Deep Learning Based Joint Hybrid Precoding and Combining Design for mmWave MIMO Systems

To maximally enhance the hybrid precoding performance, an innovative deep learning-based optimization algorithm is given for a multiuser fully connected (MUFC) structure in this article. First, a joint optimization framework based on convolutional neural network (CNN) is suggested, namely, JOCNN, wh...

Full description

Saved in:
Bibliographic Details
Published in:IEEE systems journal Vol. 18; no. 1; pp. 560 - 567
Main Authors: Liu, Fulai, Li, Xiaoyu, Yang, Xianghuan, Shi, Huiyang, Shi, Baozhu, Du, Ruiyan
Format: Journal Article
Language:English
Published: New York IEEE 01.03.2024
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Subjects:
Algorithms
Artificial neural networks
Convolutional neural network (CNN)
Convolutional neural networks
Deep learning
deep learning (DL)
hybrid precoding
Machine learning
Millimeter wave communication
Millimeter waves
millimeter-wave (mmWave)
MIMO
multiple-input multiple-output (MIMO)
Neural networks
Optimization
Precoding
Radio frequency
Training
Unsupervised learning
ISSN:1932-8184, 1937-9234
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:To maximally enhance the hybrid precoding performance, an innovative deep learning-based optimization algorithm is given for a multiuser fully connected (MUFC) structure in this article. First, a joint optimization framework based on convolutional neural network (CNN) is suggested, namely, JOCNN, which can synchronously optimize the hybrid precoder and combiner. In the JOCNN, analog precoding neural network (AP-NN), digital precoding neural network (DP-NN), analog combining neural network (AC-NN), and digital combining neural network (DC-NN) are designed to satisfy the nonconvex constant modulus constraint and the power constraint. Then, using a training strategy based on unsupervised learning, the developed JOCNN can be trained to maximize the spectral efficiency (SE). At last, the thoroughly trained JOCNN model can take in the estimated channel matrix as the input and immediately output the analog precoding matrix, the digital precoding matrix, the analog combining matrix, and the digital combining matrix. Simulation results demonstrate that the suggested method can outperform the previous works in terms of SE while requiring less computation time.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1932-8184
1937-9234
DOI:10.1109/JSYST.2024.3357712