Attention Mechanism Based Intelligent Channel Feedback for mmWave Massive MIMO Systems

TL;DR

This paper introduces EMEVNet, an attention-based neural network architecture that efficiently encodes eigenmatrix and eigenvector feedback for mmWave massive MIMO systems, improving CSI feedback accuracy with low overhead.

Contribution

The paper presents a novel attention mechanism-based neural network for joint eigenmatrix and eigenvector feedback, addressing limitations of existing CSI feedback schemes.

Findings

Significant performance improvement over traditional methods

Enhanced robustness with low feedback overhead

Effective joint encoding of eigenmatrix and eigenvector

Abstract

The potential advantages of intelligent wireless communications with millimeter wave (mmWave) and massive multiple-input multiple-output (MIMO) are based on the availability of instantaneous channel state information (CSI) at the base station (BS). However, no existence of channel reciprocity leads to the difficult acquisition of accurate CSI at the BS in frequency division duplex (FDD) systems. Many researchers explored effective architectures based on deep learning (DL) to solve this problem and proved the success of DL-based solutions. However, existing schemes focused on the acquisition of complete CSI while ignoring the beamforming and precoding operations. In this paper, we propose an intelligent channel feedback architecture using eigenmatrix and eigenvector feedback neural network (EMEVNet). With the help of the attention mechanism, the proposed EMEVNet can be considered as a dual channel auto-encoder, which is able to jointly encode the eigenmatrix and eigenvector into codewords. Simulation results show great performance improvement and robustness with extremely low overhead of the proposed EMEVNet method compared with the traditional DL-based CSI feedback methods.