Heterogeneous Edge‐Enhanced Circular Dilated Convolutional Graph Attention Network–Based PAPR Reduction Scheme for Enhanced MIMO‐OFDM Systems

ABSTRACT High spectral efficiency and robust wireless communication have been made possible by the rapid evolution of multiple‐input multiple‐output orthogonal frequency division multiplexing (MIMO‐OFDM) systems that have enhanced wireless communication in 5G and beyond. These developments allow exceptionally high spectral efficiency and provide communications that are remarkably reliable. Methods that are traditionally used to cut down peak‐to‐average power ratio (PAPR) are excessive clipping, selective mapping, and tone reservation. All of which, in one way or another, are incapable of pragmatic use due to the distortion of the signal, considerable complexity, and the requirement of additional information. In light of these limitations, this research introduces the heterogeneous edge‐enhanced circular dilated convolutional graph attention network (Het‐ECDCGAN)–based PAPR reduction framework. It draws on techniques from circular dilated convolutional neural networks (CD‐CNNs) and heterogeneous edge‐enhanced graph attention networks (HE‐GATs). The CD‐CNN successfully separates long‐range information from periodic peak structures in the time–domain, while the HE‐GAT utilizes graph‐based attention techniques to model inter‐subcarrier and inter‐antenna correlations. Binary Portia spider optimization is applied to optimize the performance of Het‐ECDCGAN. The designed framework can reach a bit error rate (BER) of up to 10 −4 , which guarantees good transmission and efficient PAPR reduction. In line with the validation of resilience at varying spectral efficiencies, modulation schemes, which include QPSK, 16‐QAM, and 64‐QAM schemes, are tested. All simulation experiments for MIMO‐OFDM signal generation, PAPR reduction, and performance evaluation of the proposed Het‐ECDCGAN framework are carried out using MATLAB.