Performance of Turbo Codes in Interleaved Flat Fading

Performance of Turbo Codes in Interleaved Flat Fading Channels with Estimated Channel State Information 48 th Annual Vehicular Technology Conference Ottawa, Canada May 19, 1998 VIRGINIA POLYTECHNIC INSTITUTE & STATE UNIVERSITY MPRG MOBILE & PORTABLE RADIO RESEARCH GROUP Matthew Valenti and Brian D. Woerner Mobile and Portable Radio Research Group Virginia Tech Blacksburg, Virginia

Introduction n n u Fading is Rayleigh distributed. u The channel is “fully-interleaved”. u Perfect channel estimates are available. n 5/19/98 Turbo codes have been shown to achieve remarkable performance over Rayleigh flat fading channels. Typical assumptions in the literature: The purpose of this presentation is to investigate the validity of these assumptions.

Prior Work n Effect of channel estimation. u Jordan and Nichols, MILCOM ‘ 96 F Noise Introduction u Hoeher, F Set variance estimation errors. Int. Symp. on Turbo Codes ‘ 97 the variance estimate equal to a constant. u Summers and Wilson, Trans. Comm. Apr. ‘ 98 F Proposed n an SNR estimator Correlated fading and channel interleaving. u Hall and Wilson, JSAC Feb. 98 F Exponentially 5/19/98 correlated channel. F Block and systematic channel interleaving.

Outline of Talk n General channel model. can be Rayleigh or Rician. u Fading is correlated using Clarke’s model. Introduction u Fading n Channel estimator u Estimates variance. u Based on an FIR filter. n Simulation study u SOVA 5/19/98 fading amplitude and noise and MAP decoding algorithms u Correlated Rician/Rayleigh fading with channel interleaving.

System Model 5/19/98 Turbo Encoder Channel Interleaver De. Interleaver Channel Estimator Turbo Decoder Deinterleaver BPSK Modulator BPSK Demod.

Channel Model System Model n Multiplicative fading amplitude: u xk and yk are i. i. d. , u Each has autocorrelation u Ratio of specular to diffuse energy: fading: = 0 F Rician fading: > 0 F Rayleigh 5/19/98

Encoder and Decoder n Encoder System Model u Constraint length K=3 RSC encoders. u Frame/interleaver size of 1, 024 bits. u Randomly designed interleaver. u Rate r=1/2. Odd/even puncturing. n Decoder u 8 5/19/98 decoder iterations. u (Improved) SOVA, Papke et al ICC ‘ 96 u Log-MAP, Robertson et al, Euro. Trans Telecomm. Mar ‘ 97

Effect of correlated fading and interleaving 1 10 n BT =. 0025, no interleaving BT =. 005, no interleaving BT =. 01, no interleaving BT =. 0025, block interleaving BT =. 005, block interleaving BT =. 01, block interleaving fully interleaved 0 10 -1 Turbo Code Performance in flat Rayleigh fading. u BER Introduction 10 u -2 10 u -3 10 u -4 10 u -5 10 -6 10 0 1 5/19/98 2 3 4 5 E b /N o in d. B 6 7 8 9 10 Parameterized by type of interleaving and BT 32 by 64 channel interleaver. 8 iterations of Improved SOVA decoding. Poor performance for all BT with no channel interleaver. Performance degrades with channel interleaver as BT decreases.

Proposed Channel Estimator FIR LPF n Fading amplitude estimator Channel Estimation u FIR filter, order N=32. u Lowpass with cutoff at fd. absolute value n Compute Sample Variance Noise variance estimator u Take 5/19/98 sample variance of estimated noise magnitude. u Constant required to unbias the estimates.

Effect of channel estimation in Rayleigh fading 0 10 SOVA, noise and fade estimates SOVA, fade estimates only SOVA, perfect channel info MAP, noise and fade estimates MAP, fade estimates only MAP, perfect channel info -1 10 n Performance in flat Rayleigh fading. u Introduction u -2 BER 10 u BT =. 005 Block channel interleaving 8 iterations of decoding: F F -3 10 u -4 10 u MAP performs 2. 0 d. B better than SOVA. Slight penalty for fade estimates: F -5 10 F u -6 10 5/19/98 0 1 2 3 4 E b /No in d. B 5 6 7 8 Improved SOVA. log-MAP. 0. 25 d. B for SOVA 0. 75 d. B for MAP No penalty for noise variance estimates.

Effect of channel estimation in Rician fading 0 10 SOVA, noise and fade estimates SOVA, fade estimates only SOVA, perfect channel info MAP, noise and fade estimates MAP, fade estimates only MAP, perfect channel info -1 10 n Performance in flat Rician fading. u Introduction u -2 u BER 10 BT =. 005 and = 1. Block channel interleaving. 8 iterations of decoding: F F -3 10 u -4 10 u MAP performs 1. 5 d. B better than SOVA. Slight penalty for fade estimates: F -5 10 F u -6 10 5/19/98 0 1 2 3 4 E b /No in d. B 5 6 7 8 improved SOVA log-MAP 0. 25 d. B for SOVA 0. 5 d. B for MAP No penalty for noise variance estimates.

Conclusion n n 5/19/98 It is important to incorporate the effects of channel correlation and interleaving when simulating turbo codes over fading channels. A simple FIR filter can be used to estimate the fades with only slight loss in performance. Performance is insensitive to noise variance estimates. MAP algorithm is considerably superior to SOVA in severe fading. u MAP is more sensitive to estimation.

Future Work n The fading amplitude estimator could be improved. Conclusion u Requires knowledge of Doppler frequency. u A Kalman filter could be used instead. n n Effects of estimating carrier phase should be considered. Estimation could be absorbed into the turbo decoding algorithm. u Estimate 5/19/98 channel after each iteration. u Use new estimates during next iteration.