Deep Joint CSI Feedback and Multiuser Precoding for MIMO OFDM Systems

TL;DR

The paper tackles the problem of achieving high downlink sum-rate $R_ ext{sum}$ in FDD MU-MIMO-OFDM with limited CSI feedback. It proposes JFPNet, a deep end-to-end framework that uses a DJSCC encoder at the UEs to compress CSI eigenvectors and a DJSCC decoder at the BS to reconstruct features, which are then fed into a joint multiuser precoding (JMP) and power allocation (PA) network to produce per-subband precoding directions and power allocations. The approach yields substantial gains over traditional and other DL-based methods, notably outperforming DJSCC_BD_WF and PF_BD_WF across a range of uplink SNRs and feedback overheads, with ablation studies confirming the critical roles of JMP and PA. The results demonstrate robustness to low feedback overhead and improved performance at low SNR, indicating practical potential for scalable MU-MIMO-OFDM in 6G systems. Overall, the work shows that end-to-end, task-oriented design of CSI feedback and MU precoding can significantly reduce feedback requirements while delivering near-optimal downlink performance.

Abstract

The design of precoding plays a crucial role in achieving a high downlink sum-rate in multiuser multiple-input multiple-output (MIMO) orthogonal frequency-division multiplexing (OFDM) systems. In this correspondence, we propose a deep learning based joint CSI feedback and multiuser precoding method in frequency division duplex systems, aiming at maximizing the downlink sum-rate performance in an end-to-end manner. Specifically, the eigenvectors of the CSI matrix are compressed using deep joint source-channel coding techniques. This compression method enhances the resilience of the feedback CSI information against degradation in the feedback channel. A joint multiuser precoding module and a power allocation module are designed to adjust the precoding direction and the precoding power for users based on the feedback CSI information. Experimental results demonstrate that the downlink sum-rate can be significantly improved by using the proposed method, especially in scenarios with low signal-to-noise ratio and low feedback overhead.

Figures (4)

• Figure 1: Architecture of the JFPNet. Data processing parts without training parameters are omitted from the figure. $\mathbf{O}^{\left({n_b}\right)}=\left\{{\mathbf{O}_1^{\left({n_b}\right)},\cdots,\mathbf{O}_k^{\left({n_b}\right)}}\right\}$ ($n_b=1,\ldots,N_b$) denotes the set of $K$ users' precoding vectors in $n_b$-th subband.
• Figure 2: Performance comparison between baseline methods and AI-based methods.
• Figure 3: Ablation experiment. In JFPNet, three primary modules constitute its framework: the DJSCC module, the JMP module, and the PA module.
• Figure 4: Network performance with different feedback overhead in different uplink SNR .