1 Multistage Symbol-by-Symbol Bayesian Interference Cancellation for UMTS-CDMA Links Affected by Severe Multipath Roberto Cusani 1, Jari Mattila, Marco Di Felice 1 1 INFO-COM Dpt., University of Rome La Sapienza, Italy Helsinki University of Technology, Commun. Lab., Finland
Outline Introduction CDMA system model Multistage parallel interference cancellation (MPIC) partial cancellation Bayesian cancellation Simulation results application to TDD-CDMA for UMTS channel estimation Conclusions
CDMA system model 3 y () 1 ( n) ( 1 d ) ( n) Code #1 Channel #1 AWGN CC 1 Code #1 $ ( 1) d ( n) Σ y ( K ) ( n ) M U D d ( K ) ( n ) Code #K Channel #K CC K Code #K $ ( K ) d ( n) Transmitted symbols Spreading Coherent Despreading Received combining sequences Data out received sequences of K users after coherent combining and despreading: K L g Í Í ( k) ( i, k) ( i) ( k ) y ( n) = g ( m) d ( n- m) + w ( n) i= 1 m=-l g {g (i,k) (n)} are S-CIRs {w (k) (n)} are noise sequences w (n) d (n) g (1,1) (n) g (1,K) (n) Σ y (n) w (K) (n) d (K) (n) g (K,K) (n) g (K,1) (n) Σ y (K) (n)
Partial interference cancellation 4 partial cancellation: partial cancellation 1) in multipath channel: y (n) y () (n) received sequences Decision Device d 1 g (1,) (n) ~ I 1 - + ~ y 0 1-p 1 y 1 (n) tentative symbol estimate from the previous stage d Decision Device () d 1 g (,1) (n) ~ I () 1 - + p 1 p 1 y 1 () (n) 1-p 1 d () out scale the cleaned signal sample by a fraction p 1 (0 < p 1 < 1) and the symbol estimate from the previous stage by 1 - p 1 ~ y () 0 ( y k ) ( n p y k ) ~ ( n I k ) ( n 1 p y k ) ( ) = ( ) - ( ) + ( - ) ~ ( n ) 1 1 1 1 0 value of fraction increases for successive stages 1) Divsalar, Simon, Raphaeli, Improved Parallel Interference Cancellation for CDMA, 1998
MPIC detector with partial interference cancellation 5 soft tentative decisions for QPSK constellation symbols {+1,-1,+j,- j} are obtained from the hyperbolic tangent non-linearity ~ d = F[ y] = % & ' tanh( a yr), if abs( yr) abs( yi) j tanh( a y ), if abs( y ) < abs( y ) I R I where α is a scale factor and y = y R + j y I = y (k) (n) multistage successive interference cancellation (MSIC) subtracts the interference with the most up-to-date symbol decisions
Multistage Bayesian (MB) interference cancellation 6 MPIC/MSIC receiver with symbol-by-symbol adaptive soft decisions for multipath channel and any complex modulation interference power estimate determines the scale factor of the non-linearity y (n) y () (n) received sequences d 1 g (1,) (n) ~ I 1 - y 1 (n) + d d () 1 g (,1) (n) ~ () I 1 y () 1 (n) + - d () out symbol-by-symbol adaptive soft decision device
Bayesian non-linearity 7 Bayesian non-linearity: c m + j s m, m=0,,m -1, are the constellation symbols the scale factor α is: s I s W { } d% = E d y == M -1 { } Í ( cm + jsm) exp a ( yrcm + yism) m= 0 M -1 Í m= 0 exp { a ( yc R m+ ys I m) } 7 a = 1 s = 1 s + s WI W I is the variance of the sum of ISI and MAI is the variance of the observation noise w (k) (n) magnitude: phase:
Estimate of the total interference 8 for a practical measure of s = s + s, we employ symbol-by-symbol: s WI m WI W I 1 6 m m J L K min y- g c + js, m= 0,, M -1 g is the main tap of S-CIR (typically real) using an averaged interference estimate gives worse performance hard-limiting (α = ) is used for the last stage
Application to TDD-CDMA for UMTS 9 ETSI-SMG proposals for UMTS at the unified chip rate of 4.096 Mcps: (now changed to 3.84 Mcps) Wideband W-CDMA (FDD) with spreading factors SF = 56,18,,4 Time-Division TD-CDMA (TDD) with: small spreading factor SF=16 small number of simultaneously active users (max K=8) multiplexing of users over 16 timeslots (now 15) on the same channel User #1 User # User #8 1 3 4 5 6 7 8 9 10 11 1 13 14 15 16 1... 1 3 4 5 6 7 8 9 10 11 1 13 14 15 16 1.... 1 3 4 5 6 7 8 9 10 11 1 13 14 15 16 1... time
Vehicular B channel, TDD downlink 10 Veh. B channel, QPSK, downlink, K=8, N=40, Q=16, Orth. Codes, 4.096 Mchip/s 5 1.E-01 BER 1.E-0 Relative power (db) 0-5 -10-15 1.E-03-0 0 9 18 7 36 45 54 63 7 81 CIR taps 1.E-04 1.E-05 ZF, channel known MPIC, Ns=4, channel known MSIC, Ns=3, channel known MB, Ns=4, channel estimated via correl. MB, Ns=4, channel known Ideal IC, channel known 5 10 Eb/No (db) 15 modulation QPSK @4.096Mcps SF=16, Orthogonal Walsh codes 8 users, block length = 40
Channel Estimation for TDD downlink 11 from signal theory: x(t) channel estimation for downlink: h(t) y(t) R xy (τ)=h(τ) R xx (τ) if R xx (τ)=δ(τ), then R xy (τ)=h(τ) Spreaded Midamble Spreaded Channel Received Received Received spreaded spreaded midamble sum of K users sum of K users Midamble the channel estimate is obtained by computing the cross-correlation between the (known) transmitted midamble and the received one cross-correlation channel taps
Vehicular B channel, TDD uplink 1 Veh. B channel, QPSK, uplink, K=8, N=40, Q=16, Orth. Codes, 4.096 Mchip/s 5 1.E-01 BER 1.E-0 Relative power (db) 0-5 -10-15 1.E-03-0 0 9 18 7 36 45 54 63 7 81 CIR taps 1.E-04 1.E-05 ZF, channel known MB, Ns=4, channel estimated via correl. MPIC, Ns=4, channel known MSIC, Ns=4, channel known MB, Ns=4, channel known Ideal IC, channel known 5 10 Eb/No (db) 15 modulation QPSK @4.096Mcps SF=16, Orthogonal Walsh codes 8 users, block length = 40
Channel Estimation for TDD uplink 13 User #1 User #K Spreaded Midamble Spreaded Midamble Spreaded Spreaded Channel Channel AWGN MAI between midambles is present Solution based on cross-correlation and multistage IC: for user #i, the cross-correlation method gives a tentative channel estimate other users regenerate the received midamble pertaining to user #i and subtract it from the received midamble before estimating (via crosscorrelation) its own channel the procedure is repeated until that a satisfactory channel estimate is obtained for all users
Conclusions 14 We propose for multipath channel: MPIC detector with symbol-by-symbol MB interference cancellation MPIC detector with partial interference cancellation realistic channel estimation procedure for TDD-CDMA both detectors have approximately the same complexity MB-canceller has a performance close to the ideal cancellation with known channel and has a minor loss with realistic channel estimates MB appears to be a good candidate for TDD and FDD receivers