Particle filtering applied to turbo equalization with channel and noise variance estimation

We use the particle filtering approach to develop a new Bayesian equalizer for turbo equalization under conditions of imperfect estimation of the channel impulse response (CIR) and the noise variance. This new equalizer is derived on the basis of fixed-lag smoothing, using the SIS (Sequential Importance Sampling) methodology to approximate the conditional probability distribution of a block of transmitted symbols, given the corresponding block of channel outputs and the estimates of the unknown parameters (channel gains and noise variance). Simulation results of performance evaluation and comparison with the MMSE equalizer proposed in [1] are provided, under different conditions of errors in the estimation of the mentioned parameters. It is shown that the proposed scheme significantly outperforms the SISO (soft-input soft-output) equalizer of [1], specially when the estimation errors are more significant.