A Novel Extreme-Learning-Machine Aided Receiver Design for THz-SM with Hardware Imperfections

Terahertz (THz) communication is promising as it can enable ultra-wide-band and ultra-high-rate for various emerging communication services. In this letter, we propose to exploit the extreme learning machine (ELM) network based regressor for simple and low-complexity joint channel estimation (CE) and signal detection (SD) for THz-band spatial modulation (THz-SM) communications impaired by hardware imperfections. Computer simulations show the performance superiority of the proposed joint CE/SD scheme when compared with the state-of-the-art schemes, and other machine learning-based ones, including the support vector machine (SVM), deep neural network (DNN) and some variants of ELM. Specifically, we show that its bit error rate (BER) performance approaches to that of the recently derived maximal likelihood (ML) SD. In addition, the robustness of the proposed scheme is validated by considering two types of background impulsive noises.