Hybrid-Domain DNN Demodulator Under Hardware Impairments on SC-FDE Transmission

We propose two deep neural network (DNN)-based demodulators for log likelihood ratio (LLR) estimation to mitigate hardware impairments. At high-frequency bands, such as the sub-terahertz (sub-THz) wave bands expected to be used in the sixth-generation mobile communication system (6G), hardware impairments such as nonlinear distortion of power amplifiers (PA) and in-phase and quadrature (IQ) imbalance of direct IQ modulators are key challenges to be addressed. We describe the signal distortion effect of nonlinear distortion and IQ imbalance in the time-domain (TD) and frequency-domain (FD) signals. We compare the two DNN-based demodulators: 1) TD-DNN, which requires TD signals as the input data, and 2) Hybrid-DNN, which requires TD and FD signals with a conventional demodulator. The numerical results indicate that the proposed DNN-based demodulators improve the coded bit error rate (BER) performance and that the Hybrid-DNN demodulator enables more efficient learning and demodulation processing by taking advantage of the characteristics of each.