Bayesian Joint Channel and Data Estimation for Correlated Large MIMO with Non-orthogonal Pilots

Abstract

We propose a novel joint channel and data estimation (JCDE) scheme for highly-correlated large multi-user multi-input multi-output (MIMO) systems with short non-orthogonal pilots. Bayesian JCDE via scalar-wise tensor products is a solid strategy for achieving multi-user detection (MUD) with extremely low computational cost, but the convergence property is significantly degraded in practical MIMO systems assuming spatially correlated fading channels. When using ultra-short pilots aiming at significant overhead reduction, the reliable MUD via JCDE becomes increasingly challenging. To address this issue, the proposed method estimates channel coefficients via a maximum a-posteriori (MAP)-like approach with the aid of long-term channel statistics. Furthermore, while the data detection is performed via probabilistic data association (PDA) to suppress the observation correlation, the channel estimation is performed via Gaussian belief propagation (GaBP) by leveraging the pseudo-orthogonality of the pilot-plus-data sequences, which makes it possible to realize low-complexity and high-accuracy Bayesian JCDE for highly-correlated MUD. Computer simulation demonstrates the validity of our proposed method in terms of bit error rate (BER) performance and computational cost.