Deep neural network based downlink power domain multi-user NOMA-OFDM signal detection

Abstract

Non-orthogonal multiple access-based orthogonal frequency division multiplexing (NOMA-OFDM) is a promising waveform-based multiple access technology for future wireless networks for multiple-user symbol transmission (MUST) in the same time-frequency resource block. However, it differs in the power domain, enhancing its spectrum efficiency. This is essential to meet the high data rate required for ever-increasing connected devices and the Internet of Things (IoT). However, NOMA-OFDM systems suffer from impairments such as imperfect successive interference cancellation (SIC) caused by channel impairments like channel fading, carrier frequency offset, and non-linearity caused by non-linear power amplifiers. This paper identifies and addresses the key impairments mentioned in the NOMA-OFDM system and proposes DNN-based estimation in offline training and detection in online testing for downlink power domain multi-user NOMA-OFDM symbols. The reported 2 dB SNR gain compared to least square-SIC/minimum mean square error SIC-based methods is a significant finding and demonstrates the robustness of the proposed DNN-aided approach against various channel impairments.