Third-Generation Systems and Intelligent Wireless Networking J.S Blogh, L Hanzo Copyright © 2002 John Wiley & Sons Ltd ISBNs: 0-470-84519-8 (Hardback); 0-470-84781-6 (Electronic) Adaptive Arrays in an FDMARDMA Cellular Network 4.1 Introduction Cellular networks are typically interference limited, with co-channel interference arising from cellular frequency reuse, ultimately limiting the quality and capacity of wireless networks [280,281] However, Adaptive Antenna Arrays (AAAs)are capable of exploiting the spatial dimension in order to mitigate this co-channel interference and thus to increase the achievable network capacity [3,6,38,242,250,282] Sincean AAA may receive signals with a high gain from one direction, whilst nulling, signals arriving from other directions, it is inherently suited to a CCI-limitedcellular network Thus a beammay be formedto communicate with the desired mobile, whilst nulling interfering mobiles [6] Assuming that each mobile station is uniquely identifiable, it is a relatively simple task to calculate the antenna array’s receiver weights, so as to maximise the received SINR The use of adaptive antenna arrays in a cellular network is an area of intensive research and adaptive antenna array’s have been studied widely in the context of both interference rejection and in single-cell situations [ 1,15,18,261,267,268] More recently, work has been expandedto cover the analysis and performancebenefits of using base stations equipped with adaptive antennaarrays across the whole of a cellular network [2,265,283] A further approach to improving the network performance the is employment of Dynamic Channel Allocation (DCA) techniques [284-2921, which offer substantially improved callblocking, packet dropping, and grade-of-service performance in comparison to Fixed Channel Allocation (FCA) A range of so-called distributed DCA algorithms were investigated by Cheng and Chuang [290] where a given physical channel could be invoked anywhere in the network, provided that the associated channelquality was sufficiently high As compromise schemes,locally optimised distributed DCA algorithms were proposed, for example, by Delli Priscoli et al [293,294], where the system imposed an exclusion zone for reusing a given physical channel around the locality, where it was already assigned In Sections 4.2.1-4.2.3 we briefly consider how an adaptive antennaarray may be mod193 194 CHAPTERCELLULAR 4.NETWORKS ADAPTIVE ARRAYS IN elled for employmentin a networklevel simulator, followed by a short overview of a variety of channel allocation schemes in Section 4.3 This section also provides abrief performance summary of the various channel allocation schemes based on our previous work [23,295], which suggested for the scenarios considered [23,295]that the Locally OptimisedLeast Interference Algorithm (LOLIA) provided the best overall compromise in network performance terms Section 4.4 presents atheoretical analysis of the performance of an adaptive antenna in a cellular network A summaryof several multipath propagation models given is inSection 4.5, with particular emphasis on the Geometrically Based Single-Bounce Statistical Channel Model [296,297] Thepotential methods of cellular network performance evaluation are described in Section 4.3.3.4, as are the parameters of the network simulated in later sections Simulation results for Fixed Channel Allocation (FCA) and two Dynamic Channel Allocation (DCA) schemes usingsingle element antennas,as well as two- and four-element adaptiveantenna arrays for Line-Of-Sight (LOS) scenarios are presented and analysed in Section 4.6.2 l Furthermore, simulation-specific details of the multipath model are given in Section 4.6.1, with the associated results obtained for the FCA and the LOLIA in the context of two, four and eight element adaptive antennaarrays presented in Section 4.6.2.2 Performance results for a network using power control over a multipath channel in conjunction with two andfour element adaptive antennaarrays are provided in Section 4.6.2.3, followed by the description of a network using Adaptive Quadrature Amplitude Modulation (AQAM) in Section 4.6.2.4 Performance results were also obtained for AQAM and the FCA algorithm as well as the LOLIA, with both two- and four-element adaptive antenna arrays Results using the ‘wraparound’ technique, describedin Section 4.6.1, which removes the cellular edge effects observed at the simulation area perimeter of a ‘desert-island’ scenario, are then presented in Sections 4.6.3.1-4.6.3.4 Finally, a performance summaryof the investigated networks is given in Section 4.7 4.2 Modelling Adaptive Antenna Arrays The interference rejection cability of an antenna array is determined by both the direction of arrival of the interference and the angle of arrival of the desired signal and therefore ultimately by the angular separation between the two The direction of arrival and angle of arrival may be used interchangably throughout ourdiscussions The numberof interferers and their signal strengths also affects the achievable attenuation of each of the interferers This section attempts to derive a simplerelationship between these factors for low-complexity modelling of an adaptive antennaarray 4.2.1 Algebraic Manipulation with Optimal Beamforming Given that the steering vector associated with the direction 8i of the ith source can be described by an L-dimensional complex vector si as [242], where L is the number of elements in the antenna array, and ti is the time delay experienced by a plane wave arriving from the i th source direction, O i , and measured from the antenna 4.2 MODELLING ADAPTIVE ANTENNA ARRAYS 195 element at the origin Then the correlation matrix, R, of the steering vector si, may be expressed as [242]: where pi is the power of the ith source, a: is the noise power and I is the identity matrix Assuming optimal beamforming under the constraint of a unit response in the wanted user's direction, then the weight vectorof the AAA is [242]: The array factor, F ( ) ,in the direction may be formulatedas [38]: L F(8)= C ZL'le-jWtl (01, (4.4) 1=1 Therefore, giventhat the desired signal arrives from the direction 80, and aninterfering signal arrives from the angle 81, the corresponding array responses are F ( & ) and F ( & ) , respectively Hence, the level ofinterference rejection, F(&,) - F ( & ) , when one desiredsignal and one interfering signal are received at a two-element antenna array, may be calculated using Equation 4.4to be: where the terms interference rejection is definedas the difference between the array response in the direction of the desired signal source andthat in the directions of the interfering source As can be seen from this equation, there is a non-linear relationship between the two angles of arrival and the achievable interference rejection Furthermore, the achievable interference rejection is independent of the desired signal's received power, po, and it is solely dependent upon the power of the interfering signal, p l Expanding this technique to either an antenna array having more elements or to catering for moreinterfering sources, or to multiple incident beams, led to overly complicated expressionswhich would be too complex to evaluate in real-time In order to avoid the associated complexity, the quantities required for interference rejection in a given scenario couldbe stored in lookup tables However, the size of the table required to store all of the information would be impractical For example, for the desired source, one dimension would be required for the angle of arrival and then another one for every interference source Two further table dimensions would be required to store the angle of arrival and interference power Therefore, the simple situation involving just oneinterferer, with a received power dynamic rangeof 40 dB, would require an array of 180 x 180 x 40 = 1,296,000 elements, at an angular resolution of l", and an interferer power resolution of dB For two interference sources this figure increases to 180 x 180 x 40 x 180 x 40 = 0.3312 x lo9 elements, which is clearly excessive CHAPTER ADAPTIVE ARRAYS 196 IN CELLULAR NETWORKS l -60 -30 30 SO Q0 l -90 Source angle (degrees) (a) Desired signal SNR = 3.0 dB, Interference SNR = 3.0 dB -60 -30 30 so 90 Source angle (degrees) (b) Desired signal SNR = 3.0 dB, Interference SNR = 12.0 dB Figure 4.1: Contour plots of interference rejection achieved using a four element antenna array with an inter-element spacing of X/2 using SMI beamforming with a reference signal length of 16 bits The angles of arrival of the signals from the desired source and the interfering source were sweptover the range, -90 degrees to +90 degrees 4.2.2 Using Probability Density Functions Due to the inherent complexities of performing large-scale network simulations, whilst invoking the required beamforming operations, we conducted an investigation into the probability distribution of the interference rejection ratio achieved by an adaptive antenna array For our initial studies a two element antenna array with the elements located A/2 apart was considered, with one desired source and one interfering source Therefore, the average interference rejection achieved in decibels, for a given source-direction and power as well as interferer-direction and power could be determined Unfortunately, as it can be seen from Figure 4.1 (a), the achievable interference rejection was not based upon a linear relationship between the twoangles of arrival Furthermore, Figure 4.l(b) illustrates that the interference rejection achieved was also related to the power, or the Signal-to-Noise Ratio (SNR), of the undesired interference source, which was dB or 12 dB As it was found in Section 4.2.1, attempting to construct a model or probability density function to cater for these parameters was not easily achievable Rather than attempting to find the Probability Density Function (PDF) relating the two angles of arrival and interference power to the interference rejection achieved, a brief study was initiated for determining the PDF of the interference rejection achieved with respect to the angular separation between the desired signal and interfering signal Figure 4.2 shows the probability density function of interference rejection achieved for one interference source and one desired source versus their angular separation As this figure shows, the distributionof the interferencerejection varies significantly, as the separation between the sources changes As a consequence of the PDF's dependence on the angular separation encountered, modelling the achievable interference rejection expressed in decibels 4.2 MODELLING ADAPTIVE ARRAYS ANTENNA 197 0.0 16 c Separation 0.014 5' 10' 0.012 A 20' X 40' h 0.01 * Y 0.008 Q h 0.006 * % a" 0.004 0.002 0.0 10 20 30 40 Interference rejection, dB 50 l t 60 Figure 4.2: The PDF of the interference rejection (dB) achieved for various angular separations of the desired signal and the interfering signal The angles of arrival of both signals were varied over the range of -90 to +90 degrees and were of equal power The antenna array consisted of two elements separated by X/2 is an arduous task Due to the complex natureof the PDF illustrated in Figure 4.2, an analysis of a smaller rangeof angles of arrival was conducted, in order to construct a piecewisevalid model The results are displayed in Figures 4.3(a) and 4.3(b) for angle of arrival spreads of f30" and *lo", respectively While these PDFs appearto be considerably simplerthan that in Figure 4.2, it was not possible to match the PDFs to any commonly known distributions Additionally, no information was available with regard to the correlation between successive interference rejection values For these reasons, and due to the difficulties associated with adding multipath, it was decided to cease work on constructing a suitable interference rejection model and instead to implement an actual SMI beamformer within the simulation program as described in the following section .8 4.2.3 Sample Matrix Inversion Beamforming The processof defining asuitable model of an adaptive antenna array was becoming increasingly complex, resulting in the decision to implement an SMI beamformer in the simulation software The SMI beamforming algorithm of Section 3.3.2.3 was chosen due to its inde- CHAPTER ADAPTIVE ARRAYS 198 IN CELLULAR NETWORKS 0.016 E Separation 0.014 0 loo e 'j E 0.012 A 20' x 40' z ,x 0.01 e 0.008 Q x 0.006 0.004 a 0.002 0.0 (a) 0.016 Angular s p r e a d = i ° l I 10 20 30 40 60 50 Interference rejection, dB (b) Angular s p r e a d = f l O O Figure 4.3: The PDF of the interference rejection achieved for the desired signal and the interfering signal angular separations of 5, 10 and 20 degrees The desired signal and the interfering signal were of equal power The antenna array consisted of two elements separated by X/2 4.3 CHANNEL ALLOCATION TECHNIQUES 199 pendence from the received signal strengths, as well as due to its fast convergence with the aid of fewdata samples andfor the sake of its good overall performance in terms of its interference rejection capability The reference signal was chosen to be eight bits in length as a compromise betweenthe quality of the sample correlation matrix, R, and the computational complexity required Since a cellular network is an interference limited system, the addition of noise to the received signal vector was neglected A result of this was that occasionally the correlation matrix, R, was non-invertible, which wasremedied by diagonally augmenting the matrix with a positive constant as it was suggested in [15,271,272] The addition of multipaths simply requiredthe direction of arrival, and the strength of the multipath rays at the antenna array to be determined before adding these received signal vectors to the total received signal vector of the antenna array In both the line-of-sight and the multipath scenarios, the transmidreceive channelwas assumed to be frequency invariant, thus allowingthe same antennapattern to be used in both the uplink andthe downlink 4.3 ChannelAllocationTechniques P.J Cherriman, L Hanzo' Channel assignment is the process of allocating a finite number of channels to the various base stations and mobile phones in the cellular network In a system using fixed channel assignment, the channels are assigned to different cells during the network planning stage, and the assignment is rarely altered to reflect changes in traffic levels A channelis assigned to a mobile at the commencement of the call and the mobile communicates with its base station on this channel until either the call terminates or the mobile leaves the current cell Dynamic channelallocation, however, assigns a channelthat best meets the channel selection criteria, which may be the channel experiencingthe minimum interference level, depending upon the cost function used With the growth in the number of subscribers to mobile telecommunications systems worldwide and the expected introductionof multimedia services in handheld wireless terminals, a tremendous demandfor bandwidth hasarisen Since bandwidthis scarce and becoming increasingly expensive,it must be utilized in an efficient manner The main limiting factor in radio spectrum reuse is co-channel interference In reduced cell-size micro/picocellular architectures, the frequency reuse distance is reduced, thereby increasing the capacity and area spectral efficiency of the system However, as the channel reuse distanceis reduced, the co-channel interference increases CO-channel interference caused by frequency reuse is the most severe limiting factor of the overall system capacity of mobile radio systems The most important techniquefor reducing co-channelinterference is power control, an issue, which will be discussed in detail in the context of adaptive modulation during our further discourse Interference cancellation techniques [298] or adaptive antennas [299-3011 can also be used to reduce co-channel interference However, a simpler and moreeffective technique used incurrent systems is employing sectorized antennas [302] Although handoversare necessary in mobile radio systems, they often cause several problems, and they constitute the major cause of calls being forcibly terminated As the cell size is decreased, the average sojourn time or cell-crossing time for a useris reduced Thisresults ' This section isbased on [ 15 l ] 200 CHAPTERCELLULAR NETWORKS ADAPTIVE ARRAYS IN in an increased number of handovers, requiring more rapid handover completion In practice a seamless handover is not always possible exceptwhen soft-handovers [303] are used in CDMA-based systems Rapid and numerous handovers require a fast backbone network between the base stations and the mobile switchingcenters, or they necessitate an increased number of mobile switching centers Clearly, the handover process is crucial with regard to the perceived Gradeof Service (GOS), and a wide range of different complexity techniques have been proposed, for example, by Tekinay and Jabbari [304] and Pollini [305]for the forthcoming future systems The related issue of time-slot reassignment was investigated by Bernhardt [306] 4.3.1 Overview of Channel Allocation The purposeof channel allocation algorithms is to exploit the variability of the radio channel propagation characteristics in order to allow increased efficiency radio spectrum utilisation, while maintaining requiredsignal quality The most commonly used signal quality measure is the signal-to-interference ratio (SIR), also known as the carrier-to-interference ratio (CIR) The signal quality measure that we have usedpreviously was the signal-to-interference+noise ratio (SINR) The SINRis approximately equalto the signal-to-noise ratio (SNR) in a noiselimited environment and approximately equalto the SIR in an interference-limited environment The radio spectrum is dividedinto sets of noninterfering physicalradio channels, which can be achieved using orthogonal timeor frequency slots, orthogonal user signature codes, and so on The channel allocation algorithm attempts to assign these physical channels to mobiles requesting a channel, such that the required signal quality constraints are met There are three main techniques for dividing the radio spectrum into radio channels The first is frequency division (FD), in which the radio spectrum is divided into several nonoverlapping frequency bands However, in practice the spectral spillage from one frequencyband to another causes adjacent channelinterference, which can be reduced by introducing frequency guard bands.However, these guard bandswaste radio spectrum,and hence there is a compromise between adjacent channelinterference and frequency band-packingefficiency Tighter filtering can help reduce adjacent channel interference, allowing the guard bands to be reduced The second techniqueis time division (TD), in which the radio spectrum is divided into disjunct timeperiods, which are usually termed time-slots However, using straight-forward rectangular windowingof the time-domain signal corresponds to convolving the signal spectrum with a frequency-domainsinc-function, resulting in Gibbs-oscillation Hence, in practical systems a smooth time-domain ramp-up and ramp-down function associatedwith a timedomain guard period is employed Therefore, there is a trade-off between complex synchronisation, time-domain guardperiods, and adjacent channelinterference The third technique for dividing the radio spectrum into channels is code division (CD) Code division multiple access (CDMA)[3941,307] has been used in military applications, in the IS-95 mobile radio system [308], and in the recently standardized Universal Mobile Telecommunications System (UMTS)[307,309] In code division, the physical channelsare created by encoding different users with different user signature sequences Inmost systemsacombination of these techniques is used Forexample, the PanEuropean GSM system [28] uses frequency division duplexing for up- and down-link trans- NIQUES ALLOCATION 4.3 CHANNEL 201 Channel Assignment Strategies Fixed Dynamic I Flexible Dynamicchannels Locally Distributed Static borrowing Simple borrowing Hybrid borrowlng e.g., LP-DDCA, LOLIA, LOMIA J Figure 4.4: Family tree of channel allocation algorithms missions, while accommodating eight TDMA users per carrier In this chapter, the term “channel” typically implies a physical channel,constituted by a time-slot of a given carrier frequency A widevariety of channel allocation algorithms have been suggestedfor mobile radio systems The majorityof these techniques canbe classified into one of three main classes: fixed channel allocation (FCA), dynamic channel allocation (DCA), and hybrid channel allocation (HCA) Hybrid channelallocation is constituted by a combinationof fixed and dynamic channel allocation, which is designed to amalgamate the best features of both, in order to achieve better performance or efficiency than I X A or FCA can provide Several channel allocation schemes and the associated trade-offs in terms of performance and complexityare discussed in detail in the excellent overview papers of Katzela and Naghshineh [310] and those by Jabbari and Tekinay et al [31l, 121 Figure 4.4 portrays the family tree for the main types of channel allocation algorithms, where the acronyms are introduced during our further discourse Zander [313] investigated the requirements and limitations of radio resource management in general for future wireless networks Everitt [314] compared various fixed and dynamic channel assignment techniques and investigated the effect of handovers in the context of CDMA-based systems 4.3.1.1 FixedChannel Allocation In fixed channel allocation (FCA), the available radio spectrum is divided into sets of frequencies Oneor more of these sets is then assignedto each basestation on a semipermanent basis The minimum distance betweentwo base stations, they have been assigned the same 202 CHAPTER ADAPTIVE ARRAYS IN CELLULAR NETWORKS set of frequencies is referred to as the frequency reuse distance This distance is chosen such that the co-channel interference is within acceptablelimits, when interferers are at least the reuse distance away from each other The assignment of frequency sets to base stations is based on a predefined reuse pattern The group of cells that contain one of each of the frequency sets is referred to as the frequency reuse cluster The grade of frequency reuse is usually characterized in terms of the number of cells in the frequency reuse cluster The lower the number of cells in a reusecluster, the more bandwidth-efficientthe frequency reuse pattern and the higher the so-called area spectral efficiency, since this implies partitioning the available total bandwidth in a lower numberof frequency subsets used in the different cells, thereby supporting moreusers across a givencell area However, small reuse clusters exhibit increased co-channelinterference, which has to be tolerated by the transceiver In FCA, the assignment of frequencies to cells is considered semipermanent However, the assignment can be modified in order to accommodate teletraffic demand changes Although FCA schemes are very simple, modifying themto adapt to changing traffic conditions or userdistributions can be problematic Hence,FCA schemes have to be designed carefully, in order to remain adaptable and scalable, as the number of mobile subscribers increases In this context, adaptability implies the ability to rearrange the network to provide increased capacity in a particular area on a long- or short-term basis, where scalability refers to the ability of easily increasing capacity across the whole network via tighter frequency reuse For example, Dahlinet al [3 151 suggested a reusepattern structure for the GSM system that can be scaled to meet increased capacity requirements, as the number of subscribers increases This is discussed in more detail in the overview paper by Madfors et al [316] Each measure invoked, in order to further increase the network capacity, increases the system’s complexity and hence becomes expensive Furthermore, such systems cannot be easily modified to provide increased capacityin the specific area of a traffic hot-spot on a short-term basis A commonly invoked reuse clustedpattern is the seven-cell reuse cluster, providing coverage over regular hexagonal shaped cells, which is shown in Figure 4.5 Each cell in the seven-cell reuse cluster has six first-tier co-channel interfering cells at a distanceD , the reuse distance By exploiting the simple hexagonal geometry seen in Figure 4.5 it can be shown that for the seven-cell cluster the reuse distance, D , is 4.58 times the cell radius T [ 1511 This reuse pattern supports the same numberof channels at each cell site, and hence the same system capacity Therefore, the teletraffic capacity is distributed uniformly across all the cells Since traffic distributions usually are not uniform in practice, such a system can leadto inefficiencies For example, under nonuniform traffic loading, some cells may have no spare capacity; hence, new calls in these cells are blocked However, nearby cells may have spare capacity Several studies have suggested techniques to find the optimal reuse pattern for particular traffic and users distributions, as exemplified by the work of Safak [317], on optimal frequency reuse with interference While such contributions are useful, a practical system would need to modify the whole network configuration every timethe traffic or user distributions changedsignificantly Therefore, suboptimalbut adaptable and scalable solutions are more desirable for practical implementations When the traffic distribution changes, an alternative technique to modifying the reuse pattern is referred to as channel borrowing,which is the subject of the next section ... Probability Density Function (PDF) relating the two angles of arrival and interference power to the interference rejection achieved, a brief study was initiated for determining the PDF of the interference... spreads of f30" and *lo", respectively While these PDFs appearto be considerably simplerthan that in Figure 4.2, it was not possible to match the PDFs to any commonly known distributions Additionally,... h 0.006 * % a" 0.004 0.002 0.0 10 20 30 40 Interference rejection, dB 50 l t 60 Figure 4.2: The PDF of the interference rejection (dB) achieved for various angular separations of the desired signal