Reconfigurable Intelligent Surface Aided High-Mobility Millimeter Wave Communications With Dynamic Dual-Structured Sparsity

Although reconfigurable intelligent surface (RIS) has been touted as a technology star for future wireless networks, the critical bottleneck still lies in the accurate acquisition of channel state information (CSI). The vast majority of state-of-the-art cascaded channel estimates entail the pilot overhead typically proportional to the number of RIS elements, which results in a long training time and thus may not be tolerable in high-mobility scenarios. In this paper, we investigate the channel tracking for RIS-aided high-mobility millimeter wave (mmWave) communications. By leveraging the angular domain representation of cascaded channels, we initially demonstrate the dynamic dual-structured sparsity (DDS), i.e., i) the angular cascaded channel matrices associated with different users share the identical non-zero rows while differ in their non-zero columns and ii) the cascaded channel support exhibits temporal correlation inherent to the dynamic nature of mobile channels. Then, a layered processing with dynamic dual-structured sparsity (LP-DDS) framework is customized to provide sparse priors for the exact distributions of cascaded channels. In this case, the joint estimate of the angular cascaded channel and Doppler shift is formulated as a compressive sensing (CS) problem while taking into account the temporal correlation of dynamic channels, which, however, is highly intractable due to the ill-conditioned sensing matrix. To tackle this issue, we propose an efficient algorithm, namely DDS-VBIMP, where in particular, both variational Bayesian inference (VBI) and message passing techniques are complemented each other to achieve parameter updates by taking full advantage of the sparse priors as captured by LP-DDS. We demonstrate through our analyses that the proposed DDS-VBIMP can significantly reduce pilot overhead. Simulation results reveal the superiority and robustness of the proposed DDS-VBIMP as compared to various benchmark schemes.