Performance Evaluation and Analysis of Deep Learning Autoencoder-Based Wireless Communication System

Abstract

Recent advancements in deep learning have led to the emergence of autoencoder-based (AE) wireless communication systems, presenting a promising approach to tackle the challenges posed by conventional mathematical models. In this research, a thorough evaluation and analysis of the performance of deep learning AE-based wireless communication systems is conducted. Specifically, our investigation focuses on employing the additive white Gaussian noise (AWGN) channel and explores the impact of varying signal-to-noise ratio (SNR) conditions during the AE training process on system performance. The results show that training the autoencoder under diverse SNR conditions, particularly with an extended number of epochs, surpasses the performance of a fixed-trained autoencoder regarding block error rate (BLER). Additionally, a comparison of BLER between multiple AE models and traditional mathematical representations used in communication systems is conducted. The obtained findings indicate that deep learning AE-based wireless communication systems exhibit promising performance compared to conventional models, underscoring their potential as an effective solution for wireless communication systems.