Link-Level Analysis of Downlink Handover Regions in UMTS

Transcription

1 Link-Level Analysis of Downlink Handover Regions in UMTS Atta Ul Quddus, Xinjie Yang and Rahim Tafazolli, Member, IEEE Abstrat This paper investigates the downlink handover (soft/softer/hard) performane of Wideband Code Division Multiple Aess (WCDMA) based 3 rd generation Universal Mobile Teleommuniation System (UMTS), as it is known that the downlink apaity of UMTS is very sensitive to the extent of overlap area between adjaent ells and power margin between them. Fators influening the handover performane suh as the orrelation between the multipath radio hannels of the two links, limiting number of Rake fingers in a handset, imperfet hannel estimation, et. that annot be modeled adequately in system-level simulations are investigated via link-level simulations. It is also shown that the geometry fator has an influene on the handover performane and exhibits a threshold value (whih depends on the orrelation between the multipath hannels assoiated with the two links in a handover) above whih the performane starts degrading. The variation of the handover gain with the losed loop power ontrol (CLPC) stepsize and spae-time transmit diversity (STTD) is also quantified. These omprehensive results an be used as guidelines for more aurate overage and apaity planning of UMTS networks. Index Terms UMTS, Code division multi-aess, Handover U I. INTRODUCTION NIVERSAL Mobile Teleommuniation System (UMTS) is urrently being deployed and optimised as new servies and features are being inorporated. It is neessary to be equipped with aurate air-interfae (link-level) and systemlevel simulators to help in the optimisation proess. The linklevel simulator provides inputs to the system-level simulator, based on whih various performane metris suh as throughput, delay, et. an be optimised. Some of the tehniques that are usually modelled at the system-level suh as soft handover () an also be modelled at the link-level. is one of the most important funtionalities of radio resoure management in CDMA based mobile ellular systems. It is used in the intra-frequeny handover and allows mobile stations to be onneted simultaneously to several base stations, whih are seleted by a pre-defined algorithm. provides spatial diversity in both the uplink and the downlink, leading to improvements in quality of servie. Proper design and tuning of is one of the main hallenges in UMTS, as it has great impat on the trade-offs between overage, system apaity and servie quality in the network. The impat of is usually investigated through systemlevel simulations. However, at the system-level, multipath fading is not usually onsidered, as the pilot hannel samples taken for handover purpose are measured on a fairly slow basis and are averaged so as to redue the number of handovers. Moreover, the use of maximum ratio ombining (MRC) in the downlink for users implies that the Rake reeiver has an unlimited number of fingers and hene all the multipaths from the base stations involved, an fully ontribute to the signal diversity. This is inaurate, as a pratial Rake handset reeiver has limited number of fingers [1]. Link-level researh on diversity performane, to some extent ould retify these inauraies. Realisti multipath hannels orresponding to more than one Node-B that are in a handover along with power margin, orretly defined other users interferene and detailed transeiver models an overome the inauraies of performane obtained at system-level. Yet there annot be found in the literature a omprehensive set of link-level results regarding, softer handover (SerHO) and hard handover () performane, partiularly as a funtion of the multipath fading orrelation between the links. This paper attempts to retify the omission. Within published literature, the authors in [2] ompare the link-level handover gain as a funtion of the relative path loss between the two Node-Bs, but they model the interferene as Gaussian noise, whih is a reasonable assumption only for the other-ell interferene (aused by distant base stations) and not for the downlink intra-ell interferene as well as the interferene aused by the adjaent handover ell. In [3], linklevel handover results are presented for four different servies in 3GPP Case 3 radio hannel but they onsider only unorrelated links with a fixed power margin. In this paper, we evaluate the performane of and ompare it with that of and no handover (that models a single ell) for a range of geometry fators (defined as ratio of power reeived from the serving ell to that from surrounding ells plus the thermal noise) and establish the range of important design parameters that influene the system apaity. The paper is organized as follows: the system model is introdued in setion II and the handover performane in setion III for a range of geometry fators for a maro as well as a miro ell. Setion IV illustrates the impat of various other fators that influene the downlink handover performane. Finally, onlusions are drawn in setion V /08/$ IEEE 2527 Authorized liensed use limited to: University of Surrey. Downloaded on April 19,2010 at 15:08:46 UTC from IEEE Xplore. Restritions apply.

2 II. SYSTEM MODEL Fig. 1 shows the modelling of the handover proess at the link-level. Two Node-Bs are modelled (having independent dediated physial data hannel DPDCH, ommon pilot hannel CPICH and orthogonal hannel noise simulator OCNS that is a mehanism for modelling intra-ell users [4]) as well as the power margin (differene) between the two radio links, an important parameter that triggers the initiation of handover proedure. A desired orrelation is introdued between the multipath radio hannels from the two Node-Bs. In this paper, we distinguish between and SerHO based on the orrelation between the two links (higher orrelation for SerHO and lower for ). In the Rake reeiver, fingers are alloated for desrambling, despreading and hannel ompensation of the signal from both the Node-Bs or only one of them, depending upon whether the UE is in or not. CPICH1 DPCH1 OCNS1 Radio hannel1 Srmb Code1 Srmb Code TABLE 1: SIMULATION PARAMETERS PARAMETER ASSUMPTION Chip Rate 3.84 Mps 3GPP Referene Channel [4] 12.2 kbps Speeh Radio Channels Used 3GPP Case 1, 3GPP Case 3 Correlation b/w the 2 Node-B s Variable between 0.00 to 0.95 Power Diff. b/w the 2 Node-B s 0 db or 3 db Number of Rake Fingers Variable between 4 and 8 Channel Estimation Ideal or through CPICH Minimum allowed DPCH E 45 db Maximum allowed DPCH E 3 db Inner Loop Power Control Step Size Variable b/w 0.5 db and 2 db Outer Loop Power Control Step Size 0.5 db Geometry Fator ( I or / Io Variable b/w 6 db and 6 db Downlink Physial Channels & Power Levels CPICH_E/Ior = -10 db DPDCH_E/Ior = variable OCNS_E/Ior = power needed to make Ior = 1 UniS Nokia UniS Nokia Channel orrelation Rake 10 1 CPICH2 DPCH2 OCNS2 Radio hannel2 x Power margin I o = σ 2 o σ 2 BG BLER 10 2 Fig. 1. Conise blok diagram of the links from two Node-Bs to one UE Typially, the multipath fading assoiated with the two radio links that are in a handover mode is assumed to be unorrelated whih is not always true partiularly for SerHO due to the o-loation of two setors of the Node-B. Due to the geometry of the environment, there is a degree of orrelation even for the multipath fading between the two radio links. Thus it is important to model the multipath fading orrelation as aurately as possible in order to have a true piture of the gains provided by handover. Herein, a method for generating a desired number N of Rayleigh fading envelopes with any desired ovariane matrix, as presented in [5], is followed. The simulation parameters are shown in Table 1 whose values are typial assumptions used for 3GPP simulations [3,4]. It should be noted that the 3GPP Case 1 is a 3 km/h twotap hannel, representative of a typial urban miro-ell whereas the 3GPP Case 3 is a 120 km/h four-tap hannel and represents the propagation harateristis of a fast moving vehile in a maro-ell [4]. An example of simulator verifiation is shown in Fig. 2 where our results (marked as UniS) are ompared to the results obtained by Nokia [3] for and and it an be seen that there is a very good math giving redibility to the results presented in the next setion Transmit E/Ior Fig. 2. Comparison with Nokia results [3] in 3GPP Case 3 Channel III. IMPACT OF GEOMETRY FACTOR ON HANDOVER In Fig. 3, the performane of, and is shown as a funtion of the geometry fator in 3GPP Case 1 radio hannel and in Fig. 4 in the 3GPP Case 3 radio hannel. E/Ior Geometry Fator Fig. 3. Performane in Case 1 Radio Channel with Correlation 0 For both sets of results, the power margin between the two 2528 Authorized liensed use limited to: University of Surrey. Downloaded on April 19,2010 at 15:08:46 UTC from IEEE Xplore. Restritions apply.

3 Node-Bs was 0 db, i.e. ( I ˆ / I Iˆ / I ) or1 o = or2 o and the orrelation between the two links was 0 as well. For the Case 1 radio hannel, CLPC was swithed-on whereas for the Case 3 radio hannel, it was swithed-off as the Case 3 radio hannel orresponds to a mobile veloity of 120 km/h and CLPC does not work well at high speeds. db in the Case 1 radio hannel and to 0 db in the Case 3 radio hannel, as expeted. The threshold value of 0 db of the geometry fator an be used as a guideline for highly orrelated environments E/Ior E/Ior Geometry Fator Geometry Fator Fig. 4. Performane in Case 3 Radio Channel with Correlation 0 It is interesting to note that in Fig. 3, the performane of beomes worse when ompared to for geometry fators greater than or equal to 6 db. A similar trend is seen in Fig. 4 where at a 6 db geometry fator, and perform almost idential. The reason for this behaviour is that experienes an interferene floor as both the Node-Bs transmit at full power and the srambling odes are not perfetly orthogonal whereas, whih is essentially a single isolated ell, suffers only from the other ells I and multipath indued interferene but not from the interferene experiened with the handover. As the geometry fator inreases (the other ell interferene dereases as in I ˆ or / Io the Î is fixed to unity), the rate of derease in or of is muh lower ompared to that of. This an be observed in Fig. 3 and Fig. 4 and is due to the fat that the suffers from a multi-aess interferene floor as pointed out earlier and eventually the E / I of interferene ( ) o beomes worse than that of. The 6 db threshold value at whih this happens and auses to suffer a loss in apaity will help in the ell planning as the geometry fator provides a measure of ell isolation. Sine, the multipath fading orrelation redues the signal ombining gain, it is expeted that with inreasing orrelation, the E will beome worse than that of at lower values of geometry fator (i.e. higher levels of the otherells interferene). This is illustrated in Fig. 5 for the Case 1 radio hannel and in Fig. 6 for the Case 3 radio hannel at 1 % BLER and for a very high orrelation value of In reality, the multipath fading orrelation may not be as high as 0.95, nevertheless it represents the worst-ase senario. Fig. 5 and Fig. 6 show that the value of the threshold indeed redues to 2 or Fig. 5. Performane in Case 1 Radio Channel with Correlation 0.95 E/Ior Geometry Fator Fig. 6. Performane in Case 3 Radio Channel with Correlation 0.95 IV. FACTORS INFLUENCING HANDOVER PERFORMANCE In this setion, the influene on the handover performane by fators suh as the power margin between the two Node-Bs (or two setors of a Node-B), multipath fading orrelation, inner loop power ontrol step size (ILPC), limited number of Rake fingers, hannel estimation and STTD are investigated. Note that the in the following results is defined as the differene of E values of and at a BLER of 1 %. All of the following results were obtained for a geometry fator of 0 db, a BLER of 1 % and using ideal hannel estimation unless otherwise speified. A. Impat of Correlation and Power Margin Table 2 summarizes the E values for the and at 1% BLER as a funtion of the multipath fading orrelation between the two radio links in 3GPP Case 1 radio hannel with the power differene between the two Node-Bs equal to 0 db and 3 db, respetively. For these results, the CLPC was also swithed-on. Sine the handover is initiated only when the power margin (or power differene) between 2529 Authorized liensed use limited to: University of Surrey. Downloaded on April 19,2010 at 15:08:46 UTC from IEEE Xplore. Restritions apply.

4 the two Node-Bs is less than or equal to 3 db, the results of Table 2 represent both the best-ase results (0 db power margin) and worst-ase results (3 db power margin). TABLE 2: CORRELATION AND POWER MARGIN IMPACT IN CASE 1 CHANNEL Correlation It is seen from Table 2 that a signifiant gain of 5.16 db is obtained at a orrelation of 0 when the power margin is 0 db. By inreasing the orrelation, the gain steadily redues and at a orrelation of 0.95, the gain has redued by 1.94 db, i.e. ( = 1.94). Moreover sine the SerHO usually experienes high values of orrelation due to the o-loation of the two setors of a Node-B whereas typially experienes low or almost no orrelation due to the physial separation of the two Node-Bs, SerHO is expeted to support 28% (.93 ( (.49) )/10 = 1 10 ) less users than. Table 2 also shows that when the power margin between the two Node-Bs is 3 db, performs about 1 db = ( ) worse as ompared to 0 db power margin, but the performs 1.53 db = ( ) better than the performane with 0 db power margin. This is logial sine in, the signal from the seond Node-B is not ombined and ats as an interferer, and the lower the interferene power, the better the performane. Beause of this, the gain over the also redues signifiantly with the inreasing power margin, e.g. at orrelation of 0 the gain redues from 5.16 db to 2.65 db. Table 3 shows the impat of the orrelation and the power margin on the and the performane in the 3GPP Case 3 radio hannel with CLPC disabled. TABLE 3: CORRELATION AND POWER MARGIN IMPACT IN CASE 3 CHANNEL E / I or Correlation A loser observation of Table 2 and Table 3 reveals that the performane of both and is improved in the 3GPP Case 3 hannel ompared to the 3GPP Case1 hannel due to availability of more multipath diversity and a greater degree of time variations in the former. However, the improvements to are less as ompared to, as ombining has already taken advantages of hannel diversity, leading to lower gains in the 3GPP Case3 hannel. Also in Case 3, the relative performane loss of SerHO as ompared to the is similar to that in Case 1 radio hannel, i.e. 28% (.20 ( (.75) )/10 = 1 10 ), at a power margin of 0 db. Thus the multipath fading orrelation affets the relative performane similarly in the two fading hannels. B. Impat of ILPC Step Size Table 4 shows the impat of the inner loop power ontrol (ILPC) step size on the handover performane for a power margin of 0 db and 3 db, respetively. The orrelation is assumed to be 0 for these results. TABLE 4: IMPACT OF ILPC STEP SIZE IN 3GPP CASE 1 RADIO CHANNEL Step Size It is seen that E does not vary muh as the stepsize hanges from 0.5 db to 2 db, whereas E improves. This is due to the fat that for a larger power ontrol step-size also helps overome the interferene aused by the signal from the seond Node-B that ats as an interferer and is not used for signal ombining. On the other hand,, due to signal ombining from the two Node-Bs is seen to be relatively independent of ILPC step-size in Case 1 hannel. TABLE 5: IMPACT OF ILPC STEP SIZE IN 3GPP CASE 3 RADIO CHANNEL Step Size Table 5 shows the performane in the 3GPP Case 3 radio hannel and it an be seen that the E shows an optimum step-size of 1 db for a power margin of 0 db as well as 3 db. The performane degrades at higher step sizes (e.g. 2 db) due to the faster variations of the 120-km/h radio hannel. Again, the is seen to perform relatively better in the Case 3 than in the Case 1 radio hannel. It is also noted from 2530 Authorized liensed use limited to: University of Surrey. Downloaded on April 19,2010 at 15:08:46 UTC from IEEE Xplore. Restritions apply.

5 Table 5 and Table 3 that in the Case 3 radio hannel, by enabling the CLPC, the performs worse ( E = db) at a step-size of 1 db as ompared to the with CLPC disabled ( E = db). Thus at high speeds, the CLPC should not be used, at least from the perspetive of. C. Impat of Channel Estimation Sine, hannel estimation (CHEST) is more hallenging at high speeds, we present the impat of CHEST through the CPICH only in the 3GPP Case 3 radio hannel having 120 km/h mobile veloity. Table 6 illustrates the impat of CHEST on the E of, and at 1% BLER, 0 orrelation and the power margin equal to 0 db. It is observed that estimating the hannel through CPICH inurs a loss in of about db. Note that the hannel estimation method that has been used in the simulations is a simple pilot averaging method. With more sophistiated 0.5/10 methods, the 11 % = ( 1 10 ) loss in apaity of an be redued still further. TABLE 6: IMPACT OF CHEST IN 3GPP CASE 3 CHANNEL WITH CLPC OFF Ideal CHEST Through CPICH Loss D. Impat of the Number of Rake Fingers A typial Rake reeiver in a UE has four to six Rake fingers. Sine, for in the 3GPP Case3 hannel, the total number of multipaths is 8, the impat of limited number of Rake fingers is shown in Table 7 by assuming a orrelation of 0 between the two links and CLPC swithed off. TABLE 7: IMPACT OF NO. OF RAKE FINGERS IN CASE 3 CHANNEL No. of Fingers E E Note that for these results, the fingers are alloated to the paths having the highest power in desending order. Due to this finger alloation mehanism, with 4 Rake fingers, the E / of suffers at most by 0.3 db = ( ), I or 0.3/10 i.e. a loss of 7 % ( 1 10 ) = in apaity, in the 3GPP Case 3 hannel, when the power margin is 0 db. E. Impat of STTD The impat of STTD on the performane is illustrated in Table 8 for 3GPP Case 1 radio hannel with CLPC enabled and a orrelation value of 0. Note that in Table 8, the STTD gain is the gain provided by STTD in and not the gain provided by over the performane. It is seen that the relative gain provided by STTD in is around 0.5 db. In [6], it is reported that STTD provides a gain of 0.8 db in a typial miro-ell without taking into aount handover, thus the 0.3 db redution ( ) in the STTD gain in seems logial as the already provides a degree of diversity and the additional diversity gain provided by STTD is expeted to be somewhat less. TABLE 8: IMPACT OF STTD IN 3GPP CASE 1 CHANNEL WITH CLPC ON Power Margin E With STTD Without STTD STTD V. CONCLUSION In this paper, the performane of (soft/softer/hard) handover in WCDMA based UMTS FDD downlink is analysed at the link-level in 3GPP Case 1 and 3GPP Case 3 radio hannels. It is shown that there is a trade-off between the signal ombining gain provided by the handover, the multi-aess interferene assoiated with the handover and the other ells interferene. Threshold values of the geometry fator are established beyond whih the performane of beomes worse than that of a single isolated ell. Fators influening the handover performane suh as the multipath fading orrelation, power ontrol step-size, hannel estimation as well as transmit antenna diversity, are investigated and their impat on the handover is illustrated in terms of the average transmit power. These link-level results will help in the system-level analysis and provide guidelines for UMTS network dimensioning. REFERENCES [1] A. Chheda, A Performane Comparison of the CDMA IS-95B and IS- 95A Soft Handoff Algorithms, in Pro. IEEE VTC, Houston, TX, vol. 2, pp , 16 th 20 th May [2] O. Salonaho and J. Laakso, Flexible Power Alloation For Physial Control Channel in Wideband CDMA, in Pro. IEEE VTC, Houston, TX, pp , May [3] Nokia, Inter ell soft handover simulation results for UE performane tests, 3GPP TSG-RAN WG4 Meeting #9, TSGR4#9(99)805, Bath, UK, Deember 7 th -10 th, [4] 3GPP TSG RAN, User Equipment (UE) radio transmission and reeption (FDD), Release 6, TS , V8.0.0 ( ). [5] B. Natarajan, C. R. Nassar, and V. Chandrasekhar, Generation of Correlated Rayleigh Fading Envelopes for Spread Spetrum Appliations, IEEE Commun. Lett., vol. 4, no. 1, pp. 9-11, Jan [6] H. Holma and A. Toskala, WCDMA For UMTS Radio Aess For Third Generation Communiations, 3 rd Ed., Chihester: John Wiley & Sons Ltd, Authorized liensed use limited to: University of Surrey. Downloaded on April 19,2010 at 15:08:46 UTC from IEEE Xplore. Restritions apply.