TELK OMNIKA Indonesian Journal of Electrical Engineering V ol. 12, No . 9, September 2014, pp . 6860 6867 DOI: 10.11591/telk omnika.v12.i9.4678 6860 Pilot-aided Joint Channel Estimation f or OFDM based Cooperative Multi-cell Netw orks Y uan W ang* 1 , Zhongpei Zhang 1 , and Yixian Chen 2 1 Univ ersity of Electronic Science and T echnology of China/ National K e y Labor ator y of Science and T echnology on Comm unications 1 No .2006, Xiyuan A v e , W est Hi-T ech Zone , 611731, Chengdu, China, +86-028-83201114 2 Southw est Univ ersity f or Nationalities/ College of Electr ical and Inf or mation Engineer ing 2 Y ihuan Lu South Sect.4, 610041, Chengdu, China, +86-028-85522012 *Corresponding author , e-mail: w ang.yuan@china.com.cn Abstract Multi-cell joint channel estimation (JCE) is the basis f or application of m ulti-ce ll cooper ativ e pro- cessing. F or w eakness of the e xisting algor ithms that po w er dela y profile (PDP) kno wledge of all m ultiple cells need be kno wn or pilot sequence sets of all cells m ust be identical, in this paper , a pilot-aided time domain JCE algor ithm is presented. And then, b y using Gener aliz ed Akaik e Inf or mation Cr iter ion (GAIC) to estimate PDPs of all cells f or reducing the signal space of channel estimation, the paper fur ther optimiz es this JCE algor ithm. Sim ulation results sho w that the proposed algor ithms ha v e good mean square error (MSE) and corresponding space frequency b loc k coded (SFBC) cooper ativ e m ulti-cell tr ansmission system has the good bit error r ate (BER) perf or mance too . K e yw or ds: Multi-cell cooper ation, Pilot-aided, Joint channel estimation, GAIC Cop yright c 2014 Institute of Ad v anced Engineering and Science . All rights reser v ed. 1. Intr oduction Recently , f or solving the se v ere inter-cell interf erence (ICI) prob lem in deplo yment of Or- thogonal F requ ency Division Mu ltiple xing (OFDM) based future wireless comm unication netw or ks , the emerging technologies based on m ulti-cell cooper ation and joint process are receiving more attention and research. Ne v er theless , the emplo yments of these technologies are highly depen- dent on the whole e xact channel state inf or mation (CSI) among the cooper ativ e BSs and the users . Theref ore the optim um channel estimation is the base of future wireless comm unication netw or ks . In cooper ativ e cellular systems , joint channel estimation (JCE), which is initially applied in a m ulti-user scheme , could be e xtended to a m ulti-cell en vironment and obtain better perf or mance as depicted in pub lications [1–3]. Ho w e v er , m ulti-cell JCE needs the po w er dela y profiles (PDP) kno wledge of m ultiple cells kno wn b y receiv er or pilot sequence sets of all cells being identical (in this paper , w e e xploit PDP to precisely represent m ultipath dela ys po w er and m ultipath tap locations [4], the belo w is same). I n gener al, the signals receiv ed from diff erent BSs or diff erent users will not ha v e the same PDP f or diff erent spatial locations of these BSs or users [5] and while the PDP kno wledge will not be kno wn a pr ior i to receiv er , these prob lems mak e the pr actical v alue of m ulti-cell JCE reduced se v erely . F or this w eakness , this paper presents a pilot-aided time domain JCE algor ithm f or cooper ativ e m ulti-cell netw or k and der iv es the corresponding Cr amer- Rao bound (CRB) when PDPs of m ultiple cells are not same and unkno wn. Then, b y using Gener aliz ed Akaik e Inf or mation Cr iter ion (GAIC) to estimate PDPs of all cells f or reducing the signal space of channel estimation, the paper fur ther optimiz es this JCE algor ithm. The rest of the paper is organiz ed as f ollo ws . Section 2 introduces a cooper ativ e m ulti-cell OFDM system model. Section 3 presents a pilot-aided time domain m ulti-cell JCE algor ithm and fur ther optimiz es it b y e xp loiting GAIC . Numer ical sim ulations are presented to v er ify perf or mance of the proposed algor ithms in section 4 and the paper will be concluded in last Section 5. Receiv ed December 25, 2013; Re vised J une 5, 2014; Accepted J une 28, 2014 Evaluation Warning : The document was created with Spire.PDF for Python.
TELK OMNIKA ISSN: 2302-4046 6861 Notation: In this paper , Bold letters represent a matr ix or a v ector ; d x e denotes the nearest integer larger than or equal to x ; ( ) H stands f or the conjugate tr anspose; diag( ) indicates the diagonal matr ix; I N denotes N N identity matr ix; jj jj is the Euclidean nor m; E [ ] represents e xpectation. 2. System Model C e l l   1 C e l l   2 U se r   u B S   1 B S   2 C e l l   3 B S   3 O t h e r   U s e r s B a c k h a u l Figure 1. An OFDM based m ulti-cell cooper ation scenar io with frequency reuse 1. Consider an OFDM system oper ating with a bandwidth of B w = 1 =T Hz ( T is th e sampling per iod) and consisting a total of N subcarr iers using QPSK modulation o v er frequency selectiv e Ra yleigh f ading channels . A cyclic prefix (CP) of length L C P is inser ted bef ore each symbol. Comb-type pilot patter n is e xploited to perf or m channel estimation i.e ., in each symbol, N p tones used as pilots to assist channel estimation are e v enly distr ib uted o v er N subcarr iers with equal po w er . Channel impulse response (CIR) has L m ultipath components , where each path is char- acter iz ed b y a comple x gain f actor h l ( t ) and a corresponding dela y l ( t ) : h ( t; ) = L 1 X l =0 h l ( t ) ( l ( t ) T ) (1) In pr actice , PDP of channel is considered to be constant o v er a fr ame , so t in (1) can be ignored and channel frequency response (CFR) v ector f or the k -th subcarr ier of the n -th OFDM symbol can be e xpressed as H ( n; k ) = L 1 X l =0 h l ( n ) exp( j 2 N k l ) ; k = 1 ; ; N (2) Since JCE does not need to distinguish do wnlink or uplink of system [3], an application scenar io of cooper ativ e Multi-cell OFDM system do wnlink is illustr ated as an e xample in Figure 1. N B S BSs each equipped with N T tr ansmit antennas cooper ativ ely tr ansmit to U users distr ib uted around N B S cells . Based on cooper ation, pilot sets of m ultiple cells are kno wn b y e v er y user and e xploited to estimate corresponding CSI, while all BSs could share these inf or mation. Theref ore joint process technique can be used to fur ther impro v e system perf or mance . Without loss of gener ality , w e discuss channel estimation f or case of only one user , be- cause , from signal processing point of vie w , m ultiple antenna wireless channels and m ulti-user wireless channels ha v e no intr insical diff erence , if the channels are assumed to be independent from each other both can be reduced to a Multiple-Input and Single Output (MISO) channel model [3]. So the system diag r am is depicted as Figure 2. Pilot-aided JCE f or cooper ativ e m ulti-cell netw or ks (Y uan W ang) Evaluation Warning : The document was created with Spire.PDF for Python.
6862 ISSN: 2302-4046 S p a c e - F r e q u en c y   en c o d er I F F T I F F T B T 1 I F F T I F F T B T N B S       FFT FFT   S p a c e - F r e q u e n c y   d e c o d e r C h a n n e l   E s t i m a t o r P D P E s t i m a t o r Figure 2. Pilot-aided time domain JCE model f or m ulti-cell cooper ation system. 3. Pilot-aided time domain joint c hannel estimation Because of the diff erence on location f or m ultiple cells , the PDPs of these cells are not same [5]. So , set channel order as L , w e represent the condition that PDPs of diff erent cells are not same with the maxim um time dela y set n (1) L ; ; ( N B S ) L o . Call f or each OFDM symbol, N p pilot symbols f p ( s ); s = 1 ; ; N p g are e v enly inser ted into the N subcarr iers at positions f k s ; k s = ( s 1) D f ; D f = N =N p ; s = 1 ; ; N p g with equal po w er E p = j p ( s ) j 2 and equal distance D f betw een tw o adjacent pilots , where k s indicates the subcarr ier inde x of pilot. 3.1. Pilot-aided time domain c hannel estimation algorithm Assuming the perf ect synchronization in m ultiple cells , the receiv ed signal v ector on pilot subcarr iers of the n -th OFDM symbol betw een the j -th tr ansmit antenna of i -th BS and the q -th receiv e antenna of user u can be e xpressed as: Y q p;u ( n ) = N B S P i =1 N T P j =1 X p;u;j ;i ( n ) H q p;u;j ;i ( n ) + W q p;u ( n ) = N B S P i =1 N T P j =1 X p;u;j ;i ( n ) F p h q u;j ;i ( n ) + W q p;u ( n ) = G p;u ( n ) h q u ( n ) + W q p;u ( n ) (3) Where , Y q p;u ( n ) = [ Y q p;u ( n; k 1 ) ; ; Y q p;u ( n; k N p )] T is the frequency-domain receiv ed v ector of N p dimension. X p;u;j ;i ( n ) = diag f p u;j ;i ( n; 1) ; ; p u;j ;i ( n; N p ) g indicates the pilot matr ix of di- mension N p N p with element p u;j ;i ( n; k ) representing the k -th pilot symbol of n -th OFDM symbol betw een the j -th tr ansmit antenna of i -th base station and the user u . H q u;j ;i ( n ) = [ H q u;j ;i ( n; 0) ; ; H q u;j ;i ( n; N p 1)] T denotes the CFR v ector betw een the j -th tr ansmit antenna of i -th BS and the q -th receiv e anten na of user u . While h q u;j ;i ( n ) = [ h q u;j ;i ( n; 0) ; ; h q u;j ;i ( n; N 1)] T is the corresponding CIR v ector . F p is the N p N dimension submatr ix der iv ed from the N -point FFT matr ix whose ( s; l ) -th element is denoted b y [ F p ] s:l = e xp ( j 2 k s l = N ) , s = 1 ; ; N p ; l = 1 ; ; N . So the cor responding equiv alent matr ix f or the total pilots of m ultiple cells can be e x- pressed as G p;u ( n ) = [ X p;u; 1 ; 1 ( n ) F p ; ; X p;u;N T ; 1 ( n ) F p ; ; X p;u; 1 ;N B S ( n ) F p ; ; X p;u;N T ;N B S ( n ) F p ] with dimension of N p N B S N T N . While the equiv alent CIR matr ix can be sho wn as h q u ( n ) = [ h q u; 1 ; 1 T ( n ) ; ; h q u;N T ; 1 T ( n ) ; ; h q u; 1 ;N B S T ( n ) ; ; h q u;N T ;N B S T ( n )] T of N B S N T N 1 dimension . W q p;u ( n ) is a noise v ector of dimension N p 1 , whose entr ies are TELK OMNIKA V ol. 12, No . 9, September 2014 : 6860 6867 Evaluation Warning : The document was created with Spire.PDF for Python.
TELK OMNIKA ISSN: 2302-4046 6863 assumed to be i.i.d. and comple x Gaussian distr ib uted with z ero-mean and equal v ar iance 2 w . Since PDPs remain constant o v er a fr ame , n of (3) can be ignored. Thereb y igno r ing the inde x of u and q , (3) can be re wr itten as Y p = G p h + W p (4) Ob viously , G p is an ill-conditioned matr ix. According to least squared (LS) cr iter ia, there is no cer tain solution of (4). F or tunately , CIR is alw a ys m uch shor ter than OFDM symbol length, i. e ., max f ( i ) L g N ; i = 1 ; ; N B S . Lettin g L b = d N p = ( N T N B S ) e and consider ing max f ( i ) L g L b , (4) can be rearr anged as Y p = G 0 p h + W p (5) where G 0 p = [ X p; 1 ; 1 F p;L b ; ; X p;N T ; 1 F p;L b ; ; X p; 1 ;N B S F p;L b ; ; X p;N T ;N B S F p;L b ] is a equiv a- lent matr ix with dim ension of N p N B S N T L b , F p;L b stands f or the matr ix F p retaining the first L b columns . Because h j ;i is modified as an L b -dimensional v ector , the equiv alent CIR matr ix can be sho wn as h = [ h 1 ; 1 T ; ; h N T ; 1 T ; ; h 1 ;N B S T ; ; h N T ;N B S T ] T with dimension of N B S N T L b 1 . So w e can obtain the corresponding LS estimator f or h : ^ h = ( G 0 p H G 0 p ) 1 G 0 p H Y p (6) F rom (6), the MSE of single tap of h with m ulti-cell JCE can be giv en b y M S E LS = 1 N B S N T L C P E [ jj ^ h h jj 2 ] = 2 w N B S N T L C P tr [( G 0 p H G 0 p ) 1 ] (7) 3.2. Pilots Design According to [6], if ( G 0 p H G 0 p ) 1 = I N B S N T L b = ( N p E p ) can be satisfied, (6) will be sim- plified as ^ h = I N B S N T L b G 0 p H Y p = ( N p E p ) with computational comple xity being g reatly reduced; While minim um MSE v alue 2 w = ( N p E p ) will be obtain. W e e xploit Chu sequence to design pilots which has perf ect or thogonality as [7] to meet this condition. First, w e arbitr ar ily choose a length of N p Chu sequence [8] f c ( s ); s = 1 ; ; N p g as the basis p ilot sequence f or fur ther design. The s -th element of a length of N p Chu sequence is e x- pressed as: c ( s ) = e j ( r ( s 1) 2 ) =N p ; f or ev en N p e j ( r ( s 1) s ) =N p ; f or odd N p ; s = 1 ; ; N p (8) in which r and N p are relativ ely pr ime . Then, let = L b , the pilot v alue p j ;i ( s ) sent from the j -th tr ansmit antenna of i -th BS can be descr ibed as: p j ;i ( s ) = A c ( s ) e j 2 s (( i 1) N T +( j 1)) N p (9) where A = p E p is the amplitude of pilot symbol. i = 1 ; ; N B S indicates the inde x of cell or BS , j = 1 ; ; N T denotes the inde x of tr ansmit antenna, s = 1 ; ; N p represents t he inde x of pilot symbol. Due to the perf ect per iodic autocorrelation proper ty of the polyphase Chu sequ-ence , the pilot sequences designed as (9) will be or thogonal to each other and leads to the condition of ( G 0 p H G 0 p ) 1 = I N B S N T L b = ( N p E p ) being satisfied. Then the minim um MSE of channel estimator will be ob tained. Cr amer-Rao Bound (CRB) reflects a lo w er bound on the error v ar iance f or an unbiased estimator . The CRB f or an arbitr ar y estimator ^ h j ;i of h with signal model as (5) is giv en Pilot-aided JCE f or cooper ativ e m ulti-cell netw or ks (Y uan W ang) Evaluation Warning : The document was created with Spire.PDF for Python.
6864 ISSN: 2302-4046 b y [9]: C R B ( ^ h j ;i ) = ( L 2 w ) = ( N p E p ) (10) While MSE of ^ h j ;i is defined as: M S E = E [ jj h j ;i ^ h j ;i jj 2 ] C R B ( ^ h j ;i ) (11) 3.3. Optim um algorithm In broadband wireless comm unications , m ultipath channel is sparse in gener al. The chan- nel estimation accur acy can be impro v ed b y once kno wing the PDP kno wledge of channel. In [10], GAIC cr iter ion is applied to MIMO sparse channel estimation prob lem to estimate PDP in an iter a- tiv e manner and obtain better perf or mance . W e e xtend this algor ithm to cooper ativ e m ultiple cells scheme . Because the PDPs of diff erent cells are not same and the PDPs of diff erent tr ansmit antennas in a cell are identical, f or each cell only the PDP of channel estimator betw een the first tr ansmit antenna and user , i.e . ^ h 1 ;i ; i = 1 ; ; N B S , needs to estimate . Then, f or estimator ^ h 1 ;i , the GAIC cost function f or a test channel order l has the f or m: GAI C 1 ;i ( l ) = L b = 2 ln( ^ 2 1 ;i ; l ) + ln(ln( L b ))( l + 1) (12) is a user specified par ameter , ^ 2 1 ;i ; l denotes the estima tion of noise v ar iance f or channel order l and is giv en b y ^ 2 1 ;i ; l = ( ^ h 1 ;i ^ h 1 ;i ; l ) H F H p ; L b X H p ;1 ;i X p ;1 ;i F p ; L b ( ^ h 1 ;i ^ h 1 ;i ; l ) =L b = ( ^ h 1 ;i ^ h 1 ;i ; l ) H N p I L b ( ^ h 1 ;i ^ h 1 ;i ; l ) =L b (13) where ^ h 1 ;i ; l is the channel estimator f or channel order l and padded with L b l z eros . X p ;1 ;i F p ; L b is the submatr ix from G 0 p of (5) with N p L b dimension. The GAIC test is e x ecuted as f ollo wing steps: Algorithm 1 Optimization f or JCE with GAIC test 1: Initially set P = L C P ; 2: Calculate the cost function GAI C 1 ;i ( L ) f or L = 1 ; ; P ; 3: Obtain the GAIC estimator as ^ L = arg min L f GAI C 1 ;i ( L ) g ; 4: Remo v e the eff ect of the ne wly estimated tap b y setting ^ h 1 ;i ( ^ L ) = 0 ; 5: Set P = ^ L 1 and repeat 1-3 to estimate the ne xt significant tap positions; 6: If ^ L 6 = 1 go to step 4. Then the set f ^ L g giv es the positions of significant channel taps and it could be utiliz ed to impro v e perf or mance of the estimator as ^ h 0 j ;i ( l ) = ^ h j ;i ( l ) ; l 2 f ^ L g 0 ; l = 2 f ^ L g ; j = 1 ; ; N T ; i = 1 ; ; N B S (14) In addition, this algor ithm m ust be satisfied with L b > L C P > max f ( i ) L g ; i = 1 ; ; N B S . TELK OMNIKA V ol. 12, No . 9, September 2014 : 6860 6867 Evaluation Warning : The document was created with Spire.PDF for Python.
TELK OMNIKA ISSN: 2302-4046 6865 4. Numberical Sim ulation results Sim ulations w ere carr ied out to demonstr ate MSE and bit error r ate (BER) perf or mance of the proposed algor ithms and the MSE is defined as (11). An OFDM system is sim ulated with f ollo wing par ameters: center frequency f c = 2 : 2 GH z , bandwidth B w = 5 M H z , n umber of o v er all subcarr iers N = 512 , CP length L C P = 24 . Each OFDM fr ame consists of N = 50 OFDM symbols . Comb-type pilot patter n is emplo y ed in syst em with n umber of pilot subcarr iers N p = 128 and pilot inter v al D f = 4 . The mobile speed is 30km/h. Assuming that the cells n umber N B S = 2 in sim ulation model, one BS per cell and each BS equipped with N T tr ansmit antennas , each user has receiv e antennas N R = 2 . The system e xploits SFBC and QPSK modulation. Multipath Ra yleigh f ading channels are considered as 3GPP-TR-25.996 SCM Case II channel model [11], with order L = 6 . The time dela y po w ers in PDPs of m ultiple cells is assumed to be e xponential with pdf ( t ) / exp( t ) . While the m ultipath tap locations in PDPs of m ultiple cells contain diff erent and are unif or mly distr ib uted o v er [0 ; 2510] ns . F or the con v enience of compar ison, let L b = 32 d N p = ( N T N B S ) e , E p = 1 and define Eb/N0 as ( N T E p ) = 2 w . P ar ameter is chosen as 2.5. Figure 3 sho ws the MSE perf or mance of the proposed pilot-aided m ulti-cell JCE algo- r ithm (denoted as P A-MC-JCE) and its optim um algor ithm with GAIC (denoted as GAIC-MC-JCE) compares with the corresponding CRB . F rom the figure , w e obser v e that the MSE of m ulti-cell JCE (indicated as MC-JCE) with PDPs kno wledge kno wn accur ately coincides with the cur v e of CRB and it means the estimator is unbiased. At the same time , because of using L b length to replace the accur ate PDPs kno wledge , the MSE perf or mance of P A-MC-JCE losses about 7.5 dB compared with CRB . Ho w e v er , due to optimization of GAIC , the MSE cur v e of GAIC-MC-JCE is close to CRB with Eb/N0 increasing. 0 5 1 0 1 5 2 0 1 0 -4 1 0 -3 1 0 -2 1 0 -1 1 0 0 Eb / N 0 (d B) M S E     P A -MC -JC E G A I C -MC -JC E PD Ps  kn o w n   MC -JC E C R B Figure 3. MSE perf or mance compar ison f or v ar ious m ulti-cell JCE algor ithms . Figure 4 and Figure 5 illuminate the BER perf or mance compar ison f or v ar ious m ulti-cell JCE algor ithms with tr ansmit antenna of one or tw o , respectiv ely . In Figure 4, with BS each equipped with one tr ansmit antenna the BER of MC-JCE with PDPs kno wledge kno wn has slight perf or mance loss as compared to case of perf ect channel kno wledge . And f or P A-MC-JCE, the loss is about 2 dB . While f or GAIC-MC-JCE the BER cur v e is close to the BER of MC-JCE with PDPs kno wn g r adually . In Figure 5, with BS each equipped with tw o antennas the BER perf or mance of both P A-MC-JCE and GAIC-MC-JCE get r apid and significant impro v ement and the cur v es ha v e been Pilot-aided JCE f or cooper ativ e m ulti-cell netw or ks (Y uan W ang) Evaluation Warning : The document was created with Spire.PDF for Python.
6866 ISSN: 2302-4046 close to 10 4 at lo w Eb/N0 (less than 12dB). But with Eb/N0 increasing, due to the e xistence of Doppler frequen cy shift, the BER cur v es of both algor ithms encounter error floors at about 20dB of Eb/N0. 0 5 1 0 1 5 2 0 1 0 -5 1 0 -4 1 0 -3 1 0 -2 1 0 -1 Eb / N 0 (d B) B E R     P A -MC -JC E G A I C -MC -JC E PD Ps  kn o w n   MC -JC E p e rf e ct   ch a n n e l   kn o w n Figure 4. BER perf or mance compar ison f or v ar ious m ulti-cell JCE algor ithms with N T = 1 . 0 5 1 0 1 5 2 0 1 0 -5 1 0 -4 1 0 -3 1 0 -2 1 0 -1 Eb / N 0 (d B) B E R     P A -MC -JC E G A I C -MC -C E PD Ps  kn o w n   MC -JC E p e rf e ct   ch a n n e l   kn o w n Figure 5. BER perf or mance compar ison f or v ar ious m ulti-cell JCE algor ithms with N T = 2 . 5. Conc lusion Based on m ulti-cell cooper ation OFDM systems , this paper firstly presents a pilot-aided time domain JCE algor ithm when PDPs of m ultiple cells are not same and unkno wn b y receiv er . Then, b y e xploiting GAIC , the paper fur ther optimiz es the algor ithm. F rom sim ulation results w e obser v e that, compared with the corresponding CRB , the MSE perf or mance of proposed m ulti-cell JCE algor ithms has a cer tain loss . This perf or mance gap pro vides a possibility to optimiz e the m ulti-cell JCE algor ithms and it is a fur ther need to study . TELK OMNIKA V ol. 12, No . 9, September 2014 : 6860 6867 Evaluation Warning : The document was created with Spire.PDF for Python.
TELK OMNIKA ISSN: 2302-4046 6867 Ac kno wledg ement The authors g r atefully ac kno wledge the helpful comments and suggestions of the re vie w- ers , which ha v e impro v ed the presentation.This w or k is par tially suppor ted b y F oundation f or the National Science and T echnology Major Project (No . 2012ZX03001027-003), National Natur al Science F oundation of China (No . 61101092), Fundamental Research Funds f or the Centr al Uni- v ersities (No . ZYGX2010J010), science f oundation of Sichuan pro vince (No . 2010FZ0082). Ref erences [1] Ragha v endr a M, Bhash y am S , and Gir idhar K. Interf erence rejection f or par ametr ic channel estimation in reuse-1 cellular OFDM systems . IEEE T r ansactions on V ehicular T echnology . 2009; 58(8): 4342-4352. [2] Xue P , Kang DK, and Kim DK. Impro v ed Multi-Cell Joint Channel Estimation f or TD-SCDMA Do wnlink . Proceedings of V ehicular T echnology Conf erence . Singapore . Ma y 2008: 1458- 1462. [3] Kang G, W ec k er le M, Costa E, and Zhang P . Time and F reque ncy Domain Joint Channel Estimation in Multi-br anch Systems . Proceedings of the 5th WSEAS Int.. Corfu, Greece . A ug 2005: 25-30. [4] Goldsmith A. 2005. Wireless Comm unications: Cambr idge Univ Pr , 2005. [5] Sath y endr an A, So w erb y KW , and Shafi M. A St atistical Approach to the Analysis of DS/CDMA Cellular Systems Emplo ying RAKE Receiv ers and Sector iz ed Antennas . IEEE T r ansactions on V ehicular T echnology . 1999; 8(1): 8-19. [6] Barhumi I, Leus G, and Moonen M. Optimal T r aining Design f or MIMO OFDM Systems in Mobile Wireless Channels . IEEE T r ansactions on Si gnal Processing , 2003; 51(6): 1615-1624. [7] Song B , Gui L, and Zhang W . Comb T ype Pilot Aided Channel Estimation in OFDM Systems with T r ansmit Div ersity . IEEE T r ansactions on Broadcasting . 2006; 52(1): 50-57. [8] Chu D . P olyphase Codes with Good P er iodic Correlation Proper ties (corresp .). IEEE T r ansac- tions on Inf or mation Theor y . 1972; 18(4): 531-532. [9] Ka y SM. 1998. Fundamentals of Statistical Signal Processing, V olume II: Detection Theor y . Ne w Jerse y: Upper Saddle Riv er , 1998. [10] Hw ang JK, Chung RL, and Tsai MF . Highly Efficient Sparse Multipath Channel Estimator with Optimal Chu-sequence Premab le f or MIMO F requency-Domain DFE Receiv er . Proceedings of the 2006 IEEE Inter national Symposium on Signal Processing and Inf or mation T echnology . V ancouv er , CA. A ug 2006: 755-759. [11] Spatial channel model f or Multiple Input Multiple Output (MIMO) sim ulations(Release 8). 3GPP TS 25.996 V8.0.0 . 3GPP; 2008. Pilot-aided JCE f or cooper ativ e m ulti-cell netw or ks (Y uan W ang) Evaluation Warning : The document was created with Spire.PDF for Python.