RISMCNet: A Two-Timescale CSI Feedback Solution for RIS-aided Multi-Carrier systems

In this letter, a novel deep learning-based solution named RISMCNet is introduced to tackle the challenge of high overhead and low accuracy in RIS-aided multi-carrier communication channel state information (CSI) feedback. The proposed method utilizes the two-timescale feature of the RIS-aided systems by designing separate feedback periods for the base station (BS)-RIS and RIS-user equipment (UE) channels. RISM-CNet leverages the sparsity of the angular-delay domain channel and apply an attention-based encoder at the UE end to compress the channel for limited feedback overhead. A multi-resolution decoder is designed at the BS end to recover high-precision CSI. Simulations results reveal that RISMCNet outperforms conventional compressive sensing (CS)-based and deep learning-based methods in CSI recovery accuracy. Moreover, RISMCNet achieves 270 to 810 times faster running times compared to CS-based methods.