Markov Chain Monte Carlo detection over time-varying frequency-selective channels

In this paper, we develop a novel list channel refinement (LCR) based Markov Chain Monte Carlo (MCMC) detector for time-varying frequency-selective channels. Information fed back from the channel decoder is exploited for both data detection and channel estimation. The LCR-MCMC detector adopts parallel Gibbs samplers to find a list of mostly likely transmitted sequences as well as matching channel impulse response (CIR). It then computes the log-likelihood ratio (LLR) of the transmitted bits based on this list. Furthermore, the estimated CIR is refined based on the LLRs to facilitate iterative joint data detection and channel estimation. A bidirectional channel estimation is performed among data segments to provide initial channel estimates. The LCR-MCMC provides a low-complexity means to approximate the optimal maximum a posterior (MAP) detection in a statistical fashion and is applicable to channels with long memory. Excellent behavior of the LCR-MCMC is presented using both synthetic channels and real data collected from actual underwater acoustic experiments.