
330
Next Generation Mobile Access Technologies
˜ (t, j) has the same distribuAs U(j) is an orthonormal transformation, n tion as n(t, j). Then, (11.7) becomes: ˜ j) + U(j)H n(t, j) = D(j)b(t, j) + n ˜ (t, j). ˜ (t, j) = U(j)H H(j)b(t, x Thus, by this procedure N orthogonal spatial channels have been formed. The channel capacity of the j channel of this system with H(j) fixed is: C4 (j) = B
N
log2 (1 + αn λn (j))
[bits/sec].
n=1
The scalar B is the bandwidth of the jth carrier. The coefficients αn may be optimised using the classical water filling technique (Raleigh and Cioffi, 1998; Telatar, 1999), so that: + αn = (μ − λ−1 n ) , where μ is set so that:
N
αn = P.
(11.8)
n=1
The function a+ is equal to zero if a < 0 and equal to a for a ≥ 0. The √ elements of b(t, j) may be scaled by the corresponding coefficient αn in order to maximise performance. • Maximum likelihood receiver (MLR): The channel matrix H(t, j) is known at the receiver only. In the absence of channel knowledge at the transmitter, it turns out that the values of αn in (11.8) should all be set to P/N (Telatar, 1999). The channel capacity for fixed H(j) is given by: C5 (j) = B log2 det(IM + (P/N )H(j)HH (j)). This result has been derived in (Telatar, 1999). The receiver will operate using the maximum likelihood principle. In this case, this means that the received signal x(t, j) for each symbol should be compared with every possible noise free received signal, which has the form H(j)b(t, j). Unfortunately, the complexity of this receiver increases exponentially with the number of transmitters N . • Linear decorrelator (LD): In this case, a separate beam pattern is set up for each transmit antenna, which nulls out the other N − 1 interfering antennas. This can be achieved by calculating the M × N pseudo-inverse matrix G(j) of the channel matrix H(j), so that G(j)H(j) = IN . This receiver is similar to the decorrelating detector MUD technique described in section 11.4.2.4. The channel capacity has been derived for that context
Smart Antennas for TDD-CDMA Systems
331
in equation (72) of (Verdu and (Shitz), 1999), so the result here is: C6 = B
N
n=1
log2 1 +
P N (G(j)GH (j))nn
,
where (G(j)GH (j))nn denotes the nth diagonal entry of the matrix (G(j)GH (j)). • BLAST receiver: This approach is a simplified implementation of the optimum receiver (Foschini, 1996). In this case, the transmit antennas are demodulated serially in the decreasing order N → 1. The beam pattern for the nth antenna is set to null out waveforms for antennas 1, 2, . . . , n − 1. By analogy with the LD receiver, this is achieved by calculating the size n × M pseudo-inverse matrix G(n, j) of the M × n matrix H(n, j), which contains only the first n columns of H. Interference from antennas (n + 1), . . . , N , which have already been demodulated, is removed by subtractive cancellation. The channel capacity of this receiver, averaged over the channel fading, has been derived in (Foschini, 1996) under the assumption of ideal interference cancellation. Here, the capacity result of equation (95) in (Verdu and (Shitz), 1999) for a BLAST receiver with a fixed channel matrix H(j) is quoted: C7 = B
N
n=1
log2
P 1+ N (G(n, j)G(n, j)H )nn
.
This result also assumes perfect cancellation. An alternative order of antennas for detection is obtained by picking the transmit antenna which is received with the best SNR at each stage (Golden et al., 1999). This is done by choosing the antenna n which corresponds to the smallest value of {(G(n, j)G(n, j)H )nn } at each stage. This choice of ordering leads to a slight reduction in channel capacity, compared to a fixed ordering.
11.6.3 Capacity comparisons In order to compare the performance of the spatial multiplexing algorithms, results for the 10% outage capacity of the different methods have been obtained. The first simulation measured the performance of the algorithms for different SNR values (P/σ 2 ) for a configuration with eight transmit and eight receive antennas. For simplicity, a flat fading channel with one channel tap and a carrier bandwidth of B = 1Hz has been considered. The results
332
Next Generation Mobile Access Technologies Channel Capacity (in bits/sec) for 10% Outage
(a) 70
1 TX 1 RX 8 TX 8 RX: ETR MLR BLAST LD
60 50 40 30 20 10 0 0
5
10 15 20 Signal-to-Noise Ratio(dB)
25
30
Channel Capacity (in bits/sec) for 10% Outage
(b) 25
1 TX 1 RX 8 TX 8 RX: ETR MLR BLAST LD
20
15
10
5
0 1
2
3
4
5
6
7
8
Number of Transmit Antennas N
Fig. 11.12. The 10% outage channel capacity of the candidate MIMO architectures: (a) different SNR values for the candidate MIMO architectures with M = N = 8 antennas (unless stated) and (b) different numbers of transmit antennas N for an c SNR value of 10dB and M = 8 receive antennas 2000 IEEE.
for the different algorithms are shown in Figure 11.12†(a), with the single transmit and receive antenna system shown for reference purposes. The results show a significant increase in outage capacity for all the architectures compared to the single antenna case. For low SNRs, the ETR scheme is slightly better than MLR and BLAST. However, when the SNR is increased, the ETR, MLR and BLAST schemes converge to the same capacity result, something that has been shown theoretically in (Foschini, 1996) and (Raleigh and Cioffi, 1998). Finally, the LD scheme is seen to perform significantly worse than the other (N, M ) architectures. This is because all the degrees of freedom at the receiver array must be used to cancel the † Figure 11.12 is reproduced with permission from: (Thompson et al., 2000).
Smart Antennas for TDD-CDMA Systems
333
co-channel interference due to other transmit antennas, which reduces both SNR and diversity gain against Rayleigh fading. The second simulation set the receiver array size M = 8 and the SNR to 10 dB. The transmit array size N was varied to determine its effect on outage capacity. The curves that were obtained are presented in Figure 11.12(b). For small values of N , there is little to choose between any of the algorithms. However, the outage capacity of the LD receiver actually starts to reduce when N > M/2, as the receiver uses up more and more degrees of freedom to cancel co-channel interference. These observations seem to match with previous analysis of the LD receiver in (Winters, 1987). The BLAST scheme results flatten out at N = 6, while there is little to choose between MLR and ETR even when N = 8. These results suggest that the ETR, MLR and BLAST schemes will generally offer similar performance in a MIMO system. In reality, the performance of BLAST will be degraded slightly by errors in the cancellation structure. In terms of complexity, the ETR architecture is simpler to implement than the BLAST algorithm and particularly the MLR detector. However, the ETR technique requires accurate knowledge of the channel matrices H(j) at the transmitter in order to operate correctly.
11.6.4 Coding issues The ETR, MLR and BLAST architectures can be used with standard coding techniques, such as convolutional, trellis or turbo codes. The coded data symbols from one encoder may be multiplexed between the different transmit antennas. Alternatively, a separate encoder may be used for each transmit antenna. It is also possible to design space-time trellis codes (Tarokh et al., 1998); (Naguib et al., 2000) to exploit the multiple transmit antennas in an optimal way. The use of an MLR receiver is required to provide optimal performance for such codes.
11.7 Discussion and conclusions This chapter has provided a tutorial overview of smart antenna arrays for TDD wireless communication systems. In the currently proliferating market of wireless cellular communications and WLANs, sharing the available spectrum among different high-capacity users effectively depends on the signalprocessing techniques being discussed in this chapter. Smart antennas are one of the key technologies expected to dramatically improve the performance of current and future wireless communication systems because they
334
Next Generation Mobile Access Technologies
have the potential to expand coverage, increase capacity and improve signal quality. Focusing on the performance in TDD CDMA and OFDM systems, we summarise our results as below: • Smart antennas can be used to reduce fading and tackle interference. This in turn allows the increase in capacity of existing or future mobile communications networks. For CDMA structures, the improvement of the SNIR directly translates into increased numbers of users in a cell. • For the design and simulation of any smart antenna system, it is important to accurately model the radio channel in the time, frequency and space domains. The angle spread of the radio channel is a key factor for determining the performance of antenna array techniques. • Two other key parameters that define array performance have been presented: array gain, which is the gain expected over thermal noise; and diversity gain, which is the reduction in multipath fading. In a multiple transmitter scenario, the gain in performance has been shown to be proportional to the number of receiver antennas M . • A number of uplink algorithms have been described and compared. In a CDMA system, fixed-beam solutions provide a simple means to obtain significant capacity gains, provided the angle spread is not too high. The space-time 2D-RAKE receiver and MMSE combiners can improve performance, although the additional gains over fixed beam techniques may not be significant. • For the downlink, it was shown that beamforming techniques, such as maximum SNR and pre-RAKE beamforming, provide significant performance gains over pure diversity techniques, such as antenna selection and space-time transmit diversity, particularly as the number of antennas is increased. • OFDM techniques have been introduced as a practical technique to obtain spectrally efficient communications with simple receiver structures. In high data rate WLAN applications, OFDM is likely to be preferred to CDMA techniques. Chapter 12 will discuss further some of the issues in TDD-OFDM systems. • Finally, spatial multiplexing techniques have been introduced as a promising new means to improve spectral efficiency by an order of magnitude or more. The eigenmode transmit receive configuration is a promising approach for use in the downlink of TDD systems, provided that accurate uplink channel estimates can be used. In conclusion, it is believed that antenna arrays can be used in TDD systems to significantly improve performance. They may be used only at
Smart Antennas for TDD-CDMA Systems
335
the base station for uplink reception and downlink transmission. This configuration offers the capability to reduce interference and increase the user capacity of TDD wireless systems. Alternatively, antenna arrays at both the base station and mobile terminal may be used in a spatial multiplexing configuration to improve the data capacity of that wireless link. In either case, antenna arrays will have an important role to play in improving the performance of future wireless networks.
12 Cellular OFDMA-TDD Harald Haas Peter Omiyi Gunther Auer
12.1 Motivation and problems High peak data rate transmission, network self-organisation and universal frequency reuse are considered important features for future cellular, ad hoc, multi-hop and hybrid wireless networks (Prehofer and Bettstetter, 2005). OFDMA is viewed as a promising modulation/multiple-access technique for providing very high data-rates and flexible resource allocation while at the same time enabling low complexity receivers (Stimming et al., 2005). Timedivision duplexing (TDD) supports traffic asymmetry very well which is inherent to packet data services. Moreover, TDD offers channel reciprocity which is exploited in this research in a novel fashion for medium access and subchannel allocation. The problems that arise from TDD are the requirement for time synchronisation and additional interference scenarios. This is particularly important as OFDMA performs poorly under conditions of universal frequency reuse because of the high CCI. Figure. 12.1 illustrates the CCI problem in a cellular network using TDD with frequency reuse of one. The figure shows two adjacent BSs (base stations), namely BS1 and BS2 with a MS (mobile station) associated with each BS, namely MS1 and MS2. MS1 is transmitting data to BS1. Consequently, MS1 causes interference to MS2, since MS2 is in receiving mode. Similarly, BS2 causes interference to BS1. Due to the potentially small spatial separation between transmitter and ‘victim’ receiver and the low path loss between BSs due to line-of-sight conditions, in a full frequency reuse network, CCI poses a major challenge on the MAC protocol design and channel assignment procedure. Additionally broadband OFDMA is susceptible to frequency selective fading which, on the one hand, can potentially compromise performance. On the other hand, frequency selectivity can be exploited constructively for subchannel assignment. If, for example, certain subchannels on the link between 336
Cellular OFDMA-TDD
337
BS1 BS2 MS1
MS2
Transmission Interference Fig. 12.1. Co-channel interference in OFDM/TDD networks with full frequency reuse.
a Tx and a vulnerable Rx are highly attenuated, the Tx may use the same subchannels on the link to its associated Rx so that data transmission on both links is not significantly affected. In the following sections a low complexity and efficient concept based on TDD properties is proposed and analysed in-depth which completely avoids the possible high interference scenarios in a cellular and ad hoc OFDMA/TDD-based network while at the same time allowing flexible use of channel asymmetry on individual links. The proposed concept exploits the frequency selective channel. The key enabler is the channel reciprocity in TDD which enables the potential interfering transmitter to determine the interference it would cause to already active coexisting links prior to sending data. Based on that information, resources (subchannels) are selected for transmission. Consequently, it will be demonstrated that the air interface is equipped with a flexible, dynamic and self-organising FDMA component. With this interference aware mechanism, CCI is effectively taken into account during the channel assignment process at the MAC protocol as well as the radio resource management stage resulting in a new cross-layer air interface design ideally suited for future wireless systems.
12.2 Interference analysis The aim is to use OFDM in cellular, multi-hop and ad hoc networks. Due to the hybrid nature of future wireless networks, full frequency reuse is preferred. Unlike spread-spectrum techniques, pure OFDM does not have an inherent interference resistance mechanism. However, since frequency planning is undesirable, the use of OFDM in cellular, multihop and ad hoc networks must rely on powerful medium access and dynamic subcarrier assignment techniques. In order to be able to assess the scale of the interference problem in a full frequency reuse scheme, the interference powers are
338
Next Generation Mobile Access Technologies
calculated analytically. The underlying assumption is that a TDD based airinterface with uncoordinated uplink and downlink transmission is used, such that all interference scenarios exist. Furthermore, it is taken into account that the propagation channel between base stations is different from the propagation channel between base stations and mobile stations due to the increased likelihood of line-of-sight (LOS) conditions between base stations. This work extends the approach in (Haas and McLaughlin, 2004). First, the actual pdf of interference is derived. Second, the analysis is done for the more general scenario of a self-organising time-division-duplex (TDD) network, where cell deployment is not regular and for all possible interference scenarios, namely MS↔MS, BS↔BS and BS↔MS (and vice versa). It is, therefore, also relevant to ad hoc and multi-hop systems. The network model consists of a service area with radius RSA , with randomly positioned cells surrounding a central “tagged” cell as shown in Figure 12.2. The cells are randomly positioned with maximum cell radius of Rcell , as a consequence of the assumed uncoordinated network deployment scenario, including operators sharing the same spectrum, together with ad hoc or multi-hop mode operation. Universal frequency reuse is assumed. It is envisaged that this models a possible 4G (fourth generation) network deployment scenario, lacking in any network planning. The size of the service area should be chosen sufficiently large to ensure that interference is negligible outside the service area, which implies that RSA ≫ Rcell .
12.2.1 Power control At the beginning, a power-controlled system that attempts to minimise interference by maintaining a target-received power Sd at the receiver is assumed. In Figure. 12.3 the model for the derivation of intercellular interference is depicted. The constant target-received power Sd can be denoted as follows: Sd = Tdx Gx ,
(12.1)
where Tdx is the transmit power of node x and Gx is the link gain between Tx node x and its associated receiver. Moreover, Gxy is the link gain between Tx node x and victim node y. The interference node x causes to node y is then, ixy = Tdx Gxy ,
(12.2)
Cellular OFDMA-TDD
339
RSA
RCELL
Fig. 12.2. Network deployment
Fig. 12.3. Interference model with power control.
where Gxy is the link gain between node x and node y. Solving (12.1) for Tdx and inserting in (12.2) results in: ixy = Sd
Gxy , Gx
(12.3)
Consequently, the intercellular interference contribution ixy , from a transmitting station x (m for a MS and b for the BS) in a neighbouring cell to a receiving station y (m for a MS and b for the BS) in the tagged cell, can be
340
Next Generation Mobile Access Technologies
expressed as: ixy = Sd gxy ,
(12.4)
G
where gxy = Gxxy is the normalised intercellular interference power from transmitter x to receiver y. Applying the standard model for radio propagation, it can be found that:
ε x y −ε x (dxy )−γx y (12.5) 10 10 , gxy = −γ x (dx ) where dxy is the distance between the transmitter that causes interference, x, and the victim receiver, y, and γxy is the propagation path loss exponent between them. Shadow fading on the same link is modelled by εxy , which is a Gaussian-distributed random variable with standard deviation σxy and zero mean. Similarly, dx is the distance between the same transmitter and the intended MS, γx is the respective path loss exponent, and εx is the Gaussian-distributed random variable with standard deviation σx modelling shadow fading effects on the desired link. The distances dxy and dx are random variables. Clearly, the statistics of the intercellular interference from transmitter x to a receiving station y is entirely dependent on the statistics of gxy . Since, from (12.5), gxy is the product of the three independent random variables ξ = 10(εx y −εx )/10 , Γxy = (dxy )−γx y and Γx = (dx )γx , the pdf of ln(gxy ) is given by: ! ∞ fξ (ϑ − τ ) fΓ (τ ) dτ fgxy (ϑ) = ! ∞ !−∞ ∞ fΓxy (z) fΓx (τ − z) dzdτ, fξ (ϑ − τ ) = −∞
−∞
(12.6)
where fξ (ϑ), fΓxy (ϑ) and fΓx (ϑ) are the pdfs of the natural logarithms of the random variables ξ, Γxy and Γx , respectively, and fΓ (ϑ) is the convolution of the latter two pdfs. The final expression for fgxy (ϑ) is obtained as follows: fgxy (ϑ) = +
$
% ln(ζ) ϑ εx exp[(2ϑεx + ε2x )/2σ 2 ]ζ −1/γx 1−Q − − (γx + γxy ) σ σ σ
2 1/γ 2 x y exp[(εxy − 2ϑεxy )/2σ ]ζ ln(ζ) ϑ εxy . ×Q − + (γx + γxy ) σ σ σ
Cellular OFDMA-TDD
341
γ
γx xy ; Q(·) is the Gaussian Q-function defined as follows: /RRV where ζ = Rcell ! ∞ −x2 /2 e √ Q(u) = dx (12.7) 2π u
The cdf (cumulative density function) is found from (12.7) as follows: ! u fgxy (ϑ) dϑ. (12.8) cgxy (u) = −∞
Figure. 12.4 shows plots of cgxy (u) for the BS↔BS (BB) links, MS↔BS (MB, BM) links, and the MS↔MS (MM) links. The cdfs are shown for different cell radii Rcell . The parameter values that are chosen for the shadowfading standard deviation and the path loss exponent reflect the fact that there are better line-of-sight conditions for BS↔BS links than MS↔BS links. The shadow fading standard deviation for the BS↔BS link is σbb = 6 dB, compared to those for MS↔BS links σm = σb = σmb = σbm = 8 dB. The path loss exponent for the BS↔BS link is γbb = 2, while those for the MS↔BS links are γm = γb = γmb = γbm = 3. The MS↔MS link has the worst line-of-sight conditions with the highest path loss exponent γmm = 4 and the highest shadow fading standard deviation σmm = 10 dB. In Figure. 12.5 a zoom of the plots in Figure. 12.4 is shown and the values for the median are plotted. The plots show that the worse the line-of-sight conditions, the lower the mean normalised intercellular interference but the higher the variance. In addition, the results show the severity of BB interference. Moreover, the median in the case of Rcell = 1000 m is about 20 dB higher than for the case where Rcell = 200 m. This suggests the use of small cells, and consequently the introduction of multi-hop and ad hoc communication. The normalised link gain can directly be translated to the required SINR. If the normalised link gain is 0 dB, this means that the level of interference and the level of the useful signal are equal, i.e. the SINR = 0 dB – neglecting background noise. If, for example, it assumed that an SINR of 10 dB is required (e.g. when using QPSK modulation), this translates into a normalised link gain of −10 dB. From the plots in Figure. 12.5 it can be seen that this target SINR is only achieved with a probability of about 45% if BS-BS interference is considered and when the cell radius is 200 m. In the case of a cell radius of 1000 m this probability is only about 10%, i.e., the outage is 90%. The situation is significantly improved for the case of BS↔MS interference. The probability that the SINR target is achieved is 75%, and 95% respectively. For MS-MS interference, the probabilities are 95%, and 98% respectively.
342
Next Generation Mobile Access Technologies
Cumulative density function
1 0.9
Monte Carlo sim. Analytical approach
0.8 0.7 0.6
BM/MB
0.5
BB
MM
0.4 0.3 0.2 0.1 0 250
200
150
100
50
0
50
100
150
200
250
Normalised interference power [dB] (a) R cell =200m
Cumulative density function
1 Monte Carlo sim. Analytical approach
0.8
0.6
BM/MB BB
MM
0.4
0.2
0 250
200
150
100
50
0
50
100
150
200
250
Normalised interference power [dB] (b) R cell =1000m
Fig. 12.4. Plots of cg xy (u).
12.2.2 Fixed transmit power The interference model for fixed power control is shown in Figure. 12.6. Since no power control is in place, and fixed transmit power is assumed, the interference at node y does not depend on the path gain Gx . The interference is directly proportional to the link gain between node x and node y: Gxy = C (dxy )−γx y 10
−ε x y 10
,
(12.9)
Cellular OFDMA-TDD
343
Cumulative density function
1
BB
Monte Carlo sim. Analytical approach
0.8
0.6
BM/MB
MM
0.4
0.2 78dB 0 100
90
80
70
8dB
42dB
60
50
40
30
20
10
0
10
Normalised interference power [dB] (a) R cell = 200 m
Cumulative density function
1
MM 0.8
BB BM/MB
Analytical approach Monte Carlo sim.
0.6
0.4
0.2
0 80
70
60
50
12dB
19dB
57dB
40
30
20
10
0
10
20
30
Normalised interference power [dB] (b) R cell = 1000 m
Fig. 12.5. Plots of cg xy (u).
where C is the path gain constant. The interference at node y is: ixy
= Tdx Gxy ,
(12.10)
= Tdx C (dxy )−γx y 10
−ε x y 10
.
(12.11)
Without loss of generality, it is assumed in the following that Tdx C = 1, and the new normalised intercellular interference power is: g˜xy = (dxy )−γx y 10
ε x y −ε x 10
.
With this definition, the pdf of g˜xy is derived as follows:
(12.12)
344
Next Generation Mobile Access Technologies
Fig. 12.6. Interference model for fixed transmit power.
f˜gxy (ϑ) =
=
exp[(ε2xy − 2ϑεxy )/2σ 2 ]ζ 1/γx y γxy $ % ! ∞ −(τ − ϑ + εxy )2 1 exp dτ × 2σ 2 σ (2π) ln(ζ) exp[(ε2xy − 2ϑεxy )/2σ 2 ]ζ 1/γx y γxy
ln(ζ) ϑ εxy , − + Q× σ σ σ
(12.13)
(12.14)
and, consequently, the cdf is: c˜gxy (u) =
!
u
f˜gxy (ϑ) dϑ.
(12.15)
−∞
In contrast to the case of power control, the interference in this case is dependent on the path gain constant C. This constant depends on the actual path loss model. In the following the UMTS (universal mobile telecommunications system) indoor path loss model is assumed, L = 37 + 10 γ log10 (d)
[dB],
(12.16)
where d is the distance between transmitter and receiver. As in the derivation of the pdf, path gains instead of path losses are used, the constant C is −37 dB. Furthermore, a constant transmit power Tdx of 30 dBm is assumed. For these parameters and a cell radius of 1000 m, a comparison of the cdf
Cellular OFDMA-TDD
345
Cumulative density function
of BS↔BS interference for the power controlled and non-power controlled systems is shown in Figure. 12.7. 1
0.8
0.6
0.4
Power control sim. No power control sim.
0.2
Analytical approach 0
120
100
80
60
40
20
0
Interference power [dBm] Fig. 12.7. Cdf of BS↔BS interference assuming a cell radius of 1 km for the power controlled and the non-power controlled systems.
When taking the median it can be seen that the system without power control results in about 10 dB less interference. This might come at the expense of reduced SINR in some cases. In any case, the SINR at the receiver is subject to variation, but link adaptation techniques are to be used to exploit these SINR variations. Hence, in the system models developed in the sequel, fixed transmit powers are assumed. 12.3 The busy-tone approach TDD networks have the advantage over traditional FDD (frequency-division duplexing) networks of flexible dynamic resource allocation, which is required to support the wide range of dynamic traffic characteristics anticipated for wireless broadband multimedia services. Almost the entire transmission link (i.e. timeslots per frame) can be dedicated to uplink or downlink traffic. Also, the link capacity can be dynamically assigned on a per time frame basis to a single or plurality of users. In addition, TDD allows the exploitation of channel reciprocity, reduces the complexity of the RF front end, and is an enabler for ad hoc wireless networks and multi-hop cellular mobile networks. When multiple BSs are operated in close proximity to one another on the same radio frequency channels, TDD networks suffer from significant
346
Next Generation Mobile Access Technologies
levels of BS↔BS and MS↔MS (i.e. same-entity) interference (Haas et al., 2000a, 2002; Haas and McLaughlin, 2001a; Lindstrom, 2001; Wie and Cho, 2001; Jeong and Kavehrad, 2002; Nasreddine and Lagrang, 2003). The TDD interference scenario is illustrated by the example shown in Figure 12.8†, showing three BSs in close proximity, and three time frames of the TDMA frame structure of the system with three timeslots per frame. BSTx 1 reserves the first two timeslots per frame for its downlink transmission. Tx The downlink transmission of BSTx 1 experiences significant MS↔MS (MS2 to MSRx 1 ) interference in the second slot of frame (i + 1) and frame (i + 2), while its transmission in the first slot of each frame experiences weaker Rx MS↔MS (MSTx 3 to MS1 ) interference. As a result, the throughput in the second slot of each frame at MSRx 1 is zero. BS1Tx DL Transmission
Tx
Tx
Tx
Tx
Tx
Tx
MS1Rx DL Reception
Rx
Rx
Rx
Rx
Rx
Rx
MS2Tx UL Transmission
Tx
MS3Tx UL Transmission
Tx Frame i
Idle Slot Transmission Reception Collision
Tx Rx Rx
Tx Tx Frame i+2
Frame i+1
time
Time Slot
Transmission Interference
BS1Tx
MS2Tx BS2Rx
MS1Rx BS3Rx MS3Tx
c Fig. 12.8. Intercellular interference scenario. 2007 IEEE
The busy-burst broadcast protocol, denoted as busy-burst TDMA (BBTDMA), has been proposed to mitigate this problem, in order to make TDD systems more robust to universal frequency reuse. The main principle is that receivers upon successful transmission of a packet transmits a busy-burst on a succeeding minislot. Potential transmitters in neighbouring cells sense † Figures 12.8, 12.9, 12.10, 12.11, and 12.12 are reproduced with permission from: (Omiyi et al., 2007)
Cellular OFDMA-TDD
347
the minislot prior to signal transmission. With this mechanism, interference to the existing link is avoided. This mechanism is applicable to both infrastructure- and non-infrastructure-based networks. In the proposed BB-TDMA protocol, timeslots are organised into fixedduration time frames, and associated with each timeslot is a minislot. The minislots can occupy any position in the time frame. The operation of BBTDMA, in the proposed cellular TDD application, is explained with the aid of a network example shown in Figure 12.9. Rx BSTx 1 transmits, as scheduled, the two packets to MS1 in the first and second timeslots of frame i. Each transmitted packet contains a user identification field, and also a one-bit ‘continue’ field, using binary signalling, that indicates the transmitter’s intention to continue or discontinue using its transmission timeslot in the next frame. Both packets are received correctly, with their ‘continue’ fields indicating that BSTx 1 still requires the use of both these timeslots in the next frame. In response, MSRx 1 broadcasts busy-bursts in the minislots of the first and second timeslots in frame i, continues transmission in the next frame i + 1. The busy and so BSTx 1 burst broadcasts in frame i inform BSTx 1 that it can continue using the same timeslots in frame i + 1 without any significant intercellular interference to its downlink transmission. In addition, this busy-burst, which presents a strong signal to MSTx 2 , keeps it from using the second timeslot for its uplink transmission, as originally scheduled by BS2 ’s scheduler to begin in frame (i + 1). Thus, MSTx 2 is preRx vented from interfering with the downlink transmission to MS1 . When BSRx 2 fails to receive the scheduled transmission, it reschedules the transmission at some future time depending on its local scheduling policy. BSTx 3 detects a weak busy-burst, which indicates to BSTx 3 that its interference to MSRx 1 would be equally low. Therefore, it transmits on the downlink in the first timeslot, as scheduled. Rx In frame (i + 1), BSTx 1 transmits two more packets to MS1 in the first and second timeslots, and, as before, when the packets are correctly received with their ‘continue’ fields indicating that BSTx 1 has more packets to send, broadcasts busy-bursts in the minislots of the first and second then MSRx 1 timeslots in frame (i + 1). In this way, the busy-burst protects the downlink packet transmissions to MSRx 1 in the first and second timeslots from intercellular interference, with the last packet indicating with the ‘continue’ field that BS1 has sent all the packets. MSRx 1 does not broadcast the busy-burst after receiving the last packet. to MSRx in the first timeslot of If the packet transmitted from BSTx 1 1 frame i had not been correctly received, then no busy-burst would have
348
Next Generation Mobile Access Technologies
Tx been broadcast to BSTx 1 in this. In this case BS1 would have to reschedule the transmission for some future time. Also, if BSTx 1 had detected a strong enough busy-burst signal in the minislot associated with second timeslot in frame (i − 1), it would not have transmitted in this timeslot in frame i, and it would have been rescheduled for some time in the future. In either would not have broadcast a busy-burst in the of these two cases, MSRx 1 second timeslot in frame i, in which case MSTx 2 would have transmitted in the second timeslot in frame (i + 1), as originally scheduled.
BS1Tx DL Transmission
Tx
Tx
Tx
Tx
Tx
Tx
MS1Rx DL Reception
Rx
Rx
Rx
Rx
Rx
Rx
MS2Tx UL Transmission
Tx
MS3Tx
Tx
Tx
UL Transmission
Frame i
Tx Tx Rx Rx
Idle Slot Transmission Tx Cancelled Reception Collision
Tx No Busy signal Busy signal
Tx Frame i+2
Frame i+1
time
Time Slot
Transmission
X
BS1Tx
Interference
MS2Tx BS2Rx
MS1Rx BS3Rx MS3Tx
Fig. 12.9. Example of intercellular interference avoidance with busy-burst broadc casts. 2007 IEEE
A busy-burst is detected as “strong” by an intending transmitter when the condition Ib ·Td ≥ Ith (12.17) Tb is satisfied, where Tb is the fixed, known busy burst transmit power, Ib is the received busy burst power from transmissions in adjacent cells, and Td is the data transmit power of the intending transmitter. The interference threshold, Ith represents the maximum allowed inter-cell interference, and is given by the following relation Ith =
Sd,tg SIRtg · Δ
(12.18)
Cellular OFDMA-TDD
349
where Sd,tg and SIRtg are the target (minimum acceptable) received power and minimum acceptable SIR (signal-to-interference ratio), respectively. As (12.17) only ensures that the interference contribution of an interfering transmitter is less than Ith , it might happen that the sum of interference from multiple interfering transmitters exceeds the maximum acceptable interference (for minimum SIR) at a vulnerable receiver. Introducing a margin Δ in (12.18) makes this effect less likely. The choice of Ith is important: while setting Ith too high compromises the interference protection toward competing links, Ith being too low makes transmitters over cautious, reducing the network throughput (Haas et al., 2006). Clearly, from (12.17), when more than one busy-burst is heard, this has the effect of lowering the effective value of Ith , which makes the transmitter more cautious. This is, in fact, desirable, as it means that when the intending transmitter detects a high level of receiver activity in its vicinity it is more cautious in transmitting and therefore interfering. However, in order to prevent the transmitter from being too cautious, the transmitter attempts to extract the strongest busy burst signal. This can be realised in that the busy-burst is constructed as a code sequence from a finite quasi-orthogonal set, where each cell selects a code that is not used by any of its neighbours. In this way, a transmitter can search across the sequence set and time offsets for the strongest signal, while receiving a minimal contribution from the other busy-burst signals. The BS can select the appropriate code by listening to the busy-burst codes from neighbouring cells and choosing a different code to that of its neighbours. The protocol also requires that a transmitting station is able to identify the busy burst from the intended receiving station in the presence of busy bursts from other cells. Interestingly, as described in the following this is possible without any encoding of the busy burst. The threshold test in (12.17) ensures that a transmitter only starts transmission of a data packet if the following condition with respect to the received busy burst power Ib is fulfilled: Ib ·Td < Ith (12.19) Tb Suppose the receiver successfully decoded the packet. For a link to be viable for communication, the following condition for the received data power Sd must be satisfied Sd =
Sb ·Td ≥ Sd,tg = Ith ·Δ · SIRtg Tb
(12.20)
350
Next Generation Mobile Access Technologies
where Sb is the received busy burst power from the intended receiver, and the ratio Sb /Tb accounts for the corresponding channel gain of the intended link. To acknowledge the successful reception, the receiver transmits a busy burst with power Tb . This means that the next time the the busy burst channel is scanned, the intended transmitter will observe an aggregate received busy burst power of Ib + Sb . As a consequence, prior to the first data packet transmission the received busy burst power is below Ith , and after a successful transmission it is, at least, by a factor Δ SIRtg above the threshold, resulting in a surge of the received busy burst power by Sd,tg Sb > = Δ · SIRtg . Ib Ith
(12.21)
If the data packet is not received successfully, the received busy burst power remains below the threshold, i.e., no Δ·SIRtg –surge will be detected. This effectively ensures reliable detection of the busy burst on the intended link simply by measuring the surge in the total received busy signal powers. It is important to note that each BS has a scheduling policy that is independent of the BB-TDMA protocol. This scheduling policy is used by the BS to determine the order in which to send/receive packets to/from the MSs in its cell, such that all intra-cellular contention is completely resolved. The embedded BB-TDMA procedure is used only to prevent a transmitter (BS or MS) from transmitting in a timeslot (predetermined by an independent scheduling policy), if the transmitter detects a busy signal from one or more receivers in adjacent cells. Therefore, BB-TDMA is used only to augment the existing intra-cell MAC scheduling policy for enhanced interference management, rather than to replace it.
12.4 Delay–Throughput Performance The performance of BB-TDMA is derived analytically and compared to state of the art blind timeslot allocation, namely fixed-slot allocation (FSA) and random time slot opposing (RTSO) (Haas et al., 2002). FSA avoids extremes of interference by preventing slot allocations that result in same-entity interference, synchronising uplink transmissions (or equivalently downlink transmissions) across all BS sites. RTSO averages out the interference experience of a user, in a manner analogous to frequency hopping but in the time domain. Simulation results are provided to verify the new analytical framework.
Cellular OFDMA-TDD
351
The network model is shown in Figure 12.2. The following assumptions are made for the analysis: • Only one user is active per cell per timeslot, so intra-cell interference is completely avoided. • A user is scheduled at most on one timeslot per frame. • A timeslot carries one packet of data and packet burst arrivals are Poisson distributed. • The system is perfectly power controlled to minimise interference. • The receive/transmit turn around time is negligible and the system is perfectly time synchronised. • In BB-TDMA, a transmitter always distinguishes between the busy burst from its intended receiver and other receivers. • Packet errors only result from inter-cell interference and not from noise, representing an interference limited environment. • The pathloss is subject to log-normal shadowing. • Small-scale Rayleigh fading is not considered. For a wideband system with sophisticated diversity and error correction coding techniques in place, this assumption appears justified. The effects of fading, however, are accounted for in the chosen SIR target SIRtg in (12.20). The throughput R and delay D in general are a function of the inter-cell interference and the process of packet burst arrivals, including possible link asymmetries between uplink and downlink. For the interference statistic we take into account different propagation scenarios, described by the pathloss exponent and the shadow fading standard deviation, for BS-to-BS, MSto-BS and MS-to-MS interference. To this end, BSs may encounter line of sight (LoS) conditions to adjacent BSs, especially if they are mounted above rooftop. On the other hand, this assumption is in most cases not applicable to other types of interference. As will be shown later on, the propagation conditions have severe implications on the pdf describing the interference statistics, and thus allow to model the performance of a TDD system more realistically. In order to derive R and D in closed form, the following probabilities are to be determined: (ul)
Pon
(ul)
Pout
(dl)
and Pon denote the probabilities that an uplink and downlink burst at the head of the queue is permitted by the protocol to begin transmission at a particular slot. (dl) and Pout denote the outage probabilities caused by inter-cell interference for packets transmitted in the uplink and downlink, respectively.
352
Next Generation Mobile Access Technologies
(yx)
Pbusy accounts for the probability that a busy burst from receiver y (m for a MS and b for BS) is detected as “strong” by the intending transmitter x (m for a MS and b for BS), when its potential interference contribution to receiver y is greater than or equal to Ith in (12.17). (yx)
Clearly, the busy-burst detection probability Pbusy is applicable only to BBTDMA. The transmission and outage probabilities Pon and Pout exist for FSA, RTSO and BB-TDMA, although, the obtained expressions are different. Furthermore, while Pon is a function of both the packet arrival process (yx) and the interference, Pbusy and Pout are fully described by the latter.
12.4.1 Throughput Performance and Packet Loss For BB-TDMA, only the first packet of a burst is transmitted in contention (i.e., without busy burst protection). Therefore, it is possible for it to be received in outage. When this occurs, repeated transmission attempts of the packet are made in contention mode until the packet is successfully received and a reservation secured for the entire packet burst. Once a packet burst has been granted access to the channel, it is protected by the receiver busy burst and so every packet in the burst is delivered without error. Let wTx denote the geometrically distributed uninterrupted transmission time for a packet burst in timeslots. Then, the mean throughput is R = λE {wTx } in packets per unit time (timeslot). The mean arrival rate into the slot is λ packets per unit time, and the expectation of the uninterrupted transmission time is E {wTx }. For FSA and RTSO, however, the situation is different, as packets will occasionally be lost due to inter-cell interference. The throughput with respect to the offered load is reduced according to (ul) (dl) (12.22) R = λE {wTx } 1 − Pout Pul − Pout (1 − Pul ) .
where Pul and (1−Pul ) denote the probabilities that a packet burst is for the uplink and downlink, respectively. While for RTSO both same-entity and other-entity interference occur, FSA completely avoids same-entity interfer(ul) (dl) ence. Hence, for FSA the outage probabilities Pout and Pout in (12.22) (oe,ul) (oe,dl) are replaced by Pout and Pout , to emphasise that the only sources of interference are “other entities” (oe). The proportion of transmitted packets lost due to outage is given by Ploss = λE {wTx } − R.
(12.23)
Cellular OFDMA-TDD
353
12.4.2 Delay Performance The delay performance, when observing an arbitrary timeslot in the tagged cell, is that of a discrete time M /G/1 queue, where M indicate a Poisson arrival process, G indicates that service times have an arbitrary statistical distribution, and ‘1’ indicates that it is a single server queue. The mean delay of a packet burst in the M /G/1 queue, including the transmission time is given by the Pollaczek–Khinchin formula (Chan et al., 1997) ) 2 * λE wserv . (12.24) D = E {wserv } + 2 1 − λE {wserv }
The packet burst service time wserv is modelled as the sum of two independent random variables, wserv = wvac + wTx , where wvac is the inter-cellular timeslot allocation delay (or server “vacation”) time experienced by a packet burst at the head of the queue before transmission. Since all packet bursts are transmitted without interruption in BB-TDMA, FSA and RTSO, the statistics of wTx are the same for all schemes. However, the statistics for wvac are protocol dependent. A frame consists of one or more timeslots. For simplicity it is assumed that a user transmits in at most one slot per frame. Then the decision to permit or restrict access is statistically independent from one slot to the next, and wvac is geometrically distributed, with first and second moments given by 1 1 E {wvac } = Pul − 1 + (1 − Pul ) −1 (12.25) (ul) (dl) Pon Pon and ) 2 * = Pul E wvac
(ul) (ul) 2 − Pon 1 − Pon
+ (1 − Pul )
(ul)2
Pon
(dl)
2 − Pon
(dl)2
Pon
(dl)
1 − Pon
.
(12.26) (ul)
In the uplink, a MS accesses a certain timeslot with probability Pon . For (ul) BB-TDMA, Pon is given by the joint probability that no strong busy burst is detected by the MS transmitter in the tagged cell from any receiver of the neighbouring cells (transmitter of the busy burst), and the probability that the first packet transmission of the burst is successfully received. As the first packet transmission is not protected by a busy burst, outage may (ul) occur with probability Pout . Let Po ul and (1−Po ul ) denote the probabilities that a timeslot is scheduled for the uplink and downlink in a neighbouring
354
Next Generation Mobile Access Technologies
cell. Provided that a single neighbouring cell is active, the probability that a busy burst from a BS and MS in the neighbouring cell is detected prior (bm) to an uplink slot in the tagged cell, is the joint probability Po ul Pbusy and (mm)
(1−Po ul )Pbusy , respectively. Then (bm) (mm) ρ1 = 1 − Po ul Pbusy − (1 − Po ul )Pbusy
(12.27)
accounts for the probability that no strong busy signal is received from a BS or a MS in a single cell. Given n active users in neighbouring cells, the probability that no strong busy burst is detected from any of the neighbouring cells amounts to ρn1 . Assuming that the number of neighbouring cell (ul) transmissions in a timeslot is Poisson distributed with parameter ρo , Pon is derived as follows (ul) Pon (ul)
=
1−
(ul) Pout
∞ [ρo ρ1 ]n exp (−ρo )
n=0
n!
(12.28)
where (1 − Pout ) is the probability that the first packet transmission of the burst is successfully received. Given the Poisson distribution of the number of co-channel transmissions, (12.28) can be re-arranged and simplified to obtain (ul) (ul) (12.29) = 1 − Pout exp(ρo ρ1 − ρo ) . Pon Substituting (12.27) into (12.29) yields, 9 : (ul) (bm) (mm) (ul) Pon = 1 − Pout exp −ρo Po ul Pbusy − ρo (1 − Po ul )Pbusy
(12.30)
The Poisson parameter ρo is a function of the traffic density in the circular region of radius RSA centred at the receiving station (BS or MS) in the tagged cell. The corresponding probability that a downlink packet at the head of the (dl) queue is permitted to begin transmission, Pon , is derived in a similar man(ul) ner as Pon to give : 9 (bb) (mb) (dl) (dl) (12.31) = 1 − Pout exp −ρo Po ul Pbusy − ρo (1 − Po ul )Pbusy . Pon
For FSA, an uplink packet burst at the head of the queue is permitted transmission in a timeslot only if the same timeslot has also been reserved for uplink transmissions in all neighbouring cells, and is delayed otherwise. The same is true for a downlink packet burst. This means, Pr {cell i in UL |cell j in UL } = 1 ∀ j .
(12.32)
Cellular OFDMA-TDD (ul)
355
(dl)
Therefore, Pon = Po ul and Pon = (1 − Po ul ). (ul) For RTSO, packet bursts are not restricted access at all, and so both Pon (dl) and Pon equal unity.
12.4.3 Modelling the Effects of Inter-Cell Interference The BB-TDMA busy burst detection process, in effect, involves estimating the value of gxy between an intending transmitter x and the most vulnerable receiver y in a neighbouring cell, and comparing this to a threshold (12.17). The most vulnerable receiver is the receiver with the largest link gain gxy to [xy] the interfering transmitter. If Ib denotes the received busy burst power [xy] from the most vulnerable receiver, then Ib is related to the total received [xy] busy busy power Ib from other cells by Ib ≈ a Ib , with a≥1. The factor a is the increase in the busy burst power from additional weaker busy burst signals. Then (12.17) can be re-expressed as [xy]
ixy
I Ith = b ·Td ≥ Tb a
(12.33)
where ixy = gxy Sd from (12.4) defines the interference from transmitter x to receiver y of a perfectly power controlled system. Eqn. (12.33) shows that the difference between using the total busy burst signal and the strongest busy burst signal maps simply to an effective lower protocol threshold for the latter. In order to simplify the subsequent analysis, it is assumed that the strongest busy burst signal is significantly greater than the sum of all other busy burst signals. Therefore, we set a=1 in the following. However, equivalent results are expected for a system using the total busy burst signal, provided the protocol threshold is appropriately scaled by a factor a≥1. Substituting ixy = gxy Sd from (12.4) into (12.33) for a=1, the probability (yx) Pbusy is given by 8 & 8 & Ith Ith (yx) Pbusy = Pr gxy ≥ = Pr ln(gxy )≥ ln Sd Sd ! ∞ fgxy (ϑ) dϑ, (12.34) = ln(Ith /S d )
where fgxy (ϑ) is given by (12.7). Substituting (12.7) into (12.34) and solving (yx) the integral numerically gives the busy channel probability Pbusy . In the following, the expression necessary to calculate the outage prob(ul) abilities on the uplink Pout is derived, as a function of the interference
356
Next Generation Mobile Access Technologies
statistics. The total instantaneous inter-cell interference for BB-TDMA and RTSO, without any restriction on same-entity interference, is given by iul (n) =
n 9 : (i) (i) (i) (i) Uul imb + (1 − Uul )ibb
(12.35)
i=1
with iul (0) = 0. Furthermore, n accounts for the number of active neighbouring cells in the current timeslot, assumed to be Poisson distributed with mean ρo (as defined in section 12.4.2). The parameter ixy is defined in (12.4) and is the interference contribution from a transmitting station x (m for a MS and b for the BS) in a neighbouring cell to a receiving station y (m for (n) a MS and b for the BS) in the tagged cell. The parameter Uul is a random variable that equals one with probability Po ul if the transmission in cell n is for the uplink, or zero otherwise. It is a function of the traffic density in the circular region of radius RSA centred at the receiving station (BS or MS) in the tagged cell. (ul) The outage probability Pout is given by (ul)
Pout =
∞ ρn exp (−ρo ) o
n=1
n!
Pr {iul (n) > Isir }
(12.36)
where Isir = Sd /SIRtg is the maximum acceptable interference, in order to attain the SIR target SIRtg , given a constant received power Sd . Since it is difficult to further simplify (12.36), an approximation is derived as follows. From (12.35), iul (1) is the interference contribution on the uplink from a single neighbouring cell, and iul (n) is the n-fold convolution of independent identically distributed random variables with the pdf of iul (1). Now given that n neighbouring cells are simultaneously active, outage occurs if the interference contribution from any one cell exceeds Isir , or if none of the individual interference contributions exceeds this threshold but their sum exceeds it. Then (12.36) can be re-expressed as (ul) Pout
$ ∞ ρno exp (−ρo ) n = 1 − Pr {iul (1) ≤ Isir } n! n=1
+ Pr {iul (1) ≤ Isir }n % * ) × Pr iul (n) > Isir iul (1) ≤ Isir ∀ n links .
(12.37)
Therefore, from (12.37) and since the second term is always positive, a
Cellular OFDMA-TDD
lower-bound on the outage probability is given as
∞ ρno exp (−ρo ) (ul) n 1 − Pr {iul (1) ≤ Isir } . Pout ≥ n!
357
(12.38)
n=1
This gives a good approximation of the outage probability and a tight lower bound for small and high values of Pr {iul (1) ≤ Isir } and for small and high values of ρo . Equation (12.38) simplifies to give (ul)
Pout ≥ 1 − exp (−ρo Pr {iul (1) > Isir })
(12.39)
where Pr {iul (1) > Isir } =
!
∞
log(Isir /S d )
Po ul fgmb (ϑ) dϑ ! ∞ + (1 − Po ul )fgbb (ϑ) dϑ. (12.40) log(Isir /S d )
The derivation for the downlink is similar to the uplink. The total instantaneous inter-cell interference for BB-TDMA and RTSO on the downlink can be expressed as n 9 : (i) (i) (i) Uul i(i) (12.41) idl (n) = + (1 − U )i mm ul bm i=1
with idl (0) = 0. The expressions for the lower bound approximations of (dl) Pout is obtained by replacing the corresponding subscripts and superscripts in (12.39). The corresponding probability that the interference exceeds Isir yields ! ∞ Pr {idl (1) > Isir } = Po ul fgmm (ϑ) dϑ log(Isir /S d ) ! ∞ + (1 − Po ul )fgbm (ϑ) dϑ. (12.42) log(Isir /S d )
For FSA, which completely avoids same-entity interference, the total instantaneous inter-cell interference is given by ioe (n) =
n
(i)
ibm .
(12.43)
i=1
Due to the use of TDMA, the assumption of same transmit powers in uplink and downlink, channel reciprocity and the fact that BSs and MSs are both randomly and uniformly located in space, the interference terms on the up(i) (i) link and downlink are identical, ibm = imb , and therefore ioe (n) is valid for
358
Next Generation Mobile Access Technologies
both uplink and downlink. Expressions for the outage probabilities which (oe) completely avoid same entity interference, Pout , are again obtained by replacing the corresponding subscripts and superscripts in (12.39). Finally, the corresponding probabilities that the interference exceeds Isir yield ! ∞ Pr {ioe (1) > Isir } = fgbm (ϑ) dϑ (12.44) log(Isir /S d )
which is again valid for uplink and downlink.
12.5 Numerical Results The pathloss parameters are summarised in Table 12.1. All other parameters are summarised in Table 12.3. Table 12.1. Channel propagation parameters. σbb
6 dB
σm , σb , σm b , σbm
8 dB
σm m
10 dB
γbb
2
γm , γb , γm b , γbm
3
γm m
4
Table 12.2. Simulation parameters. Radius of service area, RR V
5 km
Maximum transmit power
30 dBm
Target receive power
-87 dBm
Minimum SIR requirement
4.8 dBm
BB-TDMA Ithresh
-95 dBm
BB-TDMA minislot overhead Average number of interferers ρo Uninterrupted transmission time E {wTx } Probability of uplink transmission in “tagged” and other cells Pul = Po ul
10% 6 300 frames 0.5
Cellular OFDMA-TDD
359
c Fig. 12.10. Delay-throughput results for cell size Rcell = 1 km. 2007 IEEE
Figure 12.10 and Figure 12.11 show the throughput-delay plots for BBTDMA, FSA and RTSO for for Rcell = 1 km and Rcell = 200 m, respectively. As a reference, the delay/throughput curve of the classical M/G/1 queue (without server vacations) is also plotted. This is the performance of the “perfect scheduling” scenario with zero timeslot allocation delay wvac and zero packet loss from interference. The delay values are obtained from (12.24) and the throughput values from the throughput equations of Section 12.4.1. The mean delay value for all schemes includes the 300 frame duration transmission time for a packet burst. Figure 12.12 shows plots of maximum throughput versus the uplink/downlink asymmetry ratio for BB-TDMA, FSA and RTSO for Rcell = 200 m and 1 km. For Figure 12.10 and Figure 12.11, traffic symmetry between the uplink and downlink in every cell is assumed; that is Pul = Po ul = 0.5. For the results of Figure 12.12, the uplink/downlink asymmetry ratio is varied and is defined here as the ratio of uplink traffic to the total traffic, which is given by Pul for the tagged cell and Po ul for the neighbouring cells. The same asymmetry is assumed for all cells (Pul =Po ul ). The pathloss and shadow fading parameters used to generate the analytical and simulation results are equivalent to the parameters used for Figure 12.4(a) and Figure 12.4(b), and
360
Next Generation Mobile Access Technologies
c Fig. 12.11. Delay-throughput results for cell size Rcell = 200 m. 2007 IEEE
Table 12.3. System parameters Radius of service area
RSA
5 km
Max. transmit power
T d = Tb
30 dBm
Target receive power
Sd
−87 dBm
Target SIR
SIRtg = Sd /Isir
BB-TDMA threshold
Ith
Average no. interferers
ρo
Uninterrupted Tx time
E {wTx }
5 dBm −95 dBm 6 300 frames
are summarised in Table 12.1. System and network related parameters are summarised in Table 12.3. The locations of the other cells relative to the tagged cell are determined randomly and the number of cells simultaneously active is Poisson distributed, with one active user per cell per unit time (due to the use of TDMA and the assumption of perfect time synchronisation). Interference and throughput/delay statistics are measured from the tagged cell. Nodes become active in the tagged cell according to a Poisson arrival process, with
Cellular OFDMA-TDD
361
c Fig. 12.12. Maximum throughput for different asymmetry ratios. 2007 IEEE
packet burst arrival rate λ, where each active node has an infinite buffer in which packets are queued until transmission. As expected the analytical results provide an upper-bound on the throughput of each scheme in Figure 12.10 and Figure 12.11, and the simulation results closely match the analysis. BB-TDMA is shown to significantly outperform the blind schemes FSA and RTSO, offering a superior delay throughput performance close to that of the ideal M/G/1 queue. Considering Figure 12.10 with a large cell radius of Rcell = 1 km, the transmit powers tend to be high due to the potentially large distances between a transmitter and its intended receiver. Consequently, inter-cell interference also tends to be high, frequently attaining unacceptably high values. The results indicate that the busy burst broadcast mechanism of BB-TDMA effectively mitigates this problem. In contrast, both RTSO and FSA suffer severely from interference for large cell coverage. RTSO pays a higher penalty due to the high outage probabilities from same-entity BS-to-BS interference (refer to Figure 12.4(a)), which is avoided by FSA together with the less problematic MS-to-MS interference. For large cell coverage, however, the difference in performance between FSA and RTSO is not significant because MS-to-BS interference is also severe.
362
Next Generation Mobile Access Technologies
Figure 12.11 shows the impact of reducing the maximum cell radius to Rcell = 200 m. Reducing the maximum cell radius, reduces the transmit power levels to maintain the target received power within its cell, which in turn also reduces the interference contributions to adjacent cells. The results show a significant performance improvement for FSA, but only a marginal improvement for RTSO. This suggests that reducing the maximum cell radius results in a more significant reduction in other entity interference than in same entity interference. This is because BS-to-BS interference still frequently reaches extremes due to the better line-of-sight conditions between BSs, despite the reduced transmit powers. BB-TDMA, on the other hand, gets very close to the optimal throughput/delay performance. Outage and packet loss are the dominant factors limiting the throughputdelay performance of FSA and RTSO, while the timeslot allocation delay wvac is the limiting factor for BB-TDMA. To this end, the lowest mean delay of BB-TDMA is D = 350 frames (300 frames of which is the transmission time wTx ). Therefore, BB-TDMA would not be able to support services that require a mean access delay, D − wTx , of less than 50 frames for Rcell = 1 km, while RTSO and FSA achieve a mean access delay close to zero albeit at diminishing throughput. On the other hand, for a smaller cell radius, Rcell = 200 m (see Figure 12.11), BB-TDMA is also capable of supporting delay sensitive services. Figure 12.12 shows plots of maximum throughput versus the uplink/ downlink asymmetry ratio for BB-TDMA, FSA and RTSO for Rcell = 200 m and 1 km. The results show that for FSA the peak throughput is not affected by the asymmetry. This is due to the fact that the interference scenario at the receiver remains unchanged with changes in asymmetry (see interference analysis in Section 12.4.3). For RTSO, the peak throughput is maximised for asymmetry ratios of unity and zero, in which there is no same entity (BS-to-BS or MS-to-MS) interference, and thus achieves the same peak throughput as FSA. However, its throughput declines as the occurrence of same entity interference increases. Furthermore, due to the relative severity of the BS-to-BS interference compared to MS-to-MS interference RTSO performs worse for an asymmetry ratio of Pul = 0.9 than for a ratio of 0.1. BB-TDMA shows a similar trend as RTSO, for the same reasons, but with superior performance due to its interference mitigation mechanism. In the following, we apply the busy burst protocol to an OFDMA based air-interferface assuming a frequency selective and time-varying channel.
Cellular OFDMA-TDD
363
12.6 Busy-tone approach applied to cellular OFDMA/TDD 12.6.1 Existing CCI mitigation techniques It was shown in (Wong et al., 1999) how the spectral efficiency of an OFDMA system can be increased by using the frequency, time and spatial diversity of the wireless channel. However, this algorithm requires perfect instantaneous channel knowledge. A channel assignment algorithm that performs well even in fast-fading scenarios was presented in (Li and Liu, 2003), but it inherently requires high signalling overhead. The latter problem was addressed in (Gross et al., 2004); however the algorithm presented there only brings a significant improvement in very slowly varying channels with high correlation in time. The region-based allocation algorithm proposed in (Yun et al., 2005) tries to minimise the CCI in the system, but it does not allow decentralised operation and, like all the former algorithms, does not solve the hidden and exposed node problems. The hidden node problem leads to collisions and arises when a transmitter cannot sense that there is a transmission to a ‘victim’ receiver within its range due to high path loss from the transmitter in the interfered link. In an exposed node scenario, a transmitter holds back its transmission because it can sense another transmission within its range, even though the path loss to the potential victim receiver is too high and the two transmissions can be carried out concurrently, and this leads to a drop in the network capacity. Earlier protocols that solve the hidden and exposed node problems (Haas and Deng, 2002; Li and Liu, 2003) use outof-band signalling and this compromises their functionality in the frequency selective OFDMA. In (Nguyen et al., 2005), a decentralised dynamic channel assignment method is proposed for cellular OFDM-TDMA/TDD networks; however, in this protocol the frequency-selectivity of the channel is not taken into consideration. In the following section, a novel MAC protocol without the above drawbacks is proposed, which is shown to minimise the CCI and increase the throughput of the network.
12.6.2 Busy burst based algorithm To mitigate severe CCI a combined medium access and dynamic subcarrier selection and adaptation algorithm for an OFDM/TDD network is proposed. The basic working mechanism is shown in Figure. 12.14. Before sending data, the transmitter senses a particular time-multiplexed channel, the busy-tone channel, in order to find suitable subchannels for transmission, i.e. channels that do not affect co-existing, already established links. Due to the frequency selectivity of the broadband channel, some subchannels may
364
Next Generation Mobile Access Technologies
experience favourable fading and channel conditions with respect to the desired and the interference link. In the scenario illustrated in Fig 12.14, it is assumed that the mobile station MSTx 2 enters the network and wants to in the uplink, while at the same time MSRx transmit to the base station BSRx 2 1 Rx and MSRx 3 are already receiving in the neighbouring cell. Thus, MS1 and Tx MSRx 3 are potential victim receivers of MS2 . To avoid jamming, on successRx ful reception both receivers MSRx 1 and MS3 broadcast a busy-signal with known transmit power in a time-multiplexed mini-slot on those subchannels which are used for transmission. The new transmitter, MSTx 2 , listens to all subchannels of the busy-channel first and compares the received signal power on all the subchannels against a given threshold. If the busy-signal power is below the threshold, the actual channel gain to the potential victim receivers is small. This, in turn, means that when using that particular subcarrier only negligible interference would be caused. Clearly, the channel reciprocity which is only possible in TDD is the key enabler for this ‘implicit’ signalling. In the example, assume that the busy-signal received on subchannels #p and #q is below the threshold. Consequently, these subchannels are selected for transmission by MSTx 2 . The proposed MAC-frame structure which accounts for downlink and uplink transmission is depicted in Figure 12.13†. The upper part of this illustration shows transmission and reception of signals from the BS point of view. Similarly, the lower part depicts transmission and reception from the MS point of view. The structure of an uplink subframe is similar to the structure of a downlink subframe and skipped here for the sake of conciseness. Each subframe includes a busy-channel (one OFDM symbol) for busy-tone signalling. The MAC frame duration is selected so that it is within the coherence time of the channel. The entire procedure can be subdivided into two phases: (a) the link initialisation phase and (b) the continuous and dynamic subchannel adaptation phase. During the initialisation phase a new communication link is established and subchannels have to be selected. In the adaptation phase, the target receiver has received data, but channel and interference variations make it necessary to select new subchannels for transmission. 12.6.2.1 Link initialisation It is assumed that BS m wants to set up a link to MS k in the MAC frame (i − 2), i.e. the transmission request arrival is at a random time position in the MAC frame (i − 2), and it is, therefore, not guaranteed that the busychannel can be heard during this MAC frame period. Therefore, the BS has † Figures 12.13, 12.16, 12.17, and 12.18 are reproduced with permission from: (Haas et al., 2006).
Cellular OFDMA-TDD
365
Frequency
Transmission direction of a BS
4
3 2 1
Time Busy tone OFDM symbol for downlink
Frequency
Transmission direction of a MS
4
3 2 1
Time Downlink subframe
Uplink subframe
(i1)th MACframe
ith MACframe
Received busy tone
Transmitted data symbol
Transmitted busy tone
Received data symbol
Selected grid for downlink
Fig. 12.13. A BS wants to transmit a packet to a MS. It first listens to the busysignal on all subchannels. In the example, it is assumed that the received busysignal power of the first, third and fourth subchannels falls below the given threshold which are subsequently selected for data transmission. It is further assumed that the SINR on the first subchannel in the downlink at the intended MS is unacceptable. Therefore, the MS does not broadcast the busy-signal on the first subchannel in the next MAC frame, i.e. the MS transmits the busy-signal only on the third and fourth subchannels. Because of channel variations on the interference link, the busy-signal on the second subchannel, at the MAC frame i, is received below the threshold. Therefore, the second subchannel is selected for downlink transmission in MAC frame i. Note that the first subchannel is released as the required SINR c at the intended MS was not achieved in MAC frame (i − 1). 2006 IEEE
366
Next Generation Mobile Access Technologies
to defer its transmission until the next MAC-frame, i.e. the MAC frame (i − 1) when it gets the first proper chance to listen to the busy-channel. The received busy tone is compared against a given threshold as explained before. Note that the received busy-signals at this point in time can only originate from non-intended MSs or BSs (nodes which may potentially suffer interference if transmission started), as the target receiver has not yet received any data and a busy-signal could not have been transmitted. It is assumed that the received busy-signal power of some subchannels falls below the threshold (first, third and fourth subchannel in the example of Figure. 12.13) which are subsequently selected for data transmission. This means that BS m commences data transmission to MS k on these subchannels. MS k determines the signal-to-interference plus noise ratio (SINR) on each of these subchannels. Based on the required QoS requirement for that particular transmission it will decide whether to reserve the respective subchannel, or whether to ‘release’ it. In the latter case it would not transmit the busy-signal on that corresponding subchannel, whereas in the former case it would reserve the respective subchannel by ‘protecting’ it using the busy-signal. Clearly, this protection is only feasible due to the use of TDD. Note that the SINR at a particular subchannel might be low either because this subchannel on the desired link is deeply faded, or because there is high interference resulting from another transmission. In Figure. 12.13, it is assumed that the SINR on the first subchannel in the downlink is unacceptable. Therefore, the busy-tone is not broadcast on this subchannel in the next MAC-frame, so the MS transmits the busy-tone only on the third and fourth subchannel. In order to mathematically model this behaviour, define ak,m l,i−1 as the channel assignment symbol for the subchannel l, at the MAC frame (i − 1) for the link between BS m and MS k. If this subchannel is assigned, then k,m ak,m l,i−1 = 1, otherwise, al,i−1 = 0. The outcome of this assignment is obtained by comparing the received busy-tone with the threshold expressed as: ˆ m |2 ≤ Ith , 1 if |B k,m l,i−1 (12.45) al,i−1 = 0 otherwise, ˆ m is the received busy-tone signal at BS m on the subchannel l in where B l,i−1 the MAC frame (i − 1). The threshold, Ith , is a measure for the interference that this transmission would cause to other receivers in the network. At the MAC frame (i − 1), the receiver, MS k, estimates the SINR and decides if this subchannel is to be reserved. The outcome of this decision
Cellular OFDMA-TDD
367
k,m k is above ˘l,i−1 is described by bk,m l,i−1 , where bl,i−1 = 1 if the estimated SINR γ
the required SINR γreq , otherwise bk,m l,i−1 = 0, described by: k ≥ γreq ), γl,i−1 1 if (ak,m k,m l,i−1 = 1) and (˘ bl,i−1 = 0 otherwise.
(12.46)
Note that the decision for the value of ak,m l,i−1 is made by the transmitter to mitigate CCI, whereas the decision for the value of bk,m l,i−1 is made by the receiver to ensure that the required SINR is maintained. It is assumed that the receiver detects ak,m l,i−1 without errors. 12.6.2.2 Dynamic subchannel adaptation For any MAC frame greater than or equal to i, the received busy-signal powers are composed of the signal powers of the intended user, MS k, and the busy-signal powers of all other entities which are potentially subject to interference. This means that the busy-signal power for the subchannels used is different from that in the MAC frame (i − 1) in which the communication between BS m and the MS k was initiated, in the way that the intended receiver, MS k, has not transmitted a busy-signal in MAC frame (i − 1). The received busy signal in the downlink sub-frame of the MAC frame i can be written as follows: k ′ ,m ′ k ′ ,m k ˆ m = H k,m B k bk,m + Hl,i Bl,i bl,i−1 , (12.47) B l,i l,i l,i−1 l,i ∀k ′ =k
k and B ˆ m are the transmitted and the received busy-tone on the where Bl,i l,i lth subchannel and the MAC frame i of MS k and BS m, respectively. The k,m represents the CTF (channel transfer function) coefficient for notation Hl,i the subchannel l and the MAC frame i of the transmission between BS m k ′ ,m is the CTF and MS k, i.e. the desired link. Similarly, the symbol Hl,i coefficient between terminal k ′ (it could be a MS or BS of a co-existing link) and entity m. Note that this algorithm does not require channel knowledge as the decision is solely based on received busy-signal levels. When the transmitter makes its selection of a subchannel that has not been used in the previous frame, it chooses a subchannel with low interference level in the busy slot. If the receiver detects the data correctly and replies with a busy-tone, this will cause a surge (modelled as a busy-tone margin) in the busy tone level on the particular subchannel in the following frame(s). This will indicate the transmitter that the receiver confirms the subchannel is usable. Therefore,
368
Next Generation Mobile Access Technologies
an estimate of bk,m l,i−1 can be obtained from (12.47) as follows, ˆ m ) > Δ · SIRtg , ˆm − B 1 if (B l,i−1 l,i ˆbk,m = l,i−1 0 otherwise,
This principle is illustrated in
Received busy signal power
Received busy signal power
where Δ is the busy-tone margin. Figure. 12.14
(12.48)
Ith
Ith
Subchannels
Set
Subchannels
Set
Set
(a) Received busy signal before busy-tone (b) Received busy signal after busy-tone transtransmission of intended receiver mission of intended receiver.In the example, it is assumed that the right subset of A does not fulfil the SINR requirement
Fig. 12.14. Busy signal reception before and after the intended receiver has transmitted the busy-tone.
The condition for the subsequent subchannel assignment on the desired link between BS m and the MS k for subsequent MAC frames is given as follows: ˆk,m ˆm 2 1 if (¯ ak,m k,m l,i−1 |Bl,i | ≤ Ith ) or (bl,i−1 = 1), (12.49) al,i = 0 otherwise, where a ¯ is the logical complement of a. In (12.49),the condition ˆm 2 (¯ ak,m l,i−1 |Bl,i | ≤ Ith ) means the subchannel l has not been used in the previous MAC frame (i − 1) and the received busy-tone on this subchannel at MAC frame i is lower than the given threshold. If this condition is fulfilled, then this subchannel is selected for data transmission in the following MAC frame. According to this mechanism, in the example in Figure. 12.13 the second subchannel is selected for downlink transmission in MAC frame i. The condition ˆbk,m l,i−1 = 1 means that the subchannel l has been selected in the previous MAC frame and the required SINR is maintained. In this case, the subchannel l remains selected for this link (third and fourth subchannel in the example). Note that the first subchannel is released as the
Cellular OFDMA-TDD
369
required SINR γreq at MS k has not been achieved. Note that, the busytone transmission primarily serves as a reservation mechanism addressing potential other transmitters in the network. However, since all receivers, including the intended receiver, send the busy signal, this implicit feedback mechanism (referred to as reservation indicator in (Omiyi and Haas, 2004)) can additionally be used for a subchannel specific ARQ (automatic repeat request). Let C denote a set of all subchannels. The notations Aki and Bik are = 1 the sets of the subchannels within a arbitrary cell m for which ak,m l,i and bk,m = 1, respectively. In the proposed protocol, the assumption is l,i made that each subchannel can only be assigned to one user within a given cell (Jang and Lee, 2003). Thus, Aki ∩ Ani = ∅ for k = n, where k and n are the user indices. It is also clear that Bik ⊆ Aki ⊆ C.
12.6.3 Benchmark – random subchannel allocation In order to assess the performance of the new busy-tone OFDMA technique an alternative solution is used for comparison. Generally state-of-the-art OFDMA solutions do not directly account for CCI (Rohling and Gr¨ unheid, 1997; Wong et al., 1999). Most importantly and to the best knowledge of the authors no decentralised technique has been proposed which takes interference precautions, i.e. systematically avoids interference by having gained knowledge about interference that would be generated elsewhere if the particular transmitter would start sending a signal. A close family of related methods are carrier-sensing techniques such as CSMA (carrier-sensing multiple access), but sensing the channel for activity only tells the potential new transmitter something about other transmitters, and not about the more important associated receivers. This means that the new transmitter is not able to infer the level of interference it would cause to any receiver in the neighbourhood yielding the classical hidden node problem and exposed node problem. For comparison, a different dynamic subchannel assignment technique is used. The main principle of this state-of-the-art solution is depicted in Figure. 12.6.3. Initially users get randomly assigned blocks of subcarriers. Ideal frequency and time synchronisation is assumed. Subcarriers on which the required SINR is not achieved are ‘released’ for other users, i.e. these ‘released’ subcarriers are randomly reassigned to newly arriving users. As a consequence, closed-loop SINR feedback is required which is assumed to be perfect, i.e. no transmission errors and no delay.
370
Next Generation Mobile Access Technologies
User 1
New user User 2
User K
Link initialzation (1st MACframe)
Subcarrier adaptation (2nd MACframe)
Fig. 12.15. Dynamic subcarrier assignment on OFDMA based on SINR at the receiver.
12.6.4 System model A cellular network consisting of seven cells with a radius of 500 m is assumed. MSs are uniformly distributed in space. The following parameters taken from the WiMAX standard (Yaghoobi, 2004) are used: • bandwidth of the system, B = 20MHz; • sampling interval, ta = 1/B = 50ns; • FFT-length, NFFT = 256. The carrier frequency is fc = 1.9GHz. A multi-path channel with a maximum propagation delay of 0.45μs is considered. The Doppler frequency of each path is 5 Hz. The channel is therefore a slowly time-variant channel. The multi-path channels of different links are statistically independent. Perfect time and frequency synchronisation is assumed. Thus only CCI is present in the system. The path-loss model described in (Erceg et al., 1999) and (IEEE 802.16.3c01/29r4, 2001) is used: g = A + 10 γ log10 (d/d0 ) + ξ,
(12.50)
Cellular OFDMA-TDD
371
where A = 20 log10 (4πd0 /λ) with d0 = 100m, and λ is the wavelength. The quantity γ is the path-loss exponent with γ = (a − bhb + c/hb ), where hb is the height of the BS and is selected to be 80m. The constant quantities a, b, and c are selected from the terrain type A given in (Erceg et al., 1999). The lognormally distributed random variable ξ models shadowing effects and its variance is assumed to be 10 dB. Poisson distributed traffic is assumed with an average inter-arrival time of 0.1 ms and an average holding time of 0.15 s. The transmit power of all MSs and BSs is 30 dBm. The minimum SINR, γreq , which is used to select subchannels at the intended receiver is 16 dB. The length of a downlink sub-frame LDL is set to be equal to the length of a uplink sub-frame LUL , which is 20 OFDM symbols. Thus, a MAC frame consists of (LDL + LUL ) OFDM symbols, in which there are two busy-tone OFDM symbols reserved for busy-tone signalling for both downlink and uplink. The spectral efficiency of the system will be reduced by: ηp = 1 −
2 . LDL + LUL
(12.51)
The penalty factor ηp is taken into account in (12.53) for evaluation of the network throughput. The offered load of the network is defined as the average number of bits per second per cell which are requested to be transmitted. Let us suppose that there are M active mobile stations during a given OFDM symbol. The offered load of the network is then defined as: νM Mary Nmax [bits/s/cell], (12.52) L= Ts N C where ν is the traffic intensity, Mary is the number of bits per symbol, Ts is the symbol duration in seconds, NC is the number of cells in the network, and Nmax is the maximum number of subcarriers that can be assigned to one user. Clearly Nmax < NFFT , where NFFT is the total number of available subcarriers. Let |Aki | be the cardinality of set Aki . If |Aki | > Nmax , then Nmax subchannels will be randomly selected for data transmission from the |Aki | preselected subchannels. This constraint is necessary for reasons of fairness so to prevent situations when a single link uses up a large proportion of the bandwidth and forces the network to deny service to other users. 16-QAM (quadrature amplitude modulation) on all subchannels is assumed, i.e. Mary = 4. Furthermore, let us assume that MS k can successfully receive data from |Bik | subchannels (|Bik | ≤ Nmax )§, where |Bik | is the § A data packet is considered to be successfully received on a subchannel if the SINR corresponding to this subchannel is higher than γreq .
372
Next Generation Mobile Access Technologies
cardinality of the set Bik . The throughput, which is a random variable in bits per second per cell, can therefore be obtained as: Ti = Mary ηp
M 1 k |Bi | [bits/s/cell]. Ts N C
(12.53)
k=1
The data bits which had been transmitted but are then received on subchannels with SINR below the required γreq are rejected by the receiver and are considered lost for that particular link. Based on sets Aki and Bik , the rejection rate per MAC frame at the receiver can be described as: Ri = Mary ηp
M 1 (|Aki | − |Bik |) [bits/s/cell]. Ts N C
(12.54)
k=1
Note that in practice data on the subchannels with low SINR (below the given threshold) will not be lost because they can be reused on other links. This is possible because the associated busy-signal is not transmitted on these subchannels, which basically means that they remain unreserved.
12.6.5 Simulation results First, the effect of the busy-signal threshold on the system behaviour is demonstrated with the results in Figure. 12.17. Note that, background noise is neglected in this study. This appears reasonable because the network is interference limited due to network deployment with full frequency reuse. If the busy-signal threshold is very low only those subchannels are selected which produce negligible signal power at the transmitter. This means that either there is no transmission on these subchannels, or that the victim receiver is located at a very high distance from the transmitter that is scanning the busy channel. As a consequence, the likelihood that the SINR on these subchannels at the own receiver is above the required SINR is particularly high. However, only a few subchannels fulfill this condition, i.e. the system is over-cautious. As the busy-signal threshold increases, the number of subchannels in set k Aki increases up to the total number of subchannels, but at the same time the number of subchannels which are rejected increases as the interference sensitivity in the network decreases. If the busy-signal interference threshold is very high (e.g. −50 dBm), this basically means that no interference protection and awareness exist, and the system behaves almost as if only a random subchannel selection algorithm is executed. From the results in Figure. 12.17 it can be observed that there exists an optimum for the busy-signal threshold at around −95 dBm which
Cellular OFDMA-TDD
373
Fig. 12.16. Complementary cdf of throughput for different busy signal thresholds c 2006 IEEE
results in an increase of subchannel utilisation of about 30% over the case when only very little or no interference awareness is taken into account, i.e. a busy-signal threshold of −50dBm. In Figure. 12.16, the complementary cdf (cumulative density function) of throughput is depicted. For up to a throughput of around 20Mbps/cell, which corresponds to 25% of the maximum theoretical throughput (NFFT · Mary /Ts = 80 Mbps/cell), the throughput is independent of the actual busysignal thresholds due to low interference levels. However, the situation is significantly different when considering the 10th percentile. For an ‘optimum’ busy-signal threshold, a peak throughput of about 45Mbps/cell can be guaranteed. This means that 56.25% of the maximum possible data rate can be achieved by the proposed MAC protocol for the given minimum required SINR, γreq = 16dB. In other words, the spectral efficiency is about 2.25bits/Hz/cell. The proposed algorithm has also been compared to conventional OFDMA, i.e. an OFDMA system where each user is randomly assigned a fixed number of consecutive subchannels. As can be seen in Figure. 12.18,
374
Next Generation Mobile Access Technologies
Fig. 12.17. The 99th percentile of the number of subchannels selected by the transmitter, |Aki |, based on received busy-signal power (solid line), the number of channels out of set Aki , which fulfill the SINR requirement at the receiver, |Bik | (dashed line) and the number of rejected subchannels, |Aki | − |Bik | (dotted line) as a function c of the busy-signal threshold. 2006 IEEE
the proposed technique significantly outperforms the conventional OFDMA in terms of throughput. For the 10th percentile the throughput gain is about 36% ((45 − 33)/33) when assuming 16-QAM. Only for very low offered load, when there is redundancy in the network capacity, the conventional OFDMA is comparable to the proposed technique. For comparison, in Figure. 12.19 the throughput performance for 64-QAM (SINR target 26dB) is depicted. While the absolute throughput is reduced, as expected, the improvement of the busy-tone approach over random subchannel allocation is even more pronounced. Again, for the 10th percentile the throughput gain is now about 115% ((28−13)/13). This example demonstrates the powerfulness of the busy-tone concept. The more interference sensitive the system is, the higher the performance gains as result of the interference awareness of the algorithm. This clearly motivates the use of link adaptation in combination with the busy-tone mechanism.
Cellular OFDMA-TDD
375
Fig. 12.18. Comparison of random subchannel selection in OFDMA with new inc terference aware DCA algorithm assuming 16-QAM 2006 IEEE
12.7 Conclusions A detailed interference analysis for a cellular TDD-based network has been carried. It has been shown that results obtained from the analytical model and results obtained from Monte Carlo simulations perfectly match. Interference results for a C-based power control algorithm as well as for fixed transmit power have been produced. These results demonstrate that the TDD-specific BS↔BS interference is most detrimental in a cellular TDDbased network. Therefore, new MAC protocols and DSA algorithms which are specifically tailored to a TDD system are required. In this research, it has been shown that the novel time-multiplexed busy-tone concept provides a powerful fundamental mechanism for such MAC protocols and DCA algorithms. In this context, a new medium access and dynamic subchannel assignment technique for cellular and ad hoc OFDM-based systems has been proposed which exploits the frequency selectivity in broadband OFDM transmission. The key novelty is that a mechanism is put in place which allows new transmitters to determine the level of interference which they would cause to coexisting receiving neighbouring nodes. This is a very effec-
376
Next Generation Mobile Access Technologies
Fig. 12.19. Comparison of random subchannel selection in OFDMA with new interference aware DCA algorithm assuming 64-QAM
tive means to mitigate CCI. For this purpose the channel reciprocity offered by TDD is exploited. The protocol works in a fully decentralised fashion and can therefore ideally be applied to ad hoc and multihop communication (this is subject to further studies). Since every potential new transmitter is aware of the interference it will potentially cause, it can make an appropriate decision as to which subcarriers to use. This decision is based on a power threshold that is applied to the received busy-tone signal transmitted by every receiving node upon successful reception of a data burst. This eventually leads to a flexible, dynamic and self-organising FDMA component in such an OFDM-based air interface. It has been demonstrated with the initial results that the throughput gain as compared to purely SINR-based dynamic channel assignment techniques is significant up to 115% for 64-QAM. The interference awareness property of the algorithm and the results for higherorder modulation motivate the use of link adaptation in combination with the busy-tone approach.
Appendix 1 Derivation of T: Unconstrained Optimisation
The mean squared error in (10.53) can be rewritten as: ) * ` )T (UT d + v ` ) − 2dT (UT d + v `) , J = Ed,`v dT d + (UT d + v
(1.1)
` . By taking the expectation where the expectation is with respect to d and v ` and dropping the terms independent of T we are left first with respect to v with: ) * ˆ 2 J ′ = Ev` d − d = dT TT UT UTd − 2dT UTd
(1.2)
By taking the expectation over d: J
= Ed {J ′ } =
K K K
T
m,n
(U U)
T
m,i
T
i=1 m=1 n=1
n,i
−2
K K
U i,n T n,i .
(1.3)
i=1 n=1
Finally, J is minimised with respect to the each component of T: ∂E{J} ∂Tp,q T
p,q
(U UT)
= 2
K
n=1 p,q
(UT U)p,q T n,q − 2Rq,p = 0
= R
T
U UT = UT ,
(1.4)
which results in the solution: T = U−1 .
377
(1.5)
Bibliography
3GPP, TSG, RAN. (1998). Opportunity Driven Multiple Access (ODMA). Tech. rep. 3GPP, TSG, RAN. (1999a). Physical Channels and Mapping of Transport Channels onto Physical Channels (TDD). TS 25.221 V3.1.1 (1999-12). 3GPP, TSG, RAN. (1999b). Physical layer — Measurements (TDD). TS 25.225 V3.1.0 (1999-12). 3GPP, TSG, RAN. (2000a). Physical Layer Items not for Inclusion in Release ’99. Tech. Rep. TR 25.833 V1.1.0, ETSI. April. 3GPP, TSG, RAN. (2000b). Physical Layer Procedure (TDD) – TS 25.224. 3GPP, TSG, RAN. (2000c). Radio Resource Management Strategies. TS 25.922 V3.0.0 (2000-01). 3GPP, TSG, RAN. (2001). Physical Layer Procedures. TS 25.214 v4.3.0 (release 4). 3GPP, TSG, RAN. (2003). Spreading and modulation (TDD) (Release 6). TS 25.223 V6.0.0 (2003-12). 3GPP, TSG, RAN, WG4. (1999). Evaluation of Up– and Downlink Adjacent Channel Performance. TSGR4#2(99) 048. 3GPP, TSG, RAN, WG4. (2000). UTRA (UE); Radio Transmission and Reception. TS 25.102 V3.4.0 (2000–10). 3GPP, TSG, Services and System Aspects. (2004). Quality of Service (QoS) Concept and Architecture. TS 23.107 V6.1.0 (2004-01). Abu-Dayya A. A., Beaulieu N. C. (1991). Outage Probabilities of Cellular Mobile Radio Systems with Multiple Nakagami Interferers. IEEE Transactions on Vehicular Technology 40, 4 (November), 757–768. Acx A. G., Henriksson J., Hunt B. et al. (2006). Final Usage Scenarios. IST-2003-507581 WINNER Deliverable, Retrieved May 10, 2006, from https://www.ist-winner.org. 378
BIBLIOGRAPHY
379
Adachi F., Sawahashi M., Okawa K. (1997). Tree–structured Generation of Orthogonal Spreading Codes with Different Lengths for Forward Link in DS–CDMA Mobile. Electronics Letters 33, 1, 27–28. Ahlin L., Zander J. (1998). Principles of Wireless Communications. Studentlitteratur. Alamouti S. (1998). “a simple transmit diversity technique for wireless communications”. IEEE Journal on Selected Areas in Communications 16, pp 1451–1458. Alpha Concept Group ETSI SMG2. (1997). ODMA annex to alpha evaluation report. Tech. rep., Alpha Concept Group Document. December. Ananthapadmanabha R., Manoj B. S., Murthy C. S. R. (2001). Multi-hop cellular networks: The architecure and routing protocols. In VTC Fall. IEEE. Argyropoulos Y., Jordan S., Kumar S. P. R. (1999). Dynamic Channel Allocation in Interference–Limited Cellular Systems with Uneven Traffic Distribution. IEEE Transactions on Vehicular Technology 48, 1 (January), 224–232. Baier A., Koch W. (1991). Potential of CDMA for 3rd generation mobile radio. In Proceedings of the Mobile Radio Conference. 282–290. Barreto A., Fettweis G. (2001). Capacity increase in the downlink of spread spectrum systems through joint signal precoding. in Proc. ICC 4, 1142– 1146. Barreto A. N., Fettweis G. (2000). Performance improvement in DS-spread spectrum CDMA systems using a pre- and a post-RAKE. Proc. of 2000 International Zurich Seminar on Broadband Communications, 39–46. Barreto R. I. A., G.Fettweis. (2001). Transmitter precoding forSpreadSpectrum signals in frequency-selective fading channels. Proc. 3G Wireless 2001 , 939–944. Barrett M., Arnott R. (1994). “Adaptive antennas for mobile communications”. IEE Electronics and Communication Engineering Journal Vol 5, 4 (August), pp203–14. Beck R., Panzer H. (1989). Strategies for Handover and Dynamic Channel Allocation in Micro–Cellular Mobile Systems. In Proceedings of the 1989 IEEE Vehicular Technology Conference. Vol. 1. IEEE, 178–185. Boyd S., Vandenberghe L. (2003). Convex optimization. To Appear. Boyer J. (2001). Multihop wireless communications channels. M.S. thesis, Carleton University. Boyer J., Falconer D., Yanikomeroglu H. (2001). A theoretical characterization of the multihop wireless communications channel without diversity. IEEE .
380
BIBLIOGRAPHY
Brandt-Pearce M., Dharap A. (2000). Transmitter-based multiuser interference rejection for the down-link of a wireless CDMA system in a multipath environment. IEEE J. Select. Areas Commun. 18, 3 (Mar.), 407–417. Burke J. V., Lewis A. S., Overton M. L. (2001). Optimal stability and Eigenvalue multiplicity. Computer Mathematics. Burke J. V., Overton M. L. (2001). Variational analysis of non-Lipshitz spectral functions. Maths programming. Chan W. C., Lu T. C., Chen R. J. (1997). Pollaczek-Khinchin formula for the M/G/1 queue in discrete time with vacations. IEE Proc. Comput. Digit Tech. 144, 4 (July), 222–226. Chang K.-N., Kim J.-T., Yim C.-S., Kim S. (1998). An Efficient Borrowing Channel Assignment Scheme for Cellular Mobile Systems. IEEE Transactions on Vehicular Technology 47, 2 (May), 602–608. Chang R. W. (1966). “synthesis of band-limited orthogonal signals for multichannel data transmission”. Bell System Tech. J. 45, pp 1775–1796. Chaudhury P., Mohr W., Onoe S. (1999). The 3GPP Proposal for IMT– 2000. IEEE Communications Magazine 37, 12 (December), 72–81. Chebaro T., Godlewski P. (1992). About the CDMA Capacity Derivation. In Proceedings of the International Symposium on Signals, Systems and Electronics. 36–39. Chebaro T., Godlewski P. (1996). Average External Interference in Cellular CDMA Systems. IEEE Transactions on Communications 44, 1 (January), 23–25. Chen L., Yoshida S., Murata H., Hirose S. (1998). A Dynamic Timeslot Assignment Algorithm for Asymmetric Traffic in Multimedia TDMA/TDD Mobile Radio. IEICE Transactions : Fundamentals E81–A, 7 (July), 1358–1365. Chih-Lin I., Greenstein L. J., Gitlin R. D. (1993). A Microcell/Macrocell Cellular Architecture for Low– and High–Mobility Wireless Users. IEEE Journal on Selected Areas in Communication 11, 6 (August), 885–890. Choi R. L.-U., Lataief K. B., Murch R. D. (2000). CDMA pre–RAKE diversity system with base station transmit diversity. In Proceedings of the 2000 IEEE 51st Vehicular Technology Conference. Vol. 2. IEEE, 951– 955. Chu T.-P., Rappaport S. S. (1995). Overlapping coverage and channel rearrangement in microcellular communication systems. IEE Proceedings – Communication 142, 5 (October), 323–332. Cimini L. J., Foschini G. J., I C.-L. (1992). Call Blocking Performance of Distributed Algorithms for Dynamic Channel Allocation in Microcells. In
BIBLIOGRAPHY
381
Proceedings of the International Conference on Communications ICC’92. IEEE, 1327–1332. Cimini L. J., Foschini G. J., I C.-L., Miljanic Z. (1994). Call Blocking Performance of Distributed Algorithms for Dynamic Channel Allocation. IEEE Transactions on Communications 42, 8 (August), 2600–2607. Cooper G. R., McGillem C. D. (1986). Modern Communications and Spread Spectrum. McGraw–Hill. Cooper G. R., Nettleton R. W. (1978). A Spread–Spectrum Technique for High–Capacity Mobile Communications. IEEE Transactions on Vehicular Technology VT–27, 4 (November), 264–275. Corazza G. E., De Maio G., Vatalaro F. (1998). CDMA Cellular Systems Performance with Fading, Shadowing, and Imperfect Power Control. IEEE Transactions on Vehicular Technology 47, 2 (May), 450–459. COST231. (1999). Digital mobile radio towards future generation systems. COST. Cox D. C., Reudink D. O. (1974). Microwave Mobile Communications. IEEE Press, Chapter layout and control of high–capacity systems. Das S. K., Sen S. K., Jayaram R. (1997a). A dynamic load balancing strategy for channel assignment using selective borrowing in cellular mobile environment. Wireless Networks 3, 5 (October), 333–347. Das S. K., Sen S. K., Jayaram R. (1997b). A Distributed Load Balancing Algorithm for the Hot Cell Problem in Cellular Mobile Networks. In Proceedings of the 1997 6th IEEE International Symposium on High Performance Distributed Computing. IEEE, 254–263. Delta Concept Group ETSI SMG2. (1997). ODMA annex to delta evaluation report. Tech. rep., Delta Concept Group Document. December. Elkamy S. E. (1997). Wireless Portable Communications Using PRE– RAKE CDMA/TDD/QPSK Systems with Different Combining Techniques and Imperfect Channel Estimation. In Proceedings of the 1997 International Symposium on Personal, Indoor and Mobile Radio Communications PIMRC 97. 3, vol. 2. IEEE, 529–533. Erceg et al. V. (1999). An empirically based path loss model for wireless channels in suburban environments. IEEE J. Select. Areas Commun. 17, 7 (July), 1205 – 1211. Esmailzadeh R., Nakagawa E. M. (1993a). Pre–RAKE Diversity Combining for Direct Sequence Spread Spectrum Mobile Communications Systems. IEICE Transactions : Communications E76–B, 8 (August), 1008–1015. Esmailzadeh R., Nakagawa E. M. (1993b). Time Division Duplex Method of Transmission of Direct Sequence Spread Spectrum Signals for Power
382
BIBLIOGRAPHY
Control Implementation. IEICE Transactions : Communications E76– B, 8 (August), 1030–1038. Esmailzadeh R., Nakagawa M. (1993d). “pre-rake diversity combining for direct sequence spread spectrum mobile communications systems”. IEICE Trans. Commun. E76-B, 8 (August), 1008–1015. Esmailzadeh R., Nakagawa M. (Geneva 1993c). Pre-RAKE diversity combination for direct sequence spread spectrum communications systems. International Conference on Communications 1, 463–467. Esmailzadeh R., Nakagawa M., Sourour E. A. (1997). Time–Division Duplex CDMA Communications. IEEE Communications Magazine 4, 2 (April), 51–56. Esmailzadeh R., Sourour E., Nakagawa M. (1995). Pre–RAKE diversity combining in time division duplex CDMA mobile communications. In Proceedings of the 1995 6th IEEE International Symposium on Personal, Indoor and Mobile Radio Communications PIMRC’95. 3, vol. 2. Toronto, Canada: IEEE, 431–435. Esmailzadeh R., Sourour E., Nakagawa M. (1999). Pre-rake diversity combining in time division duplex CDMA mobile communications. IEEE Trans. on Veh. Technol. 48, 3 (May), 795–801. ETSI. (1998). UTRA TDD, physical layer procedures. Tech. rep., UMTS(xx.13) v0.1.0. September. ETSI 30.03, V3.2.0 (1998-04). (1998). Universal Mobile Telecommunications System (UMTS); Selection procedures for the choice of radio transmission technologies of the UMTS. TR 101 112. ETSI SMG2. (1998). The ETSI UMTS Terrestrial Radio Access (UTRA) ITU–R RTT Candidate Submission. Everitt D., Manfield D. (1989). Performance Analysis of Cellular Mobile Communication Systems with Dynamic Channel Assignment. IEEE Journal on Selected Areas in Communication 7, 8 (October), 1172–1179. Eylert B. (2000). The Extended UMTS Vision. In Proceedings of the UMTS 2000 Conference. 12. published at: http://www.umts-forum.org. Feher K. (1995). “Wireless Digital Communications : Modulation and Spread Spectrum Applications”. Prentice Hall, NJ,. Foschini G. (1996). “Layered space–time architecture for wireless communication in a fading environment when using multi–element antennas”. Bell Labs Technical Journal , pp41–59. Foschini G., Gans M. (1998). “On limits of wireless communications in a fading environment when using multiple antennas”. Kluwer Wireless Personal Communications Vol 6, 3 (March), pp 311–335.
BIBLIOGRAPHY
383
Foschini G. J., Miljanic Z. (1993). A simple distributed autonomous power control algorithm and its convergence. IEEE Transactions on Vehicular Technology 42, 4 (November), 641–646. Foschini G. J., Miljanic Z. (1995). Distributed autonomous wireless channel assignment algorithm with power control. IEEE Transactions on Vehicular Technology 44, 3 (August), 420–429. Galvan-Tejada G. M., Gardiner J. G. (2001). Theoretical Model to Determine the Blocking Probability for SDMA Systems. IEEE Transactions on Vehicular Technology 50, 5 (September), 1279–1288. Gameiro A., Morgado A., Fernandes J. (2001). Space-time pre-equaliser for the downlink channel of UMTS-TDD. IEE 3G 2001 Conference on Mobile Telecommunications Technologies, 99–103. Gantmacher F. R. (1959). Applications of the theory of matrices. Interscience publishers, Chapter 3. Garg S., Kalia M., Shorey R. (2000). MAC Scheduling Policies for Power Optimization in Bluetooth: A Master Drive TDD Wireless System. In Proceedings of the 2000 IEEE 51st Vehicular Technology Conference. IEEE, 196–200. Georgoulis S., Cruickshank D. (2001). Pre-equalization, transmitter precoding and joint transmission techniques for time division duplex CDMA. IEE 3G 2001 Conference on Mobile Telecommunications Technologies, 257–261. Gerlach D., Paulraj A. (1994). “adaptive transmitting antenna arrays with feedback”. IEEE Signal Processing Letters 1, 150–152. Ghosh A., Wolter D. R., Andrews J. G., Chen R. (2005). Broadband Wireless Access with WiMax/802.16: Current Performance Benchmarks and Future Potential. IEEE Communications Magazine 43, 2 (February), 129– 136. Gilhousen K. S., Jacobs I. M., Padovani R., Viterbi A. J., Weaver, Jr. L. A., Wheatly III C. E. (1991). On the capacity of a cellular CDMA system. IEEE Transactions on Vehicular Technology 40, 2 (May), 303–312. Goiser A. M. J. (1998). Handbuch der Spread–Spectrum Technik. Springer. Golden G., Foschini G., Valenzuela R., Wolniansky P. (1999). “Detection algorithm and initial laboratory results using V-BLAST space-time communication architecture”. IEE Electronics Letters Vol 35, No 1 (January), pp 14–16. Golub G. H., van Loan C. F. (1996). Matrix Computations. The John Hopkins University Press. Goodman D., Mandayan N. (2001). Network assisted power control for
384
BIBLIOGRAPHY
wireless data. In Proceedings of the 2001 IEEE 53rd Vehicular Technology Conference. 1022 – 1026. Goodman D. J., Valenzula R. A., Gayliard K. T., Ramamurthi B. (1989). Packet Reservation Multiple Access for local wireless communications. IEEE transactions on Communications 37, 8 (August), 885–890. Grant P., Thompson J., Mulgrew B. (1998). “Antenna arrays for cellular CDMA systems”. In “Spread Spectrum: Developments for the New Millenium”, I. O. F. Swarts, P. van Rooyen and M. Lotter, eds. Kluwer Academic Publishers, Chapter 3, pp 59–81. Gross J., Paoluzzi I., Karl H., Wolisz A. (2004). Throughput study for a dynamic OFDM-FDMA system with inband signaling. In Proc. IEEE VTC. Vol. 3. 1787 – 1791. Gudmundson M. (1991). Correlation Model for Shadow Fading in Mobile Systems. Electronics Letters 27, 23 (November), 2145–2146. Gupta P., Kumar P. R. (2000). The Capacity of Wireless Networks. IEEE Transactions on Information Theory 46, 2 (March), 388–404. Gurbuz O., Owen H. (2002). Power control based qos provisioning for multimedia in w-cdma. Wireless Networks 8, 37–47. Haardt M., Klein A., Koehn R. et al. (2000). The TD–CDMA Based UTRA TDD Mode. IEEE Journal on Selected Areas in Communication 18, 8 (August), 1375–1385. Haas A. (1927). “kortegolf outvangst ronder fading”,. Radio Nieuws, pp 1. Haas H. (2000). Interference analysis of and dynamic channel assignment algorithms in TD-CDMA/TDD systems. Ph.D. thesis, University of Edinburgh. Haas H., McLaughlin S. (2001a). A Dynamic Channel Assignment Algorithm for a Hybrid TDMA/CDMA-TDD Interface Using the Novel TSOpposing Technique. IEEE J. Select. Areas Commun. 19, 10 (October), 1831 – 1846. Haas H., McLaughlin S. (2001b). A Novel Decentralised DCA Concept for a TDD Network Applicable for UMTS. In Proceedings of the 2001 IEEE 53rd Vehicular Technology Conference. IEEE, 5 pages in CD–Rom. Haas H., McLaughlin S. (2004). A Derivation of the PDF of Adjacent Channel Interference in a Cellular System. IEEE Comm. Lett. 8, 2 (February), 102–104. Haas H., McLaughlin S., Povey G. J. R. (2000a). A Novel Interference Resolving Algorithm for the TDD TD-CDMA Mode in UMTS. In Proc. Personal, Indoor and Mobile Radio Communications 2000 11th International Symbosium. Vol. 2. IEEE, 1231–1235.
BIBLIOGRAPHY
385
Haas H., McLaughlin S., Povey G. J. R. (2000b). An Investigation on Capacity versus Guard–Bands in the TDD Mode of UMTS. In Proceedings of the 2000 IEEE 51st Vehicular Technology Conference. IEEE, 5p. on CD–Rom. Haas H., McLaughlin S., Povey G. J. R. (2000c). Capacity–Coverage Analysis of the TDD and FDD Mode in UMTS at 1920 MHz. IEE Proceedings : Communications 149, 1 (May). Haas H., Nguyen V. D., Omiyi P., Nedev N. H., Auer G. (2006). Interference aware medium access in cellular ofdma/tdd network. In Proceedings of the International Conference on Communications ICC’06. IEEE. Haas H., Povey G. J. R. (1998). Outage Probability of CDMA-TDD Micro Cells in a CDMA-FDD Environment. In Personal, Indoor and Mobile Radio Communications 1998, PIMRC ’98. 3, vol. 1. IEEE, 94–98. Haas H., Povey G. J. R. (1999). A Capacity Investigation on UTRA–TDD Utilising Underused UTRA–FDD Uplink Resources. In Proceedings of the 10th International Symposium on Personal, Indoor and Mobile Radio Communications PIMRC ’99. IEEE, A6–4. Haas H., Wegmann B., Flanz S. (2002). Interference diversity through random time slot opposing (RTO) in a cellular TDD system. In Proc. IEEE VTC. Vol. 3. 1384 – 1388. Haas Z., Winters J. H., Johnson D. S. (1997). Simulation Results of the Capacity of Cellular Systems. IEEE Transactions on Vehicular Technology 46, 4 (November), 805–817. Haas Z. J., Deng J. (2002). Dual busy tone multiple access (DBTMA)-a multiple access control scheme for ad hoc networks. IEEE Trans. Commun. 50, 6 (June), 975 – 985. H¨am¨al¨ ainen S., Holma H., Toskala A. (1996). Capacity Evaluation of a Cellular CDMA Uplink With Multiuser Detection. In Proceedings of the 1996 4th International Symposium on Spread Spectrum Techniques and Applications, ISSSTA’96. Vol. 1. IEEE, 339–343. H¨am¨al¨ ainen S., Slanina P., Hartman M., Lappetel¨ ainen A., Holma H., Salonaho O. (1997). A novel interface between link and system level simulations. In Proceedings of ACTS Summit. 506–604. H¨am¨al¨ ainen S., Toskala A., Laukkanen M. (1997). Analysis of CDMA Downlink Capacity Enhancements. In Proceedings of the 1997 International Symposium on Personal, Indoor and Mobile Radio Communications PIMRC 97. 3, vol. 1. IEEE, 241–245. Hanly S., Tse D. N. (1999). Power control and capacity of spread spectrum wireless networks. Automatica 35, 1987–2012.
386
BIBLIOGRAPHY
Hanly S. V. (1995). An algorithm for combined cell-site selection and power control to maximise cellular spread spectrum capacity. IEEE J. Select. Areas Commun. 13, 1332–1340. Hanly S. V. (1999). Congestion measures in DS-CDMA networks. IEEE Trans. Comm. 47, 3, 426–437. Harold T. J., Nix A. R. (2000a). Capacity enhancement using intelligent relaying for future personal communications systems. In Proc. VTC Fall. Paper 4.3.1.1. Harold T. J., Nix A. R. (2000b). Intelligent relaying for future personal communications systems. In IEE Colloquium Capacity and Range Enhancement Techniques for 3G. Hashem B., Sousa E. (1997). On the Capacity of a Cellular DS/CDMA System Under Slow Multipath Fading and Fixed Step Power Control. In Proceedings of the 1997 IEEE 6th International Conference on Universal Personal Communications ICUPC ’97. 2, vol. 2. IEEE, 352–355. Hassell-Sweatman C. H. W., Mulgrew B., Thompson J. S., Grant P. M. (2000). Multiuser detection for CDMA antenna array receivers using spatial equivalence classes. IEE Proceedings – Communication 147, 2 (April), 105–113. Hayashi M., k Miya, Kato O., Homma K. (1995). CDMA/TDD Cellular System Utilizing a Base–Station–Based Diversity Scheme. In Proceedings of the 1995 IEEE 45th Vehicular Technology Conference. 2, vol. 2. IEEE, 799–803. Haykin S. (1991). Adaptive Filter Theory. Prentice Hall. Helmberg C., Rendl R., Vanderbei R. J., Wolkowicz H. (1996). An interiorpoint method for semidefinite programming. SIAM Journal on Optimization 6, 342–361. Holma H., Povey G. J. R., Toskala A. (1999). Evaluation of Interference Between Uplink and Downlink in UTRA/TDD. In Proceedings of the 1999 50th IEEE Vehicular Technology Conference. Vol. 5. IEEE, 2616– 2620. Holma H., Toskala A. (2000). WCDMA for UMTS. John Wiley. Hons E., Khandani A., Tong W. (2002). An optimized transmitter precoding scheme for synchronouds DS-CDMA. in Proc. ICC 3, 1818–1822. Horikawa I., Kato O., Miya K., Hayashi M. (1997). Performance of Wideband CDMA with TDD Scheme. In Proceedings of the 1997 Asia Pacific Microwave Conference. IEEE, 145–148. Horn R. A., Johnson C. R. (1985). Matrix Analysis. Cambridge University Press.
BIBLIOGRAPHY
387
IEEE 802.16.3c-01/29r4. (2001). Channel Models for Fixed Wireless Applications. retrieved November 3, 2005, from: http://grouper.ieee.org/groups/802/16/tg3/contrib. Jakes, W. C., Ed. (1974). Microwave Mobile Communications. IEEE Press. Jalali A., Mermelstein P. (1994). Effects of Diversity, Power Control, and Bandwidth on the Capacity of Microcellular CDMA Systems. IEEE Journal on Selected Areas in Communication 12, 5 (June), 952–961. Jang J., Lee K. B. (2003). Transmit power adaptation for multiuser OFDM systems. IEEE J. Select. Areas Commun. 21, 2 (February), 171 – 178. Jang W. M., Vojˇci´c B. R., Pickholtz R. L. (1998). Joint transmitter-receiver optimization in synchronous multiuser communications over multipath channels. IEEE Trans. Commun. 46, 2 (Feb.), 269–278. Jeon H., Kwon S., Kang C. (1998). Reverse Link Capacity Analysis of a DS– CDMA Cellular System with Mixed Rate Traffic. IEICE Transactions : Communications E81–B, 6 (June), 1280–1282. Jeon W. S., Jeong D. G. (2000). Comparison of Time Slot Allocation Strategies for CDMA/TDD Systems. IEEE Journal on Selected Areas in Communication 18, 7 (July), 1271–1278. Jeong W., Kavehrad M. (2002). Cochannel Interference Reduction in Dynamic-TDD Fixed Wireless Applications, Using Time Slot Allocation Algorithms. IEEE Trans. on Comms. 50, 10 (October), 1627–1636. Joham M., Utschick W. (2000). Downlink processing for mitigation of intracell interference in DS-CDMA systems. IEEE 6th Int. Symp. on SpreadSpectrum Tech. & Applic.,New Jersey,USA, 15–19. Jorguseski L., Farserotu J., Prasad R. (2001). Radio resource allocation in third-generation mobile communication systems. IEEE Communications Magazine, 117–123. Jorguseski L., Punt J. B., Sparreboom D. (1999). A Comparison of Dynamic Channel Allocation Algorithms. In Proceedings of the FRAMES Workshop 1999. 131–136. Jovanovi´c V. M., Gazzola J. (1997). Capacity of Present Narrowband Cellular Systems: Interference–Limited or Blocking–Limited? IEEE Personal Communications 4, 6 (December), 42–51. Jubin J., Tornow J. (1987). The DARPA packet radio network protocols. Proceedings of the IEEE 75, 1 (January), 21–32. Jung P., Baier P. W., Steil A. (1993). Advantages of CDMA and Spread Spectrum Techniques over FDMA and TDMA in Cellular Mobile Applications. IEEE Transactions on Vehicular Technology 42, 3 (August), 357–364.
388
BIBLIOGRAPHY
Jung P., Blanz J. (1995). Joint Detection with Coherent Receiver Antenna Diversity in CDMA Mobile Radio Systems. IEEE Transactions on Vehicular Technology 44, 1 (February), 76–88. Kadous T. A., Sourour E. E., Elkamy S. E. (1997). Comparison Between Various Diversity Techniques of the PRE–RAKE Combining System in TDD/CDMA. In Proceedings of the 1997 47th IEEE Vehicular Technology Conference. 3, vol. 3. IEEE, 2210–2214. Kchao C., St¨ uber G. L. (1993). Analysis of a Direct–Sequence Spread– Spectrum Cellular Radio System. IEEE Transactions on Communications 41, 10 (October), 1507–1516. Kikuchi F., Suda H., Adachi F. (2000). Effect of Fast Transmit Power Control on Forward Link Capacity of DS–CDMA Cellular Mobile Radio. IEICE Transactions : Communications E83–B, 1 (January), 47–55. Kim C.-S., Kim W.-Y., Park O.-S. (1997). On the Capacity of W–CDMA System with Different Bandwidths in a Realistic Mobile Channel. In Proceedings of the 1997 IEEE 6th International Conference on Universal Personal Communications ICUPC ’97. 2, vol. 2. IEEE, 613–617. Klang, G., Ed. (2006). Identification of Radio-Link Technologies. IST2003-507581 WINNER Deliverable, Retrieved May 10, 2006, from https://www.ist-winner.org. Klein A. (1996). Multi-user detection of CDMA signals-algorithms and their application to cellular mobile radio. Fortschritt-Berichte VDI. Klein A., Baier P. (1993). Linear unbiased data estimation in mobile radio systems applying CDMA. IEEE J. Select. Areas Commun. 11, 7 (Sep), 1058–1066. Klein A., Kaleh G. K., Baier P. W. (1995). Zero Forcing and Minimum Mean–Square–Error Equalization for Multiuser Detection in Code– Division Multiple–Access Channels. IEEE Transactions on Vehicular Technology 45, 2 (May), 276–287. Klein A., Pirhonen R., Sk¨ old J., Suoranta R. (1997). FRAMES Multiple Access Mode 1 – Wideband TDMA with and without Spreading. In Proceedings of the 1997 International Symposium on Personal, Indoor and Mobile Radio Communications PIMRC 97. IEEE, 37–41. Klingenbrunn T., Mogensen P. (1999). Modelling Cross–Correlated Shadowing in Network Simulations. In Proceedings of the 1999 50th IEEE Vehicular Technology Conference. IEEE, 1407–1411. Kowalewski F. (2000). Joint predistortion and transmit diversity. IEEE Global Telecommunications Conference ’00 1, 245–249. Kudoh E. (1993). On the Capacity of DS/CDMA Cellular Mobile Radios
BIBLIOGRAPHY
389
under Imperfect Transmitter Power Control. IEICE Transactions : Communications E76–B, 8 (August), 886–893. Kunz D. (1999). Transitions from DCA to FCA Behavior in a Self– Organizing Cellular Radio Network. IEEE Transactions on Vehicular Technology 48, 6 (November), 1850–1861. Lee W. (1989a). “Mobile cellular Telecommunication Systems”. McGraw Hill, Inc. Lee W. C. Y. (1989b). Spectrum efficiency in cellular. IEEE Transactions on Vehicular Technology 38, 2 (May), 69–75. Lee W. C. Y. (1991). Overview of cellular CDMA. IEEE Transactions on Vehicular Technology 40, 2, 291–302. Lee W. C. Y. (1993). Mobile Communications Design Fundamentals, Second edn. John Wiley & Sons. Lee W. C. Y. (1994). Applying the Intelligent Cell Concept to PCS. IEEE Transactions on Vehicular Technology 43, 3 (August), 672–679. Lee W. C. Y. (1997). Mobile Communications Engineering. McGraw–Hill. Lee Y., McLaughlin S. (2004). Radio resource metric estimation for a tddcdma system supporting wireless internet traffic. Wirel. Commun. Mob. Comput. 4, 5, 513–528. Lee Y., McLaughlin S., Yun S. (2003). Radio resource metric estimation for wireless cdma communication systems maximising radio resource utilisation. In Proceedings of the IEEE Vehicular Technology Conference (VTC 2004-Spring). Lee Y., Tabulo M., Laurenson, McLaughlin S. (2003). Radio resource allocation with radio resource metric estimation for multimedia cdma systems. In Proceedings of CIC2003. Li G., Liu H. (2003). Downlink dynamic resource allocation for multi-cell OFDMA system. In Proc. IEEE VTC. Vol. 3. 1698 – 1702. Liberti J., Rappaport T. (1999). “Smart Antennas for Wireless Communications: IS-95 and Third-Generation CDMA Applications”. Prentice Hall, NJ. Liberti, Jr. J. C., Rappaport T. S. (1994). Analytical Results for Capacity Improvements in CDMA. IEEE Transactions on Vehicular Technology 43, 3 (August), 680–690. Lin Y.-D., Hsu Y.-C. (2000). Multihop cellular: A new architechcture for wireless communications. In INFOCOM 2000. IEEE. Lindstrom M. (2001). TDD resource allocation through inter-mobile interference avoidance. In Proc. IEEE VTC 2001-Spring. Vol. 2. 1027–1031. Liu H., Zoltowksi M. (1997). “Blind equalization in antenna array CDMA
390
BIBLIOGRAPHY
systems”. IEEE Transactions on Signal Processing Vol 45, 1 (January), pp161–172. Lo K.-R., Chang C.-J., Chang C., Shung C. B. (1997). A Combined Channel Assignment Strategy in a Hierarchical Cellular Systems. In Proceedings of the 1997 IEEE 6th International Conference on Universal Personal Communications ICUPC ’97. IEEE, 651–655. Lozano A., Cox D. C. (1999). Integrated Dynamic Channel Assignment and Power Control in TDMA Mobile Wireless Communication Systems. IEEE Journal on Selected Areas in Communication 17, 11 (November), 2031–2040. Lucatti D., Pattavina A., Trecordi V. (1997). Bounds and Performance of Reuse Partitioning in Cellular Networks. International Journal of Wireless Information Networks 4, 2, 125–134. Luediger H., Zeisberg S. (2000). User and Business Perspectives on an Open Mobile Access Standard. IEEE Communications Magazine, 160–163. Ma L., Sunay M. O., Honkasalo Z. C., Honkasalo H. (1997). A Simulation Bed to Investigate the Feasibility of a TDD DS CDMA Residential Indoor System Underlay. In Proceedings of the 1997 International Symposium on Personal, Indoor and Mobile Radio Communications PIMRC 97. 3, vol. 2. IEEE, 286–291. MacDonald V. (1979). The Cellular Concept. The Bell Systems Technical Journal 58, 1 (January), 15–43. Makrakis D., Murthy K. M. S. (1992). Spread slotted aloha techniques for mobile and personal satellite communication systems. IEEE Journal on Selected Areas in Communications 10(6), 985–1002. Meurer M., Baier P., Weber T., Lu Y., Papathanasiou A. (2000). Joint transmission:advantageous downlink concept for CDMA mobile radio systems using time division duplexing. IEE Electronics Letters 36, 900–901. Meyer C. M. (2001). Matrix Analysis and Applied Linear Algebra. SIAM. Michail A., Ephremides A. (2000). Energy Efficient Routing for Connection– Oriented Traffic in Ad–hoc Wireless Networks. In Proc. Personal, Indoor and Mobile Radio Communications 2000 11th International Symbosium. 2, vol. 2. IEEE, 762–766. Mihailescu C., Lagrange X., Godlewski P. (1999). Dynamic Resource Allocation For Packet Transmission in TDD TD–CDMA Systems. In Proceedings of the 1999 IEEE Vehicular Technology Conference. IEEE, 1737–1741. Miya K., Kato O., Homma K., Kitade T., Hayashi M. (1998). Wideband CDMA Systems in TDD–Mode Operation for IMT–2000. IEICE Transactions : Communications E81–B, 7 (July), 1317–1325.
BIBLIOGRAPHY
391
Mohr W. (1998). Investigation of the UMTS Radio Interface in the ACTS FRAMES Project. In Proceedings UMTS workshop. 1, vol. 1. IEEE ComSoc Germany, Schloss Reisensburg near Ulm, Germany: University of Ulm, 33–64. Mohr W. (2002). WWRF — The Wireless World Research Forum. Electronics & Communication Engineering Journal 14, 6 (December), 283–291. Mohr W., Kottkamp M. (1996). Downlink Performance of IS–95 DS–CDMA under Multipath Propagation Conditions. In Proceedings of the 1996 4th International Symposium on Spread Spectrum Techniques and Applications, ISSSTA’96. IEEE, 1063–1067. Moore E. E. (1965). Cramming more components onto integrated circuits. Electronics 38, 8 (April). Morgado A., Gameiro A., Fernandes J. (2002). Constrained space-time zeroforcing pre-equalizer for the downlink channel of UMTS-TDD. Personal, Indoor and Mobile Radio Communications 2002, PIMRC ’02 3, 1122– 1126. Morgado A., Pinho P., Gameiro A., Fernandes J. (2001). Pre-equalisation technique for the interference cancellation in the UMTS-TDD downlink channel. IEEE Vehicular Technology Conference 2001-Fall 2, 878–881. Moshavi S. (1996). Multi-User Detection for DS-CDMA Communications. IEEE Communications Magazine, 124–136. Naguib A. (1996). “adaptive antennas for cdma wireless networks”. Ph.D. thesis, Standord University. Naguib A., Paulraj A. (1995). “Performance enhancement and trade–offs of smart antennas in CDMA cellular networks”. In IEEE Vehicular Technology Conference (VTC), Chicago (USA). 40–4. Naguib A., Seshadri N., Calderbank A. (2000). “increasing data rate over wireless channels”. IEEE Signal Processing Magazine,. Nakajima N., Yamao Y. (2001). “development for 4th generation mobile communications”. Wireless Comm. and Mobile Computing 1, pp 3–12. Nanda S., Goodmann D. J. (1992). Third Generation Wireless Information Networks. Kluwer Academic Publishers, Chapter Dynamic Resource Acquisition: Distributed Carrier Allocation for TDMA Cellular Systems, 99–124. Narula A., Trott M., Wornell G. (1999). “Performance limits of coded diversity methods for transmitter antenna arrays,”. IEEE Transactions on Information Theory 45, 7, pp 2418–2433. Nasreddine J., Lagrang X. (2003). Time slot allocation based on a path gain division scheme for TD-CDMA TDD systems. In Proc. IEEE VTC 2003-Spring. Vol. 2. 1410–1414.
392
BIBLIOGRAPHY
Nesterov Y., Nemirovskii A. (1994). Interior-point polynomial algorithms in convex programming. SIAM Studies in Applied Mathematics. Nguyen V. D., Omiyi P. E., Haas H. (2005). Decentralised dynamic channel assignment for cellular OFDM/TDD networks. In Proc. Internat. OFDM Workshop. Vol. 3. 255 – 259. Ojanper¨ a T., Prasad R. (1998). An Overview of Air Interface Multiple Access for IMT–2000/UMTS. IEEE Communications Magazine 36, 9 (September), 82–95. Ojanper¨ a T., Ranta P. A., H¨ am¨al¨ ainen S., Lappetel¨ ainen A. (1997). Analysis of CDMA and TDMA for 3rd Generation Mobile Radio Systems. In Proceedings of the 1997 47th IEEE Vehicular Technology Conference. 3, vol. 2. IEEE, 840–844. Oloffson H., Almgren M., Johansson C., H¨ oo¨k M., Kronestedt F. (1997). Improved interface between link level and system level simulation applied to gsm. In Proceedings of the 1997 IEEE 6th International Conference on Universal Personal Communications ICUPC ’97. 79–83. Omiyi P., Haas H., Auer G. (2007). Analysis of TDD Cellular Interference Mitigation using Busy-Bursts. IEEE Trans. on Wireless Commun. 6, 7 (July), 2721 – 2731. Omiyi P. E., Haas H. (2004). Improving time-slot allocation in 4th generation OFDM/TDMA TDD radio access networks with innovative channelsensing. In Proc. IEEE ICC. Vol. 6. 3133 – 3137. Omura J. K., Yang P. T. (1992). Spread Spectrum S–CDMA For Personal Communication Services. In MILCOM’92. IEEE, 269–273. Ortigoza-Guerrero L., Lara-Rodriguez D. (1996). Dynamic channel assignment strategy for mobile cellular networks based on compact patterns with maximised channel borrowing. Electronics Letters 32, 15 (July), 1342–1343. Papathanasiou A., Meurer M., Weber T., Baier P. (2000). A novel multiuser transmission scheme requiring no channel estimation and no equalization at the mobile stations for the downlink of TD-CDMA operating in the TDD mode. IEEE Vehicular Technology Conference Fall 2000 , 203–210. Papoulis A. (1991). Probability, Random Variables, and Stochastic Processes. McGraw–Hill. Paulraj A. J., Ng B. C. (1998). Space–Time Modems for Wireless Personal Communications. IEEE Personal Communications 1, 5 (February), 36– 48. Paulraj A. J., Papadias C. B. (1997). Space–Time Processing for Wireless Communications. IEEE Signal Processing Magazine 14, 6 (November), 49–83.
BIBLIOGRAPHY
393
Pehkonen K., Holma H., Keskitalo I., Nikula E., Westman T. (1997). A Performance Analysis of TDMA and CDMA Based Air Interface Solutions for UMTS High Bit Rates Services. In Proceedings of the 1997 International Symposium on Personal, Indoor and Mobile Radio Communications PIMRC 97. 3, vol. 1. IEEE, 22–26. P´erez-Romero J., Alonso L., Agust´ı R. (2000). Average block error probability in the reverse link of a packet ds/cdma system under rayleigh fading channel conditions. IEEE Communications Letters 4, 4, 116–118. Perkins C. E. (2001). Ad hoc networks. Addison Wesley. Pickholtz R. L., Milstein L. B., Schilling D. L. (1991). Spread Spectrum for Mobile Communications. IEEE Transactions on Vehicular Technology 40, 2 (May), 313–322. Povey G. J. R. (1994). Capacity of a cellular time division duplex CDMA system. IEE Proceedings: Communications 141, 5 (October), 351–356. Povey G. J. R., Holma H., Toskala A. (1997). TDD–CDMA Extension to FDD–CDMA Based Third Generation Cellular System. In Proceedings of the 1997 IEEE 6th International Conference on Universal Personal Communications ICUPC ’97. 2, vol. 2. IEEE, 813–817. Prakash R., Shivaratri N., Singhal M. (1999). Distributed Dynamic Fault– Tolerant Channel Allocation for Cellular Networks. IEEE Transactions on Vehicular Technology 48, 6 (November), 1874–1888. Prasad R., Jansen M. G., Kegel A. (1993). Capacity Analysis of a Cellular Direct Sequence Code Division Multiple Access System with Imperfect Power Control. IEICE Transactions : Communications E76–B, 8 (August), 894–905. Prasad R., Kegel A. (1991). Improved Assessment of Interference Limits in Cellular Radio Performance. IEEE Transactions on Vehicular Technology 40, 2 (May), 412–419. Prehofer C., Bettstetter C. (2005). Self-Organization in Communication Networks: Principles and Design Paradigms. IEEE Communications Magazine 43, 7 (July), 78–85. Proakis J. G. (1995). Digital Communications, 3rd edn. Electrical Engineering, McGraw-Hill. Proakis J. G., Salehi M. (1994). Communication System Engineering. Prentice Hall. Punt J. B., Sparreboom D., Brouwer F., Prasad R. (1998). Mathematical Analysis of Dynamic Channel Selection in Indoor Mobile Wireless Communication Systems. IEEE Transactions on Vehicular Technology 47, 4 (November), 1302–1313.
394
BIBLIOGRAPHY
Raleigh G., Cioffi J. (1998). “Spatio-temporal coding for wireless communication”. IEEE Transactions on Communications Vol 46, 3 (March), pp 357–366. Rappaport T. S. (1996). Wireless communications. Prentice Hall. Ringel U., Irmer R., Fettweis G. (2002). Transmit diversity for frequency selective channels in UMTS-TDD. Spread Spectrum Techniques and Applications, IEEE ISSSTA ’02 3, 802–806. Rohling H., Gr¨ unheid R. (1997). Performance comparison of different multiple access schemes for the downlink of an OFDM communication system. In Proceedings of the 1997 47th IEEE Vehicular Technology Conference. Vol. 3. 1365 – 1369. Rouse T., McLaughlin S., Band I. (2005). Congestion-Based Routing Strategies in Multihop TDD-CDMA Networks. IEEE J. Select. Areas Commun. 23, 3 (March), 668 – 681. Rouse T. S., McLaughlin S. (2001). Coverage-capacity analysis of Opportunity Driven Multiple Access in UTRA TDD. In IEE 3G 2001. 252–256. Rouse T. S., McLaughlin S., Band I. (2002). Capacity and power investigation of multi-hop relaying networks in TDD-CDMA based systems. In ICC 2002. Salbu Research and Development (Proprietary) Ltd. (1978). Adaptive communication system. SA Patent. Sampath A. S., Kumar P. S., Holtzman J. M. (1995). Power control and resource management for a multimedia cdma wireless system. In Proceedings of the 1995 6th IEEE International Symposium on Personal, Indoor and Mobile Radio Communications PIMRC’95. Sanada Y., Seki K., Wang Q., Kato S., Nakagawa M., Bhargava V. K. (1995). A Transmission Power Control Technique on a TDD–CDMA/TDMA System for Wireless Multimedia Networks. IEICE Transactions : Communications E78–B, 8 (August), 1095–1103. Sandouk A., Yamamoto T., Katayama M., Ogawa A. (1999). An integrated voice/data cdma packet communications with multi-code cdma scheme. IEICE Trans. Fundamentals E82-A(101). Sari H., Vanhaverbeke F., Moeneclaey M. (2000a). Channel Overloading in Multiuser and Single–User Communication. In Proc. Personal, Indoor and Mobile Radio Communications 2000 11th International Symbosium. IEEE, 1106–1111. Sari H., Vanhaverbeke F., Moeneclaey M. (2000b). Extending the Capacity of Multiple Access Channels. IEEE Communications Magazine 38, 1 (January), 74–82.
BIBLIOGRAPHY
395
Scharf L. (1991). Statistical signal processing : detection, estimation, and time series analysis. Addison-Wesley series in electrical and computer engineering, Addison-Wesley Pub. Co. Schilling D. L. (1991). Spread spectrum for commercial communications. In IEEE Communications Magazine. 66–79. Schwartz S. C., Yeh Y. S. (1982). On the Distribution Function and Moments of Power Sums With Log–Normal Components. The Bell System Technical Journal 61, 7 (September), 1441–1462. Seidenberg P., Althoff M. P., Schulz E., Herbster G. (1999). Statistics of the Minimum Coupling Loss in UMTS/IMT–2000 Reference Scenario. In Proceedings of the 1999 50th IEEE Vehicular Technology Conference. Vol. 2. IEEE, 963–967. Shannon C. (1949). “Communication in the presence of noise”. Proceedings of the IRE Vol 37, No 1 (January), pp 10–21. Shannon C. E. (1948). A Mathematical Theory of Communication. Bell System Technical Journal 27, 379–423 and 623–656. Shin S. M., Cho C.-H., Sung D. K. (1999). Interference–Based Channel Assignment for DS–CDMA Cellular Systems. IEEE Transactions on Vehicular Technology 48, 1 (January), 233–239. Sidenbladh T. (2000). Current status of 3G licensing. In Proceedings of the UMTS 2000 Conference. UMTS Forum, 15. published at: http://www.umts-forum.org. Steiner B. (2000). Performance Aspects of Transmit antenna diversity (TxAD) techniques for UTRA–TDD. In Proceedings of the 2000 IEEE 51st Vehicular Technology Conference. Vol. 2. IEEE, 1165–1169. Stimming C., Cheng T., Rohling H. (2005). Flexible Self-Organized Resource Allocation in Cellular OFDM Systems. In Proceedings of the 10th International OFDM Workshop InOWo’05. 41 – 45. Strang G. (1976). Linear Alegbra and its Applications. Academic Press. St¨ uber G. L., Yiin L.-B. (1991). Downlink Outage Predictions for Cellular Radio Systems. IEEE Transactions on Vehicular Technology 40, 3 (August), 521–531. Sunay M. O., Honkasalo Z. C., Hottinen A., Honkasalo H., Ma L. (1997). A Dynamic Channel Allocation Based TDD DS CDMA Residential Indoor System. In Proceedings of the 1997 IEEE 6th International Conference on Universal Personal Communications ICUPC ’97. 2, vol. 2. IEEE, 228– 234. Sung C. W., Wong W. S. (2001). Power control and rate management for wireless multimedia cdma systems. IEEE Transaction on Communications 49(7), 1215–1226.
396
BIBLIOGRAPHY
Takeo K. (1996). Improvement of Coverage Probability by Subband Scheme in CDMA Macro–micro Cellular System. In Proceedings of the 1996 7th IEEE International Symposium on Personal, Indoor and Mobile Radio Communications PIMRC’96. 3, vol. 1. IEEE, 93–97. Takeo K., Sato S., Ogawa A. (2000). Optimum Cell Boundary for Uplink and Downlink in CDMA Systems. IEICE Transactions : Communications E83-B, 4 (April), 865–868. Tam W. M., Lau F. C. M. (1997). Capacity Analysis of a CDMA Cellular System with Power Control Schemes. In Proceedings of the 1997 IEEE 6th International Conference on Universal Personal Communications ICUPC ’97. 2, vol. 2. IEEE, 608–612. Tan P. T., Soh C. B., Gunawan E., Soong B. H. (1998). Dynamic Flow Model for Borrowing Channel Assignment Scheme in Cellular Mobile System. Wireless Personal Communications 6, 3 (March), 249–264. Tang Z., Cheng S. (1994a). Interference cancellation for DS-CDMA systems over flat fading channels through pre-decorrelating. PIMRC’94 , 435–438. Tang Z., Cheng S. (1994b). Pre-decorrelating single user detection for CDMA systems. IEEE Vehicular Technology Conference ’94,IEEE 44th 2, 767–769. Tarokh V., Jafarkhani H., Calderbank A. (1999). “Space-time block codes from orthogonal designs”. IEEE Transactions on Information Theory Vol 45, 5 (July), pp 1456–1467. Tarokh V., Seshadri N., Calderbank A. R. (1998). “space-time codes for high data rate wireless communication: Performance criteria and code construction”. IEEE Transactions on Information Theory 44, pp 744– 764. Telatar E. (1999). “Capacity of multi-antenna Gaussian channels”. European Transactions on Telecommunications. Thompson J., Grant P., Mulgrew B. (1998). “Performance of antenna array receiver algorithms for CDMA”. Elsevier Signal Processing Vol 68, 1, pp 23–42. Thompson J., Grant P., Mulgrew B. (1999a). “Algorithms for coherent diversity combining of M -ary orthogonal signals”. IEEE Journal on Selected Areas in Communications Vol 17, 11 (November), pp 1886–1899. Thompson J., Grant P., Mulgrew B. (1999b). “The effects of user distribution on CDMA antenna array receivers”. IEE Proceedings on Radar, Sonar and Navigation 146, 1 (February), 25–30. Thompson J., Grant P., Mulgrew B. (2000). “downlink transmit diversity schemes for cdma networks”. IEE Proceedings on Communications, 147, 6 (December), 371–380.
BIBLIOGRAPHY
397
Thompson J., Grant P., Sweatman C. H., Mulgrew B. (2000). “adaptive antenna strategies for multimedia communications”. In In Proceedings of the Millennium Conference on Antennas and Propagation (AP 2000), Davos (Switzerland),. Thompson J. S., Grant P. M., Mulgrew B. (1996). Smart Antenna Arrays for CDMA Systems. IEEE Personal Communications 3, 5 (October), 16–25. TIA/EIA/IS–95. (1995). Mobile Station–Base Station Compatibility Standard for Dual–Mode Wideband Spread Spectrum Cellular System. Telecommunication Industry Association. Toh C.-K. (2001). Maximum battery life routing to support ubiquitous mobile computing in wireless ad hoc networks. IEEE Communications Magazine, 138–147. Tsoulos G., Beach M., McGeehan J. (1997). “wireless personal communications for the 21st century, european technological advances in adaptive antennas”. IEEE Communications Magazine 35, 9 (September), pp 102– 109. Tuttlebee W. H. W. (1992). Cordless Personal Communications. IEEE Communications Magazine 30, 12 (December), 42–53. van Rooyen, P., Lotter, M. et al. and van Wyk, D., Eds. (2000). “SpaceTime Processing for CDMA Mobile Communications”. Kluwer Academic Publishers. Veen B. V., Buckley K. (1988). “Beamforming: A versatile approach to spatial filtering”. IEEE ASSP Magazine, 4–24. Veeravalli V. V., Sendonaris A. (1999). The Coverage–Capacity Tradeoff in Cellular CDMA Systems. IEEE Transactions on Vehicular Technology 48, 5 (September), 1443–1450. Verdu S. (1998). Multiuser Detection. Cambridge University Press. Verdu S. (2000). Wireless Bandwidth in the Making. IEEE Communications Magazine, 53–58. Verdu S., (Shitz) S. S. (1999). “Spectral efficiency of CDMA with random spreading”. IEEE Transactions on Information Theory Vol 45, No 2 (March), pp 622–640. Viterbi A. (1995a). Principles of Spread Spectrum Communication. Addison–Wesley. Viterbi A. J. (1991). Wireless Digital Communication: A View Based on Three Lessons Learned. IEEE Communications Magazine 29, 8 (September), 33–36. Viterbi A. J. (1995b). Performance Limits of Error–Correcting Coding in
398
BIBLIOGRAPHY
Multi–Cellular CDMA Systems With and Without Interference Cancellation. In Code Division Multiple Access Communications, S. G. Glisic and P. A. Lepp¨ anen, eds. Kluwer Academic Publishers, 47–52. Viterbi A. J., Viterbi A. M. (1994). Other–Cell Interference in Cellular Power–Controlled CDMA. IEEE Transactions on Communications 42, 2/3/4 (February), 1501–1504. Viterbi A. J., Viterbi A. M., Gilhousen K. S., Zehavi E. (1994). Soft Handoff Extends CDMA Cell Coverage and Increases Reverse Link Capacity. IEEE Journal on Selected Areas in Communication 12, 8 (October), 1281–1287. Viterbi A. M., Viterbi A. J. (1993). Erlang Capacity of a Power Controlled CDMA System. IEEE Journal on Selected Areas in Communication 11, 6 (August), 892–900. Vodafone Ltd. (1997a). Initial results from ODMA simulations. Tech. rep., Epsilon Concept Group Document. October. Vodafone Ltd. (1997b). Notes on the simulation of ODMA. Tech. rep., Epsilon Concept Group Document. August. Vodafone Ltd. and Salbu (pty) Ltd. (1996). ODMA - Opportunity Driven Multiple Access. Tech. rep., Sophia Antipolis Workshop. December. Vodafone Ltd. and Salbu (pty) Ltd. (1997a). Characteristics of Opportunity Driven Multiple Access. Tech. rep., Epsilon Concept Group Document. March. Vodafone Ltd. and Salbu (pty) Ltd. (1997b). ODMA - system gain from fast fading. Tech. rep., Epsilon Concept Group Document. May. Vojˇci´c B. R., Jang W. M. (1998). Transmitter precoding in synchronous multiuser communications. IEEE Trans. Commun. 46, 10 (Oct.), 1346– 1355. Wallich P. (2002). Digital hubub. IEEE Spectrum 39, 7 (July). Walsh G. (1975). Methods of optimization. John Wiley & Sons Inc. Wang W., Wu X., Yin Q. (2001). A novel space-time pre-multiuser downlink transmission scheme suitable for TDD-CDMA systems. IEEE Vehicular Technology Conference 2001-Spring 2, 185–189. Wathan F., Irmer R., Fettweis G. (2002). On transmitter-based interference mitigation in TDD-downlink with frequency selective fading channel environment. Proc. Asia-Pacific Conference on Communications (APCC), Bandung, Indonesia. West K. A., St¨ uber G. L. (1994). An Aggressive Dynamic Channel Assignment Strategy for a Microcellular Environment. IEEE Transactions on Vehicular Technology 43, 4 (November), 1027–1038.
BIBLIOGRAPHY
399
Wie S.-H., Cho D.-H. (2001). Time slot allocation scheme based on a region division in CDMA/TDD systems. In Proc. IEEE VTC 2001-Spring. Vol. 4. 2445–2449. Wilson R. J. (1987). Introduction to Graph Theory. Longman Scientific & Technical. Winters J. (1987). “On the capacity of radio communication systems with diversity in a Rayleigh fading environment”. IEEE Journal on Special Areas in Communications Vol 5, 5 (June), pp871–8. Winters J., Salz J., Gitlin R. (1994). “The impact of antenna diversity on the capacity of wireless communication systems”. IEEE Transactions on Communications Vol 42, 2/3/4 (February/March/April), pp1740–50. Winters J. H. (1998). Smart Antennas for Wireless Systems. IEEE Personal Communications 1, 5 (February), 23–27. Wong C. Y., Cheng R. S., Lataief K. B., Murch R. D. (1999). Multiuser OFDM with adaptive subcarrier, bit, and power allocation. IEEE J. Select. Areas Commun. 17, 10 (October), 1747 – 1758. Wong D., Lim T. J. (1997). Soft Handoffs in CDMA Mobile Systems. IEEE Personal Communications, 6–17. Wong W., Sousa E. S. (1998). Frequency Selection Strategies for Hybrid TDD/FDD–CDMA Cellular Networks. In Proceedings of the International Conference on Universal Personal Communications ICUPC ’98. IEEE, 1152–1156. Wong W., Sousa E. S. (1999). Feasibility Study of TDD– and FDD–CDMA Frequency Sharing Cellular Networks. In Proceedings of the International Conference on Communications ICC’99. IEEE, 531–535. Wothke W. (1993). Testing structural equation models (pp. 256-93). Newbury Park, CA: Sage., Chapter Nonpositive definite matrices in structural modeling. Wozencraft J., Jacobs I. (1993). “Principles of Communication Engineering”. Waveland Press,. Wu J. S., Chung J. K., Sze M. T. (1997). Analysis of uplink and downlink capacities for two–tier cellular system. IEE Proceedings – Communication 144, 6 (December), 405–411. Xiao L., Johansson M., Boyd S. (2001). Simultaneous routing and resource allocation via dual decomposition. In 39th Annual Allerton Conference on Communication, Control, and Computing, Monticello, Illinois. Yaghoobi H. (2004). Scalable OFDMA physical layer in IEEE 802.16 wireless MAN. Intel Technology J. 8, 3, 201 – 212. Yamao Y., Suda H., Umeda N., Nakajima N. (2000). Radio Access Network Design Concept for the Forth Generation Mobile Communication System.
400
BIBLIOGRAPHY
In Proceedings of the 2000 IEEE 51st Vehicular Technology Conference. IEEE, 2285–2289. Yang G., Li C., Hu J. (2002). “td-scdma: Way to the future”. IEEE Vehicular Technology Society News 49, 3 (August), pp 13–18. Yates R. D., Huang C.-Y. (1995). Integrated power control and base station assignment. IEEE Transactions on Vehicular Technology 44, 3 (August), 638–644. Ye Y. (1997). Interior Point Algorithms: Theory and Analysis. Discrete Mathematics and Optimization. Wiley, New York. Yeung K. L., Yum T.-S. P. (1995). Cell Group Decoupling Analysis of a Dynamic Channel Assignment Strategy in Linear Microcellular Radio Systems. IEEE Transactions on Communications 43, 2/3/4 (February– April), 1289–1292. Yongsheng Z., Daoben L. (1997). Power Control based on Adaptive Prediction in the CDMA/TDD system. In Proceedings of the 1997 IEEE 6th International Conference on Universal Personal Communications ICUPC ’97. IEEE, 790–794. Yun S., Park S. Y., Lee Y., Alsusa E., Kang C. G. (2005). Spectrum efficient region-based resource allocation with fractional loading for FH-OFDMA cellular systems. IEE Electron. Lett. 41, 13 (June), 752 – 754. Yung Y.-F., Bentler P. M. (1994). Bootstrap-corrected adf test statistics in covariance structure analysis. British Journal of Mathematical and Statistical Psychology 47, 63–84. Zander J. (1992). Distributed cochannel interference control in cellular radio systems. IEEE Transactions on Vehicular Technology 41, 3, 305–311. Zander J., Frodigh M. (1992). Capacity Allocation and Channel Assignment in Cellular Radio Systems Using Reuse Partitioning. Electronics Letters 28, 5 (February), 438–440. Zeng M., Annamalai A., Bhargava V. K. (1999). Recent Advances in Cellular Wireless Communications. IEEE Communications Magazine 37, 9 (September), 128–138. Zvoran Z., Brady D. (1995). Suboptimal multiuser detector for frequencyselective Rayleigh fading synchronous CDMA channels. IEEE Trans. Commun. 43, 3 (Mar.), 154–157.
Index
16-QAM, 372 3GPP scenarios, 200 ACI, 54, 60, 61, 65, 69, 74, 75, 77, 78, 80, 82, 84, 86, 93, 180 ACIR, 64, 79, 80, 86, 88, 89, 94 ACLR, 64 ACS, 64 aggregated data rate, 260 antenna arrays, 300 downlink beamforming, 322 fixed beam, 314 Max SNR beamforming, 322 space-time 2D-RAKE, 314 space-time combiner, 314 ULA, 306 ARQ, 369 asynchronous, 55 beamwidth, 317 BER, 278, 279 BLAST, 331 Bluetooth, 16, 37 capacity, 15, 19, 22, 23, 54, 55, 57, 59, 65, 77, 79, 149 capacity v coverage, 79, 175 comparisons, 333 degree of confidence, 261, 263 multiuser, 310 pole capacity, 56, 58, 60, 78, 161, 187 relative capacity, 58, 79 Shannon, 308 SIMO capacity, 309 theoretical capacity, 59 theoretical upper capacity, 61 CCI, 54, 61, 89, 90, 363 CDMA, 15, 136 channel models, 303 angle spread, 305 angular spread, 309 delay spread, 305
doppler, 305 large scale propagation effects, 303 multipath, 304 small scale propagation effects, 303 uplink, 306 Cholesky, 174 COST 231, 168 CSMA, 369 DCA, 89, 95, 100, 113, 118, 138, 214, 217, 363 centralised DCA, 95, 104, 111, 112, 118, 119, 135 congestion, 223 decentralised DCA, 135, 138, 143, 148, 150, 152, 154 decentralisedDCA, 136 distributed DCA, 135, 151 dynamic sub channel-assignment, 370 dynamic subchannel allocation, 367 fast DCA, 143 multi-hop DCA, 214 decision feedback detector, 316 decorrelating detector, 316 DECT, 136 diversity antenna selection diversity, 322 downlink, 321 space-time transmit, 321 Doppler, 371 downlink, 16, 28, 30, 38 dynamic channel selection, 136 Eigenmode transmit/receive, 329 ETSI, 160 WWW, 236 FDD, 16, 30, 32, 37–39, 42–45, 150 FDMA, 15, 17, 136 fixed slot allocation, 350 idle frame, 142 interference BS↔BS, 345
401
402
Index
BS-BS, 341 inter-cellular, 339, 341 MS-to-MS, 346 same-entity, 346 interference cancellation, 296 ISI, 278, 281, 286 Kalman, 264, 265 linear decorrelator, 330 link adaptation, 228 MAI, 281, 287, 316 matched filter, 287 Maximum likelihood receiver, 330 minislot, 346 MMSE, 284, 312, 317 2D combining, 315 MRC, 315 MUD, 312 multi-hop, 157, 184, 186 multicode, 240, 258 multihop, 179, 337, 341 multislot, 240, 255 multiuser detection, 271–273, 284, 299 ODMA, 157–161, 165, 166, 168–171, 174, 177, 178, 181, 183, 184, 190, 228 OFDM, 326, 337, 371 spatial multiplexing, 327 OFDMA, 336, 363, 369 opposed synchronisation, 143 ORACH, 166 other entity interference, 138 outage probability, 259 Pareto, 236 path loss, 340 peer-peer, 187, 212 Perron-Frobenious, 196, 220 Poisson, 236 positive definiteness, 174 power control, 24, 26, 51, 52, 178, 196, 338, 342 closed loop, 32 downlink, 39 downlink power control, 147, 148, 151, 155 open-loop, 32 precoded blockwise, 276 precoding, 271–273, 275, 277–285, 287, 294, 296 bitwise, 275, 281, 287 blockwise, 275, 281 decorrelating pre-filters, 287 joint transmission, 281 power scaling, 279 pre-RAKE, 289 transmitter, 283 QoS, 260, 264 quasi synchronous, 148 quasi-synchronous, 141
RAKE, 283, 312 2D RAKE, 317 Pre-RAKE beamformer, 323 random time slot allocation, 350 real time traffic, 264 relaying, 159, 160, 167, 168, 175, 178, 180, 184 resource management, 228 mapping function, 229 RMMF, 250, 255 resource metrics, 228 BER, 235, 245, 249, 255 BLER, 235, 250 burst level SIR, 233 burst SIR, 246 channel characteristics, 232 estimation, 229, 261 link quality, 234 link quality information, 231 mapping function, 231 region, 229 routing, 186 congestion, 197, 214 congestion based, 196 interference, 192 local, 190, 192 path loss, 190 Salbu, 160 same entity interference, 138 SDMA, 19 shadowing, 172 Shannon, 15, 25, 27 SINR, 341, 345, 366 closed loop, 370 smart antennas, 301 Spread Spectrum, 15, 20, 22, 31 direct sequence, 20 FM, 21 frequency hopping, 21 time hopping, 21 successive interference cancellation, 316 synchronous, 55 frame synchronisation, 61, 62, 64, 65, 75, 89 opposed TS synchronisation, 76 synchronisation error, 76 synchronisation factor, 63, 64, 88, 138 TS opposing, 89, 94 system throughput, 253 TD-SCDMA, 16 TDD busy burst, 346, 350 busy-tone, 363, 367 OFDM, 363 TDMA, 15, 19, 136 TS opposing, 136, 138, 143 uplink, 16, 18, 22, 24, 30, 38 Viterbi decoder, 316 WiMAX, 16, 370 WLAN, 37