A parallel reduced-complexity filtering algorithm for sub-optimal Kalman per-survivor processing

Per-survivor processing (PSP) [Raheli et al., 1991] has been proposed for overcoming the difficulties of conventional decision-directed channel estimation for maximum likelihood sequence estimation (MLSE) in rapidly varying channels. Optimal PSP requires elaborate statistical channel modelling and complex Kalman filtering techniques. In the present paper, a novel filtering algorithm is presented for sub-optimal Kalman PSP. Simplification of the channel ARMA model and the introduction of a prediction-feedback mechanism into the Kalman recursions lead to a reduced-complexity filtering algorithm with parallel structure. Computer simulations are used to evaluate the performance of the technique in an MLSE decoder for frequency-selective Rayleigh fading channels. Results indicate that for an O(N²) reduction in algorithm complexity only a small increase in error rate occurs