Deep Learning for Channel Estimation of 6G RIS-NOMA Systems with Hardware Impairments

Reconfigurable intelligent surfaces (RIS) integrated with non-orthogonal multiple access (NOMA) are essential technologies for enhancing spectrum efficiency and coverage in sixth-generation (6G) wireless networks. Accurate channel estimation in RIS-NOMA systems is challenged by hardware impairments and user mobility, which introduce distortions and dynamic channel conditions. This paper proposes a hybrid convolutional neural network-long short-term memory (CNN-LSTM) model designed to improve channel estimation in RIS-NOMA environments affected by hardware impairments. The model leverages CNNs to extract spatial features from high-dimensional channel matrices and employs LSTMs to capture temporal variations resulting from user movement. Additionally, a dataset generation algorithm is developed to simulate time-varying RIS-NOMA scenarios with realistic hardware impairments. We evaluated the proposed model performance across various noise factors, demonstrating its capability to maintain accurate channel estimation under different hardware impairment levels and quantization bit settings, as measured by root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). Across different settings in RIS-NOMA systems, the proposed model achieved its best performance with an RMSE of 0.00186, MAE of 0.00148, and MAPE of 0.06501. Our findings highlight the model’s adaptability and robustness, providing valuable insights for reliable channel estimation in complex RIS-NOMA systems.