Channel Estimation for RIS Communication System with Deep Scaled Least Squares

Reconfigurable intelligent surfaces (RIS) are widely used in auxiliary millimeter wave communication systems due to their low energy consumption and low cost, accurate channel state information (CSI) is very important for channel estimation. However, the complexity of channel estimation is significantly increased by the fact that RIS are typically used as passive reflectors, that RISs are not capable of signal processing, and that the high complexity of RISs is caused by the abundance of reflective elements. This work suggests a deep systolic least squares channel estimation approach to solve this issue by lowering the guiding frequency overhead and increasing the channel estimation precision. We transform the problem of cascade channel estimation into the problem of noise elimination. By using scaled least square (SLS) algorithm, we can get the channel matrix containing noise, using a ResU-Net network to reduce the noise. The simulation results show that compared with the existing channel estimation method, the ResU-Net network algorithm proposed in this paper reduces the pilot frequency overhead and can significantly improve the accuracy of channel estimation.