Self-Supervised Zero-Shot Noise2Noise Framework for Improved Channel Estimation in RIS-Aided Multi-User Systems

Accurate channel estimation is crucial for the proper operation of reconfigurable intelligent surfaces (RIS). This paper introduces a convolutional neural network (CNN) approach for multi-user RIS channel estimation that incorporates the zero-shot noise-to-noise (N2N) methodology within its architecture. In contrast to techniques that rely on clean training data, the proposed method learns from the noisy data itself to figure out how to remove the noise. The proposed zero-shot N2N self-learning demonstrates improved performance and a fast convergence rate in the RIS channel estimation.