Deep Learning-Based Hybrid System for Multiuser MIMO Systems

Abstract

Hybrid computation is an important step in multiple-user mm Wave MIMO systems to reduce complexity and expense while obtaining an acceptable sum-rate. Prior research on hybrid precoding was frequently driven by optimization or pessimistic methods. These techniques either offer more intricacy or function less than optimally. Furthermore, the quality of the channel information plays a significant role in how efficient these tactics are. In this article, we introduce a deep learning (DL) method that improves productivity while requiring less computation time than current approaches. In reality, we employ MIMO using a convolutional neural network (CNN) (CNN-MIMO) in order to generate an analogue precoder and combiners using a false channel matrix as the source. The process is split into two main sections. We first create a successful detection method, which allows us to choose the best precoder and combiners from a predetermined coding system while achieving the highest possible total rate as result labels, the chosen compiler and precoders are employed when the input-output pairs are generated during the CNN-MIMO training stage. We assess the effectiveness of the applied methodology using a variety of intricate computations and demonstrate that the suggested DL framework performs better than conventional methods. CNN-MIMO provides a potent hybrid precoding in the presence of channel matrix errors. technique. The suggested method is also easier to calculate than an alternative relying on codebooks and optimization.