Deep Neural Network Denoising for Enhanced CSI Feedback in FDD Multi-User MIMO Systems

In 5G and 6G communication systems that employ multi-user MIMO architectures, accurate channel estimation remains a critical challenge. The inherent sparsity of the channel and the effects of random projection further complicate this task. Compressive sensing (CS) techniques have been explored to address these issues; however, their reliance on operations such as the channel state information network (CsiNet) often leads to increased computational complexity. In this paper, we used a deep learning (DL) based method utilizing a deep neural network denoising (DNNet-DeNo) framework for improved channel state information (CSI) feedback in FDD multi-user MIMO systems. Simulation results demonstrate that the proposed method consistently outperform existing DL-based algorithms across various signal to noise ratio (SNR) circumstances. The normalized mean square error (NMSE) and cosine similarity ($\rho$) versus SNR analyses further highlight the superior performance and robustness of our model compared to existing DL-based methods.