ROYAL INSTITUTE OF TECHNOLOGY KUNGL TEKNISKA HÖGSKOLAN. Department of Signals, Sensors & Systems Signal Processing S STOCKHOLM

Transcription

1 A Comparson of Two Systems for Down Lnk Communcaton wth Antenna Arrays at the Base Per Zetterberg 95{1{1 In: IEEE Wreless Communcaton Systems Symposum, 1995 IR{S3{SB{9522 ROYAL INSTITUTE OF TECHNOLOGY Department of Sgnals, Sensors & Systems Sgnal Processng S-1 44 STOCKHOLM KUNGL TEKNISKA HÖGSKOLAN Insttutonen för Sgnaler, Sensorer & System Sgnalbehandlng 1 44 STOCKHOLM

2 A Comparson of Two Systems for Down Lnk Communcaton wth Antenna Arrays at the Base Per Zetterberg Royal Insttute of Technology S{1 44 Stockholm, Emal: Abstract In ths paper we wll compare the performance of the down-lnk part of two systems employng antenna arrays at the bases. One of systems uses a 1/3 channel reuse pattern but also reuses the channels three tmes wthn each cell. The other systems reuses all channels n all cells 1/1, but does not reuse channels wthn a cell. Thus, the maxmum number of users n the two systems are equal. In order to account for the ncreased nterference, the 1/3 system drects antenna pattern nulls aganst same cell co-channel users whle the 1/1 system drects nulls aganst co-channel users n other cells. The performance of the two systems are compared are as a functon of the azmuth angular wdth (seen from the base) of the multpaths generated at the moble. The result heren ndcates that wth ecent power control, nter-cell handover and synchronzaton the 1/1 system has better performance than the 1/3 system. I. Introducton The use of antenna arrays at the base statons has been proposed as a means of ncreasng the capacty (spectrum ecency) of moble cellular networks. For hgh moblty systems as GSM, AMPS and IS-54 the capacty enhancement achevable wth an antenna array wll be gven by the downlnk enhancement. Ths s because the processng applcable to the downlnk s lmted to beam steerng type of technques, whle n the uplnk hgh performance dversty combnng technques can be used. Downlnk cellular system performance are treated n [6], [7] and [1]. The three papers all shows that substantal capacty gans can be acheved but that the downlnk performance s very dependent on the propagaton characterstcs. Reduced channel reuse dstances s nvestgated n all three papers but [9] also looks at allocatng multple mobles on the same channel wthn the same cell. The results n [9] ndcates that multple mobles n a cell s more ecent than reduced reuse dstances. However, drected nulls aganst users n other cells s not consdered n [9]. In ths paper we compare a system whch uses a thrd of the avalable spectrum n a 12 o sector cell (1/3), wth a system that uses all the avalable spectrum (1/1) n all 12 o sectors. However, the 1/3 system reuses the channels three tmes wthn each cell and thus the maxmum number of users n the two systems are equal. The 1/1 system s operated wth and wthout nter-cell nulls. The nuence of uplnk power control s also nvestgated. The uplnk power control actually aects the downlnk performance snce t nuences whch nterferng mobles the base \sees" n the uplnk. Ths work was supported n part by the Swedsh Natonal Board for Industral and Techncal Development (NUTEK). Part of the work were conducted at and nanced by the Center of Personal Communcatons (CPK) at Aalborg Unversty, Denmark. Smulatons s performed assumng the ar nterface of GSM. Plots of power control ranges, uplnk dynamc range requrements, and downlnk outage probabltes (dened here as the probablty of unacceptable speech qualty) are gven. The outage probablty of the two systems are compared are as a functon of the azmuth angular wdth (seen from the base) of the multpaths generated at the moble. Ths s mportant snce the performance of beam steerng arrays s very senstve to the multpath propagaton. The results heren shows that the 1/1 system has better performance than the 1/3 system, provded that nullng, ecent handover and power control s employed. If nether of these three condtons are met, the 1/3 system s superor. Analytcal approxmatve expressons of the outage probabltes are also derved and compared wth smulaton results. A. Propagaton Modelng II. Prelmnares In ths secton, we dene the model of the multdmensonal channel between the lnear array of unformally spaced antenna elements at the base, and the sngle antenna element at the moble. The poston of the moble s dened by ts polar coordnates (r; ) seen from the base, where = o represents the broadsde drecton. Flat fadng (no tme dsperson) s assumed for smplcty. The transfer functon conssts of three factors: path loss, shadowng, and fast fadng. The path loss and the shadowng are common to all the antenna elements. The path loss s modeled as (R=r) where s the path loss exponent, r s the dstance between the base and the moble, and R s the cell radus. The shadowng s modeled by a factor L whch has log-normal dstrbuton [4]. The standard devaton of 1 log L s denoted by db (the mean s zero). The gan and the phase of the m antenna elements of the array are stacked nto a vector denoted ~v(; ). The vector ~v(; ) s a complex normally dstrbuted random vector wth mean zero and covarance matrx parameterzed by and, where s to be dened. Ths models the at fadng of the channel. The fadng s due to the superposton of a large number of rays. When the moble moves the phases of the rays wll appear random whch n turn mposes the fadng. The complex covarance matrx, ~R vv (; ), of ~v(; ) s derved n [9] under the assumpton that the azmuth angular perturbatons from of the multpaths are normally dstrbuted wth standard devaton. The obtaned expresson s

3 [ ~R vv (; )] k;l p 2 ()[R vv (; )] k;l = p 2 () exp(? 1 2 ~2 (l? k) 2 ) exp(j(l? k) 2 sn()) (1) where ~ = 2 cos() and an deal lnear array wth 9 o dentcal element antenna patterns p(x) s assumed. In ths paper we assume, p(x) = cos(x), whch closely ts the element pattern of the prototype lnear array presented n [3]. In order to smplfy notaton we ntroduce the normalzed fast fadng vector v(; ) dened as v(; ) = cos?1 ()~v(; ). The covarance matrx of v(; ) s mplctly gven by (1). Under the assumptons above the baseband sgnal u RX (t) receved at the moble s gven by u RX (t) = G 1 2 x TX;T (t)v(; ) (2) where x TX;T (t) s the transposed vector of base band sgnals transmtted from the ndvdual antennas at the base and G s gven by ( ( R r G = ) L cos 2 () f jj 9 o ; (3) otherwse: The varable G s the power averaged over the fast fadng. We wll refer to G as the path gan or the strenght of the moble. In the downlnk systems proposed below, t s assumed that G, and, have been estmated from the uplnk sgnals. Ths may be done as proposed n [8]. Fnally we ntroduce a dstance dependence n as = R=(2r). III. A Transmsson Algorthm In ths secton, an algorthm for transmsson from the base (to moble) s descrbed. The algorthm wll be employed by all the systems proposed n ths paper. Concder the stuaton where q base statons are transmttng smultanously (at the same frequency) to q mobles. The propagaton parameters of the :th moble seen from the k:th base are denoted G k; ; k; and k;. The :th moble s connected to the :th base. However, the framework allows that several mobles are allocated to the same base, and thus the :th and the k:th base may physcally be the same base. Let us wthout loss of generalty concder transmsson of the sgnal s 1 (t) from base 1 to moble 1. The stuaton s depcted n Fgure 1. The antenna element ntputs x TX 1 (t) are chosen as a weghtng vector w 1 tmes the (scalar) waveform s 1 (t).e x TX 1 (t) = w 1 s 1 (t): (4) Seen from the vewpont of the q? 1 co-channel mobles the sgnal transmtted from base 1 s nterference. Therfore these mobles are refered to as the nterfered mobles. We now chose the crteron functon w 1 = arg y mnf =2 I =G ; g (5) where I s the nterference (averaged over fast fadng) at the :th nterfered moble. A smlar crteron functon has prevously been proposed n [2]. Usng (2) and (4) the followng expresson s obtaned for I Usng (6) n (5) yelds, where I = G 1; w 1R vv ( 1; ; 1; )w 1 : (6) w 1 = arg y mnfy M 2 yg; (7) M 2 = =2 G 1; G ; R vv ( 1; ; 1; ): (8) Ths speces w 1 wthn a scalar factor. Ths factor s chosen such that the power delvered at the desred moble s G 1;1.e where G 1;1 w 1 M 1w 1 = G 1;1 (9) M 1 = R vv ( 1;1 ; 1;1 ): (1) Gven all the requred enttes, the weghtng vector w 1 can be obtaned as the soluton of a generalzed egenvalue problem. In general, the necessary parameters G 1; =G ;, 1; and 1; are only known for a subset of the nterfered mobles, say = 2; : : : ; p + 1. We refer to these mobles as the dented nterfered mobles. In order to account for the remanng mobles the approxmaton =p+2 G 1; G ; R vv ( 1; ; 1; ) constant I (11) s made. Ths approxmaton s reasonable f the angles 1; are well spread. The constant n (11) s obtaned from loose reasonng and wll vary dependng on applcaton. Wth the above dentons and assumptons the weghtng vector w 1 s gven by w 1 = 1 e; (12) (e M 1 e) 1=2

4 where e s the domnatng egenvector to the generalzed egenvalue problem wth M 1 e = M 2 e (13) M 1 = R vv ( 1;1 ; 1;1 ) (14) M 2 = Base 1 p+1 X =2 G 1;1 ; 1;1 ; 1;1 G 1; G ; R vv ( 1; ; 1; ) + constant I: G 1; ; 1; ; 1; G ; User 1 The th user Fg. 1. Illustraton of the concdered stuaton. (15) The th base IV. Two Capacty Enhancement Approaches In ths secton we propose a number of systems whch all use the transmsson algorthm of the prevous secton. The systems der n terms of channel allocaton, power control, synchronzaton and dentcaton of nterfered users. In terms of capacty enhancement strategy the systems can be dvded nto two groups: same cell frequency reuse (SCFR) and reduced cluster sze (RCS). The two groups of systems are descrbed n the followng sectons. A. Same Cell Frequency Reuse (SCFR) In ths approach the capacty enhancement s acheved by allocatng of d mobles wthn a cell, on the same channel. The frequency reuse plan s the same as n a correspondng one antenna (per sector) base staton system. Thus, the base statons transmts to d mobles smultaneously on the same channel. To make ths feasble, dynamc channel allocaton s appled to separate the co-channel users n the cell (wth respect to angle). In prncple, the deployment of SCFR n the downlnk does not mply that the same capacty enhancement strategy has to be used n the uplnk. However f t s used, channel allocaton and power control must be employed to combat near far effects n the uplnk processng. Snce some cellular systems dctate a one to one relaton between uplnk and downlnk channels, e.g GSM, the development followng assumes that the mobles smultaneously accessng the same channel n the downlnk also does that n the uplnk. A.1 Channel Allocaton and Power Control The mobles n the cell are sorted wth respect to ther path gan G ;, to ther desred base. Then, they are dvded nto groups say? 1 ;? 2 ; : : : such that all the mobles n? 1 are stronger than all the mobles n? 2 and so on. Subsequently, power control s then appled such that all mobles n a group are receved wth power equal to ther mean path gan (averaged n db). The groups are allocated n tme-frequency space such that ther dssmlar receved power does not negatvely eect the uplnk processng. The dynamc channel allocaton algorthm of [1] s appled on each of the groups. Ths algorthm begns by sortng the mobles n the group wth respect to ther azmuth angular poston ;. Assumng that the sorted lst of angular postons s 1 ; : : : ; ncd, the mobles wth angles ; +nc ; : : : ; +(d?1)nc wll be allocated to the same channel. If frequency hoppng s appled, the same channel users jumps together from frequency to frequency. A.2 User Trackng and Nullng The users on the same channel wthn the cell are consdered as the dented nterfered users n the framework of Secton III. Snce the consdered base and the dented nterfered users desred base are the same base n ths case, t follows that G 1; = G ; and thus the factors G 1; =G ; n (15) dsappear. The constant n (15) s chosen from the followng crude reasonng: Assume that only mobles n the rst ter of co-channel cells are dsturbed. Assume that the path gans to these mobles are gven by G 1; = (R=D) where D s the dstance between the consdered base and the rst ter of co-channel base statons. Assume further that these mobles have receved desred power G ; = 1. Geometry yelds that the number of sgncant undent- ed nterfered mobles s gven by q? p? 1 = d l where l = 6 and l = 3 usng 1/1 and 1/3 reuse plannng respectvely. Wth these assumptons we obtan =p+1 SIR?1 = (1=D) p+ld X =p+1 Thus the constant n (15) s chosen as B. Reduced Cluster Sze (RCS) R vv ( ; ) (R=D) ldi: (16) constant = (R=D) ld: (17) In ths approach the capacty enhancement s obtaned by changng the frequency allocaton plan.e the cluster sze. For nstance, by gong from a 1/3 to a 1/1 reuse pattern, a threefold capacty enhancement s obtaned. The antenna array base statons are employed to compensate for the ncreased nter-cell nterference. Several versons of ths approach are nvestgated. The versons appear as derent derent power control settngs and drected nulls or not. The development below assumes that uplnk co-channel mobles are downlnk co-channel mobles,.e mobles that are nterferers n the uplnk are nterfered

5 n the downlnk. For TDMA systems ths mples that the tmeslots of the base statons are synchronzed. The treatment n Sectons IV-B.1 to IV-B.3 assumes for smplcty that frequency hoppng s not appled. The mpact of frequency hoppng s gven treated n Secton IV-B.4. The development makes the reasonable assumpton that only nterferng mobles that are receved sgncantly strong may be dented and tracked by the base staton. B.1 Power Control Consder the stuaton depcted n the Fgure 1 agan. Assume for a moment that no other base or mobles than the two present n the plot exst. If the antenna array gan s dsregarded, the uplnk sgnal to nterference rato at base 1 s gven by SIR up 1 = G 1;1 P up 1 =(G 1;P up ); (18) where P up 1 and P up are the power transmtted by user 1 and user respectvely. Smlarly, the downlnk sgnal to nterference rato at the :th nterfered moble s gven by Heren, the power control law SIR down = G ; =G 1; : (19) P up = G?e r ; : (2) s used, where e r s a desgn parameter. If e r = 1 then SIR up 1 = SIR down. Thus a moble that s receved strongly at base 1, wll be vulnerable to transmsson from base 1. Ths s a desrable property snce t mples that base wll have a good chance of dentfyng the moble wth poor downlnk SIR and drect a null n the transmt pattern towards that moble. Another advantage wth e r = 1 s that all mobles wthn the cell are receved equally strong at the base and therefore, the adjacent channel nterference s small. In practce usng e r = 1 may be prohbted by the power control range allowed by the moble. In that case a smaller e r s employed. B.2 Dynamc Channel Allocaton When e r 6= 1 the power of the mobles may vary sgnfcantly whch causes adjacent channel nterference problems. We propose that ths problem s solved usng dynamc channel allocaton as follows: Sort the mobles n the subcell wth respect to ther path gan to the base G ;. Dvde the mobles nto groups say? 1 ;? 2 ; : : : such that all the mobles n? 1 are stronger than all the mobles n? 2 and so on. Fnally, allocate the groups n tme-frequency space such that the dssmlar power of the groups does not negatvely eect the uplnk processng. Wth e r = 1 random channel allocaton s smply employed. B.3 User Trackng and Nullng We wll consder two versons wth respect to nullng: wth and wthout drected nulls. Wth Drected Nulls The requred enttes of dented nterferng users n (15) are G 1; =G ; ; 1; and 1;. When power control wth e r = 1 s used G 1; =G ; s the power receved n the uplnk from the :th user and thus all the parameters can be derved from the uplnk data. When e r 6= 1 however, the :th base must nform base 1 about the value of G ;. Ths n turn requres that the base 1 s able to dentfy the home cell of the nterferng user. In order to chose the constant n (15) we assume that the base s able to dentfy nterferng users stronger than P mn, and that the value of P mn s known by the base. Snce the power of the desred user ders among the channels P mn wll be+ channel dependent. Thus seen from the vewpont of base 1, the nterferng users that satsfy G 1; G?e r ; P mn (21) are not dented. Equaton (21) n (11) yelds =p+2 SIR?1 = =p+2 =p+2 G 1; G?1 ; R vv ( 1; ; 1; ) P mn G 1?e r ; R vv ( ; ) (:1) 1?e r P mn(r? p)i (22) = const I (23) where the approxmaton s based on the assumpton G ; = :1 whch s approxmately the medan value of G ;. Based on geometry the parameter r s set to 11 n the 1/1 reuse case and 5 wth 1/3 reuse. Wthout Drected Nulls In ths case nterferng mobles are not dented n the framework of Secton III,.e p =. It follows that the constant n (15) can be chosen arbtrarly. Wth ths approach synchronzed bases s not a requrement. B.4 Inuence of Frequency Hoppng Let us consder the GSM system [5]. In ths system a user uses the same tme slot n each TDMA frame but changes frequency accordng to a pseudo random pattern. The tme and frequency of the uplnk tmeslot s gven by the downlnk tmeslot wth a delay of two burst perods and a frequency translaton of 45MHz respectvely. The frequency hoppng mples that a user \meets" derent nterferers n each TDMA frame. Let us for a moment regard all frequency channels multplexed on the same tme slot n the TDMA frame as a sngle wdeband channel. Wth ths denton of the channel we may say that uplnk co-channel mobles are downlnk co-channel mobles. The dea s to track the users (.e estmate G 1; =G e r, ; 1; and

6 1; n the framework of secton III) as they jump from frequency to frequency. Ths s possble f the base s able to determne the home cell of the nterferng users whch n turn wll enable the nfrastructure to nform the base of the frequency hoppng pattern of the nterferng moble. In the downlnk the base transmts wth nulls n the drecton of the dented nterfered users whch are usng the same frequency n the partcular tmeslot. Note that the mnmum power enablng dentcaton of nterferng users P mn wll depend on the power of all the users that are usng the channel (where channel s n the sense dened above). V. Results A 1/3 Same Cell Frequency Reuse (SCFR) system wth, d = 3, users per channel s compared wth a 1/1 Reduced Cluster Sze (RCS) system. The RCS system s operated wth and wthout ntercell nulls and usng the two choces e r = 1 and e r = :3. The ar nterface s GSM wth frequency hoppng and dscontnuous transmsson [5]. The user actvty factor s :5. The smulaton consders 48 users n each subcell usng sx 2kHz carrers wth eght tmeslots multplexed on each. Ths s approxmately seven tmes less bandwdth than than the correspondng one antenna per sector system would requre for the 48 users, [1]. The base antenna arrays have ten elements m = 1 wth half a wavelength spacng = :5. The propagaton exponent used s = 3:5 and the standard devaton of the log-normal fadng s db = 8. However, the log-normal fadng between a moble and two base statons s correlated wth correlaton coecent :5.e Ef1 log(l 1;)1 log(l 2;)g = db 2 =2. No mobles are closer than r ; = :35R(cos( ; )= cos(3 o )) 2= to the base. The regon nsde ths nner lmt s 1% of the cell area. The handover s geometry and antenna pattern based. Ths means that a moble s connected to t's geometrcally closest base wth a bas accordng to the antenna element patterns. In practce ths handover strategy corresponds to a sgnal strength based handover wth very large tme averagng wndow. The SCFR system uses four power groups wth twelve mobles each. Group 1 uses the rst and second tmeslot, group 2 the thrd and fourth and so on. The RCS systems wth nulls uses a P mn (see secton IV-B.3) whch s set to 3dB less than the medan power of the sx desred users n a TDMA tme slot. Wth e r = 1 random channel allocaton s employed whereas wth e r = :3 the power control of secton IV-B.1 s employed wth eght groups consstng sx mobles each. Group number one s allocated to rst tme slot n the TDMA frame, group number two to the second and so on. More detals of the smulaton procedure s gven n [9]. In Fgure 2, hstograms of the (relatve) power control settngs at the mobles s plotted. The data s obtaned from one smulaton and consders only user at one base. In Fgure 3, a hstogram of the rato of the strongest sgnal, to the weakest desred sgnal n a TDMA tmeslot (6 2kHz bandwdth). The data s obtaned from one Number of mobles Number of mobles Number of mobles Power transmtted from the moble (db) Power transmtted from the moble (db) Power transmtted from the moble (db) Fg. 2. Power control settngs at the 48 users n subcell 1a; upper: SCFR ; mddle: RCS,er = 1; lower: RCS,er = :3 Number of tmeslots Number of tmeslots Number of tmeslots Strongest sgnal to weakest desred sgnal (db) Strongest sgnal to weakest desred sgnal (db) Strongest sgnal to weakest desred sgnal (db) Fg. 3. Dynamc range requrements n the 6 2kHz spectrum; upper: SCFR ; mddle: RCS,er = 1; lower: RCS,er = :3 smulaton and consders the sgnals receved at one base staton. In Fgure 4 the estmated outage probablty, dened as the probablty that a user experences bad speech qualty, s plotted as a functon of the multpath angular spread, see secton II. The smulaton results n Fgure 4 are obtaned from 7 smulatons accordng to the procedure n [9]. In [9] approxmatve analytcal expressons of the outage probablty are derved for the SCFR approach and the RCS approach wth nullng and e r = 1. Results usng these expressons are also plotted n Fgure 4. From Fgure 2 we conclude that the RCS system wth e r = :3, s consstent wth the allowed 2dB power control range of GSM, whle SCFR system requres a lttle more and RCS wth e r = 1 system requres much more. Fgure 3 ndcates that the uplnk processng must be able to suppress nterference that are about 15dB stronger that than the desred sgnal n all three cases. To acheve ths the base may explot frequency selectve lterng and antenna array gan. It appears as f a dgtalzaton of the entre bandwdth s not feasable. The results n Fgure 4 shows that the RCS system wth

7 Outage Probablty Angular Spread at r=.5r (degrees) c1 c2 c3 c4 c5 c6 c7 Fg. 4. Outage probabltes, c1: RCS, 1/1 reuse, er = :3, no nullng, smulaton. c2: RCS, 1/1 reuse, er = 1, no nullng smulaton. c3: SCFR, 1/3 reuse, Three mobles per channel, d = 3, smulaton. c4: RCS, 1/1 reuse, er = :3, wth nullng, smulaton. c5: SCFR, 1/3 reuse, Three mobles per channel, d = 3, analytcal. c6: RCS, 1/1 reuse, er = 1, wth nullng, analytcal. c7: RCS, 1/1 reuse, er = 1, wth nullng, smulaton. nullng has the lowest outage probablty f e r = 1 s appled. However ths superorty s lost f e r = :3 s used. The derence between the smulaton results and the correspondng analytcal results are approxmately one unt of standard devaton, except for the SCFR and = 1 o where the analytcal result s optmstc. We beleve that that ths s because the analytcal result assumes that the same cell users are well separated n angle whch becomes crtcal wth large angular spreads. However, we nd that the analytcal expressons are good vehcles for assessng the performance of ther correspondng approaches. Fgure 4 also supports the concluson n [1], that SCFR has better performance than RCS wthout drected nulls. The standard devaton of the errors n the smulated outages n Fgure 4 s about 1:5%. However, the smulatons of the derent approaches uses the same realzaton of the user postons and path gans and thus the results should gve a good ndcaton of the performance derences between the approaches. In [9] the systems are also smulated usng a more ecent handover. The results shows that all system mproves sgncantly. Ths s partcularly true for the RCS system. References [1] C Carnehem, S.O Jonsson, M Ljungberg, M Madfors, and J Naslund. \FH-GSM Frequency Hoppng GSM". In Proceedngs IEEE Vehcular Technology Conference, pp. 1155{1159, [2] D Gerlach and A Paulraj. \Base Staton Transmttng Antenna Arrays for Multpath Envronments". submtted to IEEE Transactons on Sgnal Processng, [3] B Johannson. \Planar Antenna Array for Adaptve Beamformng". In Nordc Rado Symposum 1995, pp. 177{18, Saltsjobaden Sweden, Aprl [4] W.C.Y Lee. \Moble Communcatons Desgn Fundamentals". Wley, New York, [5] M Mouly and M.B Pautet. \The GSM System for Moble Communcatons". Mchel Mouly and Mare-Bernadette Pautet, 49,rue Louse Bruneau, F-9112 Palaseau France, ISBN [6] T Ohgane. \Spectral Ecency Evaluaton of Adaptve Base Staton for Land Moble Cellular Systems". In Proceedngs IEEE Vehcular Technology Conference, pp. 147{1474, [7] G Ralegh, S.N. Dggav, V..K Jones, and A Paulraj. \A Blnd Adaptve Transmt Antenna Algorthm for Wreless Communcaton". In Proceedngs IEEE Internatonal Conference on Communcatons, [8] T Trump and B Ottersten. \Maxmum Lkelhood Estmaton of Nomna drecton of Arrval and Angular Spread Usng an Array of Sensors". In Proceedngs of COST 229 Adaptve Systems, Intellgent Approaches, Massvely Parallel Computng and Emergent Technques n Sgnal Processng and Communcatons, Vgo, Span, January [9] P Zetterberg. \A Comparson of two Systems for Down Lnk Communcaton wth Antenna Arrays at the Base, Report Verson". Techncal Report IR-S3- SB-9521, Sgnal Processng, Royal Insttute of Technology, Sweden, Avalable by Mosac: Document URL: or by anonymous ftp to: elxr.e.kth.se drectory /pub/sgnal/reports., [1] P Zetterberg and B Ottersten. \The Spectrum Ecency of a Basestaton Antenna Array System for Spatally Selectve Transmsson,". IEEE Transactons on Vehcular Technology, vol. 44, no. 3,, August Bography VI. Conclusons In ths paper we have compared Same Cell Frequency Reuse (SCFR) aganst Reduced Cluster Szes (RCS), for ncreased the spectrum ecency usng base staton antenna arrays. The results shows that RCS can outperform SCFR provded that nullng and ecent power control s employed. Unfortunately, the used power control algorthm s not consstent wth current GSM speccatons. The assumed propagaton model assumes a cluster of rays n the vcnty of the moble. A derent propagaton model may very well lead to other conclusons. Per Zetterberg (S'92) was born n Uppsala, Sweden, He receved the M.S. degree n electrcal engneerng from Lulea Unversty, Lulea, Sweden, n Snce then he has been wth the Sgnal Processng group at the Royal Insttute of Technology (KTH), Stockholm, Sweden, as a graduate student. Zetterberg s currently a guestresearcher at the Center of Personal Communcatons n Aalborg, Denmark. Zetterbergs research nterests are n the areas of wreless communcatons and sensor array sgnal processng.