1274 LETTER A Novel Adaptive Chael Estimatio Scheme for DS-CDMA Che HE a), Member ad Xiao-xiag LI, Nomember SUMMARY This paper proposes a adaptive chael estimatio scheme, which uses differet movig average legth ad pilot gai for differet mobile eviromets. It is based o MSE method ad extesive simulatios uder various eviromets for WCDMA physical layer. The scheme applies a computatioally efficiet ad easily implemeted pilot filter o WCDMA forward chael. For differet mobile chael eviromets, the optimal combiatio of movig average legth ad pilot gai for low SNR is achieved. The simulatio results illustrate that the adaptive scheme ca achieve much lower BER compared with two other adaptive schemes, especially whe the speed of mobile user is high. Ad the BER performace of the proposed scheme is isesible to the mobile speed. key words: DS-CDMA, pilot chael, movig average, pilot gai 1. Itroductio For limited wireless spectrum resources, digital spread spectrum techology ca achieve high spectrum efficiecy ad support more multi-access users due to frequecy reuse. Wide-bad code divisio multi-access based o direct sequece spread spectrum is oe of the key techologies i 3G systems. Coheret demodulatio, which requires less SNR, is more attractive tha o-coheret demodulatio due to the fact that low SNR required by sigle user ca icrease the system capacity. Accurate chael estimatio is required to implemet coheret demodulatio uder multi-path fadig eviromet. The chael is modeled as multi-path fadig model ad the fadig rate icreases as the carrier frequecy ad the speed of mobile statio icrease. The 3rd Geeratio System demads reliable high speed data commuicatio for more dyamic mobile speed zoes of users, compared with the 2d Geeratio System. Thus the chael estimatio scheme for WCDMA is required to adapt to dyamic mobile speed from pedestria (5 km/h) to high speed vehicle (above 300 km/h). Ad such scheme should provide reliable data commuicatio for low SNR situatio, especially whe mobile speed is high. For WCDMA, pilot chael is utilized to provide the mobile chael iformatio for coheret demodulatio. I mobile systems that use pilot chael durig demodulatio o fadig chaels, pilot filter is used to achieve the Mauscript received Jue 8, 2004. Mauscript revised August 23, 2004. The authors are with the Dept. of Electroic Eg., Shaghai Jiaotog Uiv., Shaghai 200030, P.R. Chia. This work is supported by the Natioal Natural Sciece Foudatio o Chia (o.60272082) ad the Shaghai Wireless Commuicatio Ceter project. a) E-mail: chehe@sjtu.edu.c DOI: 10.1093/ietcom/e88 b.3.1274 amplitude ad phase of fadig [1]. I DS-CDMA, geerally a bak of low-pass filters is used to estimate the chael respose o each rake figer before rake combiig [2]. Pilot filters i receivers are usually fix-bad. O forward chael, movig average ad first-order IIR filter have bee used for easy implemetatio [2] [5]. For piloted aided chael estimatio, two elemets are cosidered. First, the badwidth of pilot filter must be larger tha the maximum Doppler frequecy f D of fadig chael. If it is too arrow, the estimates ca ot follow the chage of chael. But if it is too broad, extra oise is itroduced at the output of pilot filter. For fixed f D, the optimal pilot filter ca be achieved accordig to Wieer Theory, but it is ot optimal whe f D is chagig. Secod, from the perspective of system capacity as little as pilot power should be trasmitted because pilot sigal is umodulated ad trasmits o iformatio. Dyamic pilot power adjustmet is discussed i [2] uder the assumptio that the chael is ot distorted by pilot filter. Yet this ideal situatio assumptio ca ot be applied to real mobile eviromet, especially whe the mobile speed is high. I [6] a aalytical discussio is carried out to achieve the optimal pilot to data power ratio. I this paper, we propose a adaptive chael estimatio method based o movig average method [2]. It chooses the optimal legth M of movig average filter ad pilot power gai A from a pre-calculated table for differet f D to process pilot sigals for imperfect chael estimatio. The optimal combiatio of movig average legth ad pilot gai is achieved by MSE method ad simulatios. This method ca compesate the performace loss caused by fixbad pilot filter, ad also it is easy to implemet ad computatioally efficiet. The simulatio results show that the proposed scheme ca achieve better BER performace tha the adaptive scheme proposed i [3]. 2. CDMA Chael Estimatio System Model I this paper, we cosider chael estimatio for DS- CDMA. It is assumed that QPSK modulatio is used. The for user k the symbol sequece o traffic chael after direct sequece spreadig is give by: Xd(t) = Ec(k) x(k)h(t T c) [a (I) (k)cos(2π f 0 t) + a (Q) (k)si(2π f 0 t)] (1) where E c (k) is the chip eergy of kth user, x (k)isbiaryof Copyright c 2005 The Istitute of Electroics, Iformatio ad Commuicatio Egieers
LETTER 1275 Fig. 1 Pilot-aided modulatio for sigle user. rate ad it works simultaeously with data despreadig at rake receiver. Geerally o each rake figer, chael estimate multiplies despreaded data for multi-path combiig durig each symbol period. I fact, it is ot ecessary to perform multi-path combiig immediately as show i Sect. 3. The chael estimates ad despreaded data may be delayed for several symbol periods ad chael estimates ca be smoothed. Pilot filter ca perform this task. 3. Movig Average Chael Estimatio iput of kth user durig th chip, T c is the chip period, h(t) is impulse respose of trasmit filter, a (I) (k) ada (Q) (k) is the data spread sequece o I ad Q chael, respectively, ad f 0 is the carrier frequecy. Base statio uses pilot chael to provide referece for all users ad the symbol o pilot chael is always 1. The pilot sigal is give by Xp(t) = A Ec(k) h(t Tc) [b (I) cos(2π f 0 t) + b (Q) si(2π f 0 t)] (2) where A is the pilot gai, b (I) ad b (Q) is the pilot spread sequece o I ad Q chael, respectively. The pilot ad data spread sequece are mutually orthogoal. The legth of spread sequece is l c. Sigle user eviromet is assumed as i Fig. 1. The received sigal after fadig chael is give by: S (t) = Ec l (Ab (I) α l cos(2π f 0 (t δ l ) + φ l ) + Ec l (Ab (Q) + x (1)b (I) a (I) (1))h(t Tc) + x (1)b (Q) a (Q) (1))h(t Tc) α l si(2π f 0 (t δ l ) + φ l ) + (t) (3) Where α l δ l ad φ l is the amplitude fadig, delay ad phase error o lth path, respectively, ad (t) is AWGN with variaceσ 2. The pilot spread sequece multiplies data sequece to idetify the specific base statio. It is assumed that the chael is sigle-path, matched filter is used at the receiver ad o timig error occurs. The the sigals o I ad Q chael after pilot despreadig are fially give by: y (I) p () = Ec(Ab (I) b (Q) + x (1)a (Q) (1)b (I) b (Q) )α si φ + Ec(A + x (1)a (I) (1))α cos φ + I () y (Q) p () = Ec(Ab (I) b (Q) + x (1)a (I) (1)b (I) b (Q) )α cos φ + Ec(A + x (1)a (Q) (1)α si φ + Q () (4) where the chael parameters α ad φ are assumed to be ivariat durig symbol period T B, I () ad Q () isthe white Gaussia oise o I ad Q chael, respectively. For expressio coveiece, Eq. (4) ca be rewritte as: y p () = y (I) p () + jy (Q) p () (5) Chael estimate is to maipulate y p () over symbol Movig average chael estimatio uses average filter for smoothig pilot despeaded sigal y p () over M symbol duratio (l c M chip duratio). The output is provided to rake receiver as chael estimates. The chael estimate durig mth symbol period ca be expressed as: ȳ M p (m) = 1 lc s=0 =1 s=0 y p,m+s () 1 = EcAlc(αm+s cos φ m+s + jα m+s si φ m+s ) 1 + lc s=0 =1 ( m+s I () + j m+s ()) = α m cos φ m + j α m si φ m + m I + j m Q (6) where y p,m+s () isthevalueofy p () at(m + s)th symbol period, α m+s ad φ m+s are the chael parameters at this period, ad m+s I () m+s Q () is the correspodet white Gaussia oise. The gai achieved by smoothig the pilot sigal over M symbol periods ca be obtaied by comparig the SNR ch of chael estimates durig oe symbol period ad the SNR ch,m of Eq. (6). SNRch= EcA2 α 2 σ 2 /lc α 2 SNRch, M = σ 2 /E c A 2 lcm SNRch, M/SNRch= M (7) Sice pilot sigal is ot modulated by data, the value of M ca be large to achieve more oise reductio gai. But M ca ot be too large, because i such sceario the chael estimates ca ot follow the chael chage. Whe M icreases, the badwidth of pilot filter decreases. For pilot filter its badwidth must be larger tha Doppler frequecy f D. So optimal M should be achieved. Aother way to improve SNR ch,m is to icrease pilot gai A, but too much pilot gai is a waste of system resource ad provides little SNR performace improvemet whe A icreases to a certai level as show i Sect. 5. Thus it is eeded to achieve optimal combiatio of M ad A for differet mobile speed zoes. Q
1276 4. Achievig Optimal M ad A for Movig Average Filter MSE method is used i this paper to obtai the optimal M for movig average filter for differet mobile statio speed zoe. It ca be achieved by miimizig the followig equatio with respect to M. F(M) = E ȳ M p (m) β(m dt) 2 where β(m) = α m cos φ m + jα m si φ m ad d t is the group delay of pilot filter. For M movig average filter d t is expressed as: M 1 dt = (9) 2 is the floor operatio. The spectrum of β(m)isgive by [3] σ 2 β (8) S β ( f ) = (10) π fd 2 f 2 where σ 2 β is the variace of β(m). Its autocorrelatio fuctio is [3] R β (τ) = σ 2 β J 0(2π f D τ) (11) where J 0 is Bessel fuctio of the first kid with zero order. The Eq. (7) ca be rewritte as F(M) = E ȳ M p (m) β(m dt) 2 p(k)p(l)j 0 (2π(k l) f D T b ) k l + J 0 (0) + 2N 0 p(k) 2 l c Ecσ 2 β k 2 p(k)j 0 (2π(k d t ) f D T b ) (12) k where p() is the impulse respose of movig average pilot filter ad N 0 is the power spectrum desity of oise over a symbol period. It is assumed M is odd. So for differet mobile statio speed zoes, that is, f D zoes, the optimal M ca be obtaied either by computig ad comparig the correspodig F(M) values, or by comparig the simulated bit-error rate(ber) directly, which is the method used i the simulatio below. As the mobile statio speed icreases, the optimal BER would degrade if the pilot gai A is kept costat, thus it will be better to icrease A so that the BER is maitaied at a reasoable costat value. We ca store the resultig M ad A values i a pre-calculated table; thereafter, the value of M ad A ca be updated adaptively accordig to the diagram show i Fig. 2. May literatures provide methods for Doppler frequecy estimatio [7]. Fig. 2 Movig average adaptive chael estimatio i DS-CDMA. 5. Simulatio Results The proposed chael estimatio method is applied to DS- CDMA forward chael for simulatio. The symbol rate is 14400 symbol/s, carrier frequecy f 0 is 2 GHz, spread factor is 16, ad QPSK modulatio is used. Oly oe user with oe pilot ad data chael is cosidered, ad the chael is assumed to be sigle-path. Jake s model is used to geerate Rayleigh fadig chael [8]. Based o such assumptios, a series of Mote-Carlo simulatios are carried out to verify the efficiecy of the proposed method. Figure 3 ad Fig. 4 show the ifluece of movig average legth M o system BER performace. Whe the mobile speed is low (5 km/h), the curve with M = 9 ca achieve better BER performace tha others ad greater M provides lower BER. While the mobile speed is high (150 km/h), the curve with M = 5 achieves lower BER istead of M = 9. This ca be explaied by otig that whe the fadig rate is low (the chael chages slowly), the chael is ot distorted by the pilot filter ad therefore Eq. (6) is satisfied; while the fadig rate is high, usig a large M prevets the pilot filter output from followig the chage of the chael, thus degradig estimatio performace. Based o the discussio i Sect. 4 ad simulatios, we obtai the desired combiatio of M ad A as show i Table 1, whe the SNR is low ( 13 db). I this table, BER 0 is the bit-error rate for the optimal M whe A = 0.5; if BER 0 > 0.15, A is icreased i steps of 0.05 to keep the BER below 0.15, ad the resultig A ad BER are also listed i the table. Figure 5 shows the BER performace of differet adaptive schemes, where M A ADP meas the adaptive scheme proposed i Table 1 ad M ADP meas the adaptive scheme that oly chages M for differet speed zoes, while keepig A=0.5. It ca be see that the proposed adaptive scheme ca achieve better BER performace tha the oe that oly adjust movig average legth M (A=0.5) ad the BER ca be kept costat regardless of the mobile speed. Compared with the case where the maximum A=0.95 is always used, the proposed scheme results i a comparable BER while savig much pilot power i low-speed sceairo. We compare the proposed scheme with the oe proposed i [3]. The adaptive scheme i [3] uses the followig first-order IIR pilot filter to process pilot sigal.
LETTER 1277 Fig. 3 System BER performace for differet M. (v=5km/h) Fig. 5 BER performace of differet adaptive schemes. (SNR= 13 db) Fig. 4 System BER performace for differet M. (v=150 km/h) Table 1 The optimal M ad A for differet speed. v(km/h) 5 40 70 90 M 29 11 11 9 BER 0 0.0425 0.1092 0.1690 0.1787 A 0.5 0.5 0.65 0.6 BER 0.0425 0.1092 0.1466 0.1460 v(km/h) 100 150 210 240 M 7 7 5 3 BER 0 0.1816 0.2115 0.2207 0.2333 A 0.7 0.8 0.95 0.95 BER 0.1500 0.1466 0.1466 0.1466 a p(z) = (13) 1 (1 a)z 1 It divides the speed of mobile statio ito 5 zoes: {[0,30 km/h), [30,60 km/h), [60,90 km/h), [90,120 km/h), [ 120 km/h)}, ad the correspodet optimal a for pilot filter is {1/16, 1/8, 1/4, 1/4, 1/2} respectively for SNR= 13 db ad A=0.5. Figure 6 shows a compariso of the results, where IIR ADP meas the adaptive proposed i [3]. It Fig. 6 BER Performaces of the proposed two adaptive schemes ad the scheme usig IIR filter. (SNR= 13 db) ca be show that our adaptive scheme that adjusts M oly achieves slightly better BER performace tha the adaptive scheme i [3], while the adaptive scheme i Table 1 performs much better tha both, because it ca keep the BER costat by icreasig the pilot gai A whe the mobile speed icreases. 6. Coclusio This paper proposes a pilot-aided adaptive chael estimatio scheme for DS-CDMA system, which is based o movig average processig of pilot sigal ad adaptive pilot gai adjustmet. The optimal combiatio of movig average legth ad pilot gai is achieved uder the sceario of imperfect chael estimatio. The simulatio results yield that the proposed scheme achieves lower BER tha the other two adaptive schemes uder the same mobile eviromet ad shows isesibility to the mobile speed. The chael es-
1278 timatio scheme is easy to implemet ad is attractive for WCDMA system which requires reliable data commuicatio for dyamic mobile eviromets, especially whe the mobile speed is high. Refereces [1] J.G. Proakis, Digital commuicatios, third ed., McGraw-Hill, New York, 1995. [2] G. Che, X.H. Yu, ad J. Wag, Adaptive chael estimatio ad dedicated pilot power adjustmet based o the fadig-rate measuremet for a pilot-aided CDMA system, IEEE J. Sel. Areas Commu., vol.19, o.1, pp.132 140, Ja. 2001. [3] J.O. Hyuk ad J.M. Cioffi, A adaptive chael estimatio scheme for DS-CDMA systems, IEEE-VTS Fall VTC 2000, vol.6, pp.2839 2843, Jue 2000. [4] P. Komulaie ad V. Haikola, Adaptive filterig for fadig chael estimatio i WCDMA dowlik, The 11th IEEE Iteratioal Symposium o Persoal, Idoor ad Mobile Radio Commuicatios, vol.1, pp.549 553, 2000. [5] J.K. Cavers, A aalysis of pilot symbol assisted modulatio for Rayleigh fadig chaels, IEEE Tras. Veh. Techol., vol.40, o.4, pp.686 693, Nov. 1991. [6] P. Schramm, Aalysis ad optimizatio of pilot-chael-assisted BPSK for DS-CDMA systems, IEEE Tras. Commu., vol.46, o.9, pp.1122 1124, Sept. 1998. [7] R. Narasimha ad C. Cox, Speed estimatio i wireless systems usig wavelets, IEEE Tras. Commu., vol.47, o.9, pp.1357 1364, Sept. 1999. [8] W.C. Jakes, Microwave Mobile Commuicatio, Wiley, New York, 1974.