Deep-Learning-Assisted Channel Estimation for Adaptive Parameter Selection in mMIMO-SEFDM

Abstract

This article introduces a massive multiple-input—multiple-output (mMIMO) system that utilizes spectrally efficient frequency division multiplexing (SEFDM) and incorporates a deep neural network (DNN) for enhanced SEFDM channel estimation. Unlike existing studies on DNN-based channel estimation, this research employs estimated channel feedback to dynamically adjust SEFDM signal characteristics at the transmitter, thereby improving the system’s adaptability. This adaptive mechanism optimizes the SEFDM compression value and modulation order based on real-time channel conditions, significantly enhancing the symbol error rate (SER). Detailed simulations demonstrate that higher modulation techniques experience substantial performance degradation with increased subcarrier compression in SEFDM. The proposed DNN-based channel estimation and adaptive parameter selection outperform traditional linear schemes, utilizing a more stable SEFDM system to achieve significant spectral efficiency (SE) compared to conventional orthogonal frequency division multiplexing (OFDM).