Variational Bayesian Inference for Time-Varying Massive MIMO Channels: Estimation and Detection

TL;DR

This paper introduces a variational Bayesian inference method for joint channel estimation and data detection in time-varying massive MIMO systems with high-mobility users, improving accuracy and adaptability.

Contribution

It develops a novel VB-based approach with online and block processing strategies for better channel tracking in high-mobility scenarios, especially when time correlation coefficients are unknown.

Findings

VB approach outperforms LMMSE, KF, and EP in SER and NMSE

Online and block strategies effectively track time-varying channels

Interleaved structure enhances online processing performance

Abstract

Massive multiple-input multiple-output (MIMO) stands as a key technology for advancing performance metrics such as data rate, reliability, and spectrum efficiency in the fifth generation (5G) and beyond of wireless networks. However, its efficiency depends greatly on obtaining accurate channel state information. This task becomes particularly challenging with increasing user mobility. In this paper, we focus on an uplink scenario in which a massive MIMO base station serves multiple high-mobility users. We leverage variational Bayesian(VB) inference for joint channel estimation and data detection(JED), tailored for time-varying channels. In particular, we use the VB framework to provide approximations of the true posterior distributions. To cover more real-world scenarios, we assume the time correlation coefficients associated with the channels are unknown. Our simulations demonstrate the efficacy of our proposed VB-based approach in tracking these unknown time correlation coefficients. We present two processing strategies within the VB framework: online and block processing strategies. The online strategy offers a low-complexity solution for a given time slot, requiring only the knowledge of the parameters/statistics within that time slot. In contrast, the block processing strategy focuses on the entire communication block and processes all received signals together to reduce channel estimation errors. Additionally, we introduce an interleaved structure for the online processing strategy to further enhance its performance. Finally, we conduct a comparative analysis of our VB approach against the linear minimum mean squared error(LMMSE), the Kalman Filter(KF), and the expectation propagation(EP) methods in terms of symbol error rate(SER) and channel normalized mean squared error(NMSE). Our findings reveal that our VB framework surpasses these benchmarks across the performance metrics.