Integration of Deep Learning and Compressive Sensing Algorithm for Beamspace Channel Estimation in mmWave Massive MIMO Systems

Document Type

Article

Publication Title

IEEE Access

Abstract

In the realm of millimeter wave (mmWave) communication, massive multiple-input-multiple-output (mMIMO) systems have grabbed greater attention due to advancements in the 5G cellular networks. This mMIMO with a lens antenna array has proven its efficacy by limiting the number of radio frequency (RF) chains, thus minimizing hardware cost, MIMO dimension, and power consumption. However, channel estimation is tedious with limited RF chains. In addition, the use of compressive sensing (CS) algorithms for signal recovery hampers their practical implementation due to several iterations causing high complexity. To address these concerns, this proposal investigates and develops a deep learning (DL)-based beamspace channel estimation approach. This is achieved by refining the learned approximate message passing (LAMP) network by deriving a new soft threshold shrinkage function to develop a new mixture of Gaussian distribution-LAMP (MoG-LAMP) algorithms. This captures the prior distribution of the beamspace channel. In each MoG-LAMP layer, the Gaussian variance parameters of the sparse channel are updated by a deep neural network, which efficiently learns complicated channel sparsity structures. Furthermore, this developed algorithm is extended to propose a hybrid approach, which is a combination of LAMP and MoG-LAMP, to showcase its effectiveness in achieving higher channel estimation accuracy. The simulation results are verified with the DeepMIMO dataset and the Saleh–Valenzuela (SV) channel model. The developed MoG-LAMP algorithm outperforms conventional methods, delivering satisfactory accuracy even at low SNRs. The proposed hybrid approach yields improvements of 38% and 26% over the LAMP and MoG-LAMP algorithms, respectively, at an SNR of 10 dB.

First Page

181200

Last Page

181216

DOI

10.1109/ACCESS.2025.3622485

Publication Date

1-1-2025

This document is currently not available here.

Share

COinS