A Particle Filtering Blind Equalization Algorithm for Frequency-Selective Mimo Channels with Unknown Noise Variance

This paper introduces a new fully Bayesian, particle-filter-based blind equalization algorithm for frequency-selective MIMO channels. By treating the noise variances observed by each receiver as unknown independent random variables, the proposed algorithm offers increased robustness in comparison to previous particle-filter-based methods that relied on the exact knowledge or on suboptimal estimates of those quantities. We also innovate by considering the use of convolutional codes for user separation in MIMO channels. Via numerical simulations, we verify that the proposed approach performs closely to the optimal (MAP) receiver based on the BCJR algorithm, outperforming a linear trained method for medium to low noise levels.