Joint CSI Estimation-Feedback-Precoding via DJSCC for MU-MIMO OFDM Systems

As the number of antennas in frequency-division duplex (FDD) multiple-input multiple-output (MIMO) systems increases, acquiring channel state information (CSI) becomes increasingly challenging due to limited spectral resources and feedback overhead. In this paper, we investigate the impact of the feedback channel on CSI feedback in a multi-user MIMO orthogonal frequency-division multiplexing (OFDM) scenario, where the received downlink pilot signal is directly utilized as the source for CSI feedback in a joint design with CSI feedback and precoding. Considering the influence of the feedback channel, we propose an end-to-end joint CSI estimation-feedback-precoding network based on a deep joint source-channel coding architecture with an adaptive number of users. Experimental results demonstrate that, under the same feedback and CSI estimation overheads, the proposed joint multi-module end-to-end network achieves a higher multi-user downlink spectral efficiency than traditional algorithms based on separate architecture and partially separated artificial intelligence-based network architectures under comparable channel quality. Furthermore, compared to conventional separate architecture, the proposed network architecture with joint architecture reduces the computational burden and model storage overhead at the UE side, facilitating the deployment of low-overhead multi-module joint architectures in practice. Meanwhile, the network designed at the BS achieves user-number adaptability without increasing the number of trainable parameters, thereby reducing both model storage and distribution overhead by requiring only a single set of parameters for different numbers of users. While slightly increasing storage requirements at the base station, it reduces computational complexity and precoding design delay, effectively reducing the effects of channel aging challenges.

💡 Research Summary

This paper addresses the critical challenge of acquiring Channel State Information (CSI) with low overhead in Frequency-Division Duplex (FDD) Multi-User MIMO Orthogonal Frequency-Division Multiplexing (MU-MIMO-OFDM) systems, a problem exacerbated by the increasing number of antennas. The authors propose a novel end-to-end deep learning network named JEFPNet (Joint CSI Estimation-Feedback-Precoding Network) that fundamentally redesigns the traditional fragmented CSI acquisition pipeline.

The core innovation lies in a holistic, joint design philosophy. Instead of treating pilot design, channel estimation (CE), CSI compression/feedback, and precoding design as separate modules, JEFPNet integrates them into a single trainable network based on a Deep Joint Source-Channel Coding (DJSCC) architecture. A key departure from prior work is that the network at the User Equipment (UE) side takes the received downlink pilot signals directly as input, bypassing the need for explicit, potentially error-prone, traditional channel estimation. The network is trained end-to-end to extract and feedback “task-oriented semantic information” optimized for the final system objective—maximizing downlink spectral efficiency—rather than just minimizing CSI reconstruction error. The DJSCC framework jointly handles source compression and channel coding, enhancing robustness against feedback channel imperfections and mitigating the “cliff effect” common in separated designs.

A second major contribution is the network’s inherent adaptability to a variable number of users. Leveraging a Transformer-based structure, JEFPNet is designed to handle a dynamic set of user inputs without increasing the number of trainable parameters. This means only one model needs to be stored and deployed at the BS and UE, regardless of the number of active users within a defined range. This eliminates the storage, distribution, and switching overhead associated with maintaining multiple models for different user counts, a significant advantage for practical deployment.

Simulation experiments demonstrate the superiority of the proposed joint architecture. Under identical feedback overhead constraints, JEFPNet achieves higher multi-user downlink spectral efficiency compared to traditional separated architectures and partially integrated AI-based baselines. The performance gap is particularly pronounced under conditions of significant channel estimation error or poor feedback channel quality, highlighting the robustness gained from joint optimization. Furthermore, the architecture reduces computational burden and storage requirements at the UE side by eliminating explicit channel estimation. On the BS side, using a single parameter set for varying user counts reduces precoding design delay, effectively alleviating the challenges posed by channel aging.

In summary, this work presents a comprehensive, user-adaptive, and deployment-friendly framework that jointly addresses CSI estimation, feedback, and precoding through deep learning. By moving beyond modular design and focusing on end-to-end semantic extraction for the end task, it offers a promising pathway for low-overhead, high-performance CSI management in future large-scale MIMO systems.