NN-CLEAN Boosts Channel Modeling in Low-SNR Wireless Systems

This paper introduces Neural Network-Assisted CLEAN (NN-CLEAN), a hybrid framework designed to enhance channel modeling in challenging low-SNR wireless communication systems. The method addresses the limitations of traditional Maximum Likelihood Estimation algorithms like CLEAN, which suffer from high computational complexity due to exhaustive grid searches, and purely data-driven deep learning approaches that lack physical grounding. NN-CLEAN integrates a multi-head residual network directly into the iterative CLEAN extraction loop. By replacing the computationally intensive grid search with rapid neural network passes and delegating residual subtraction to exact mathematical models, NN-CLEAN efficiently isolates physical multipath parameters without accumulating non-physical errors. Monte Carlo simulations demonstrate that NN-CLEAN achieves over 96% accuracy at 5 dB SNR, matching traditional Grid-Search CLEAN, but with a massive reduction in computational complexity. Its near-flat scaling with batch sizes makes it a robust, real-time solution for MIMO channel estimation.