Analysis of CNN-LSTM Assisted Ris Cooperative Noma System

This work evaluates the performance gain of a reconfigurable intelligent surface (RIS)-assisted massive multiple-input multiple-output (M-MIMO) cooperative non-orthogonal multiple access (C-NOMA) system under time-selective Nakagami-m fading channels, focusing on symbol error rate (SER) and outage probability (OP) in high-mobility environments. A baseline user-pairing strategy reveals that RIS phase shift cooperative relaying significantly enhances the end-to-end reliability of the far users compared to the traditional NOMA networks. However, performance is still limited due to the presence of the residual successive interference cancellation (SIC) and channel state information estimation errors. Simulation results demonstrate the significant impact of fading severity and Doppler-induced spectral broadening on the reliability of RIS-NOMA networks, exposing the constraints of conventional optimization-based and single-model deep-learning (DL) methodologies. To mitigate these problems, a convolutional neural network long short-term memory (CNN-LSTM) hybrid framework is considered. The proposed framework integrates CNNs for spatial feature extraction with LSTM units to capture temporal dependencies in time-selective fading channel conditions. Through extensive simulations, the results confirm that the proposed scheme outperforms conventional NOMA, stacked long short-term memory (S-LSTM), and bidirectional long-short term memory (Bi-LSTM) DL benchmarks. Specifically, it attains a near-user and far-user SER reduction of up to 3.5 × 10–³ and 1.39 × 10–² at 20 dB SNR, respectively, while consistently attaining the minimal OP across all conditions. These outcomes indicate that DL-enabled spatiotemporal modeling, coupled with RIS-assisted beamforming, provides a robust, scalable, and energy-efficient solution for fifth-generation (5G) and beyond 5G (B5G) networks. The proposed framework is well-suited for time selective channel, high-speed vehicular communications and mission-critical applications defence application, facilitating the design of intelligent, low-latency, and resilient SIC estimators.