A Deep Learning Approach to Evaluating SISO-OFDM Channel Equalization

Abstract

Channel equalization is crucial to the efficiency of wireless network systems. To improve communication reliability and decrease computing complexity, 5G networks have made great progress with the help of deep learning (DL). When applied to 5G and future networks, deep learning has been proven to increase system performance while decreasing computational complexity. ZF is often used to acquire a channel equalizer because of its inexpensive cost and lack of statistical expertise; nonetheless, ZF has a large equalizer error. Because of the limitations of the Minimum Mean Square Error (MMSE) method, deep learning models provide a more convenient solution for solving the channel equalizer issue. Since deep learning may give a better performance-complexity trade-off, it can be used to enhance MMSE and ZF channel equalizers. Its generalization and resilience also make it an attractive tool for use in this context. To address the shortcomings of the ZF and MMSE equalizers, this thesis focuses on developing a DL-based channel equalizer. An Orthogonal Frequency-Division Multiplexing (OFDM)-based single-input-single-output (SISO) system is used to measure the DL-based equalizer's efficacy. Simulation results show that the DL-based channel equalizer can achieve at least a 3 dB gain in SNR over the MMSE equalizer for a SER = 10-3 in low and high selective channel models, respectively, validating the performance of the benchmarked equalizer using different frequency selectivity levels. Furthermore, the DL-based equalizer results in a drastic decrease in computing complexity in contrast to the ZF and MMSE equalization techniques.