Recursive Gated CNN-Based Beamfocusing Algorithm for Massive MIMO System

doi:10.1109/tgcn.2026.3740665

Recursive Gated CNN-Based Beamfocusing Algorithm for Massive MIMO System

2026-10-06

0

PRE

AI

DOI：10.1109/tgcn.2026.3740665

Original

Original request for help

Save

Abstract

En 中文

Considering the beam splitting problem caused by frequency-independent phase shifter (PS) in massive MIMO, this paper proposes a recursive gated convolutional neural network (RGCNN)-based beamfocusing (BFC) algorithm, named RGCNN-BFC algorithm. Specifically, the channel state information is preprocessed by computing the mean channel covariance matrix (MCCM). To compensate for the phase deviation, a precoding structure combining TTD and PS is constructed to generate frequency-dependent phase shifts. The delay, phase, and digital precoding labels are generated by maximizing the array gain under this architecture. Then, a RGCNN structure is proposed to improve the prediction accuracy of the precoding labels. By multiple rounds of feature interaction, RGCNN can capture more complex spatial dependencies, which can better learn the global features of the input MCCM. Finally, the well-trained RGCNN can rapidly and exactly output the predicted weight vector that can compensate for the phase shifts without complex matrix operations. Simulation results demonstrate that the proposed RGCNN-BFC algorithm can mitigate the beam split effect and achieve near-optimal achievable rate performance.

Keywords:

Massive MIMO

beamfocusing

true time-delays (TTDs)

convolutional neural network

beam split

recursive gated convolution

Recursive Gated CNN-Based Beamfocusing Algorithm for Massive MIMO System

Abstract

Journal

Organization

Cited Papers