Deep Learning-based Channel Predictor for RIS-assisted NOMA

This paper explores the integration of non-orthogonal multiple access (NOMA) and reconfigurable intelligent surfaces (RIS) as key technologies to address the wireless communications challenges. NOMA improves spectral efficiency by enabling resource sharing among multiple users, while RIS enhances signal quality and coverage with lower energy consumption. However, this integration introduces complexity and non-linearity in channel estimation. To tackle this, we employ deep learning (DL) models, specifically convolutional neural networks (CNN) and long short term memory (LSTM) networks, to improve channel state prediction. Our main contribution is a new DL model with additional layers for more accurate magnitude and phase prediction. Simulations demonstrate that the proposed model reduces average inference time by 17%, decreases the number of training parameters by over 35%, and showcases signal-to-noise ratio (SNR) gains for fixed bit-error rate (BER).