Channel Estimation for Cell-Free mmWave Massive MIMO Through Deep Learning

The combination of cell-free massive multiple-input multiple-output (MIMO) systems along with millimeter-wave (mmWave) bands is indeed one of most promising technological enablers of the envisioned wireless Gbit/s experience. However, both massive antennas at access points and large bandwidth at mmWave induce high computational complexity to exploit an accurate estimation of channel state information. Considering the sparse mmWave channel matrix as a natural image, we propose a practical and accurate channel estimation framework based on the fast and flexible denoising convolutional neural network (FFDNet). In contrast to previous deep learning based channel estimation methods, FFDNet is suitable a wide range of signal-to-noise ratio levels with a flexible noise level map as the input. More specifically, we provide a comprehensive investigation to optimize the FFDNet based channel estimator. Extensive simulation results validate that the training speed of FFDNet is faster than state-of-the-art channel estimators without sacrificing normalized mean square error performance, which makes FFDNet as an practical channel estimator for cell-free mmWave massive MIMO systems.