Refined Channel Estimation for Cell-Free Massive MIMO in Dense Urban Environments Using MMSE and Deep Feedforward Neural Networks

This paper investigates a Cell-Free Massive Multiple-Input Multiple-Output (MIMO) system for next-generation Fifth/ Sixth Generation (5G/6G) wireless networks, focusing on channel estimation under dense-urban conditions with numerous obstacles (e.g., buildings, trees). We first employ a linear Minimum Mean Square Error (MMSE) estimator to derive the channel state information (CSI) between user equipments (UEs) and access points (APs). While MMSE performs effectively in moderate-to-high signal-to-noise ratios (SNRs), it shows limitations at low SNR or in harsh propagation scenarios where path loss and obstructions severely degrade the pilot signals. To overcome these shortcomings, we introduce a Deep Feedforward (DFF) neural network with multiple wide hidden layers (e.g., 3000, 2000, 1500 neurons) to predict the residual mismatch between the true channel and the MMSE estimate. By adding the learned correction to the initial MMSE estimate, our approach substantially reduces the bit error rate (BER) across all tested SNR levels, with pronounced improvements in challenging conditions. We demonstrate the effectiveness of this refined estimator in a 1000m×1000m setting containing 200 UEs and 200 APs, highlighting its promise for high-density Cell-Free Massive MIMO deployments in future 5G/6G networks.