Deep Learning-Powered Equalization with Autoencoders for Improved 5G Communication
Adomeas Asfaw Tafere, Tewodros Seble Hailemariam, Tsegamlak Terefe Debella · 2025
Fifth Generation wireless technology, commonly referred to as 5G, has significantly enhanced communication speed, capacity, and latency, revolutionizing various industries and enabling transformative applications. The complexities of wireless environments, including multipath propagation, fading, and interference, pose a challenge to these benefits. The purpose of this research is to mitigate errors caused by multipath propagation and fading in 5G communication systems. Often, these errors cause Inter Symbol Interference (ISI) and other forms of distortion. To address these issues, an equalizer tailored to 5G communication systems is proposed and evaluated. In this regard, we propose an autoencoder which utilizes deep learning techniques to extract complex features from received signals. We emphasized on mitigating errors within the context of the International Telecommunication Union’s (ITU/IMT) 2020 channel model and Quadrature Amplitude Modulation (QAM) schemes, specifcally 16-QAM and 64-QAM. Simulated results are used to analyze the performance of the proposed equalizer using key metrics such as constellation plots, Symbol Error Rate (SER), Bit Error Rate (BER), and convergence rate. Results demonstrate that the designed autoencoder achieves strong error mitigation capabilities, showing an SER of approximately 10−4and a BER of 10−4for the 16-QAM and an SER of approximately 10−3and a BER of 10−4for the 64-QAM on 5G downlink outdoor to indoor communication system.