Angle-Delay Domain Hybrid Model-Driven and Data-Driven Downlink CSI Acquisition for FDD Massive MIMO Systems

This paper investigates the downlink channel state information (CSI) acquisition in the frequency division duplex (FDD) massive multi-input multi-output (MIMO) systems, which is significantly challenged by low pilot overhead. Specifically, to fully exploit the inter-antenna and inter-subcarrier correlations, we formulate CSI acquisition as the compressed sensing (CS) problem in the angle-delay domain. Based on this model, we propose the angle-delay domain hybrid model-driven and data-driven (ADHMD) algorithm, in which the received signals are processed by the model-driven stage and data-driven stage serially. In the model-driven stage, the excellent hybrid message passing (HMP) algorithm is employed to perform initial estimation. However, this algorithm tends to produce noticeable deviations in the angle-delay tap positions, which results in significant space-frequency domain CSI acquisition performance degradation, especially under low pilot overhead. To address this issue, we present the angle-delay domain channel refinement network (ADCRN) based on multiple attention in the data-driven stage to refine the initial estimation produced by the model-driven stage. Then, the more accurate full bandwidth space-frequency channel is obtained by the Fourier transform of the refined angle-delay channel. Simulation results demonstrate that the proposed ADHMD algorithm for CSI acquisition outperforms the state-of-the-art in various scenarios.