Superimposed Pilot-based Channel Estimation for RIS-assisted IoT Systems Using Lightweight Network

Abstract

Channel estimation (CE) in Internet of Things (IoT) systems faces challenges of low spectral efficiency, high energy consumption, and blocked propagation paths. To address these issues, this paper proposes a superimposed pilot-based CE scheme with reconfigurable intelligent surface (RIS) assistance. The pilot signal is superimposed on the uplink user data to improve spectral efficiency and reduce user equipment (UE) energy consumption. RIS is introduced to enhance communication robustness in the complex propagation environments. At the base station (BS), dedicated lightweight neural networks (NNs) are developed for CE and symbol detection (SD) to alleviate computational complexity and processing delay. The limited learning ability of these lightweight NNs is addressed by employing conventional CE and SD methods for initial feature extraction. This enables the NNs to learn along with the extracted features, reducing the required training set size. The proposed scheme improves spectral efficiency, reduces energy consumption, computational complexity, and processing delay, while enhancing the performance of both normalized mean square error (NMSE) of CE and the bit error rate (BER) of SD. Simulation results demonstrate the robustness of the proposed method against various parameter settings.