Channel Estimation for Active RIS-Assisted Millimeter-Wave MIMO Systems

Most studies of RIS-assisted millimeter-wave massive MIMO systems are based on known channel state information, in which channel estimation is challenging due to the fact that RISs are usually passive element relays, resulting in a high guide frequency overhead. First, a compressive sensing algorithm (SOMP) is used to reconstruct the channel matrix considering the common sparsity of the millimeter-wave MIMO channels in the angular domain between different subcarriers. Secondly, a dual-channel denoising residual dense network (DCRDN) is proposed, which not only can simultaneously input the real and imaginary parts of the two-dimensional noisy channel matrix into the complex convolutional network for training, but also its dual-channel structure can be designed to extract and fuse different levels of feature information. The final simulation results show that the algorithm proposed in this paper can obtain better NMSE performance for different signal-to-noise ratios.