Extreme Learning Machine-Based Feature Refinment for Channel Estimation in RIS-ISAC Systems
Alice Faisal, Ibrahim Al-Nahhal, Octavia A. Dobre, Hyundong Shin · IEEE Transactions on Vehicular Technology · 2025
Integrated sensing and communication (ISAC) systems have emerged as a key enabler for future wireless networks, aiming to optimize spectral resource utilization for both sensing and communication tasks. The incorporation of reconfigurable intelligent surfaces (RIS) with ISAC enables more efficient utilization of resources, improving the quality of communication and the accuracy of sensing. A critical aspect of deploying such systems reliably is accurate channel estimation. Traditional deep learning methods, though effective, often struggle with the intricate characteristics of communication channel matricies. This work introduces an innovative two-stage channel estimation approach for RIS-ISAC systems. The first stage focuses on feature refinement using an extreme learning machine framework to process the received signals and extract essential channel features. The second stage employs these refined features to estimate the desired channels through dedicated neural networks specifically designed for sensing and communication tasks. The numerical simulations demonstrate that the proposed two-stage approach significantly outperforms the existing techniques across various system configurations. Moreover, the proposed method achieves a remarkable computational complexity reduction as compared to the state-of-the-art works. The results prove the robustness and efficiency of the proposed approach, facilitating more robust RIS-assisted ISAC deployments.