Channel estimation using learned-VAMP network for RIS-assisted mm-wave MIMO systems

This paper addresses the challenge of channel estimation for Reconfigurable Intelligent Surfaces assisted Millimeter Wave Multi-User Multiple-Input Multiple-Output Systems. The task is complex because of the large number of antennas at the Base Station and the passive nature of RIS elements, which lack active transmitter/receiver and signal processing capabilities. In this paper, the mmWave channel is generated using the Saleh-Valenzuela channel model dataset. While compressive sensing algorithms like Orthogonal Matching Pursuit and Approximate Message Passing can be used for channel estimation, their performance is limited by fixed shrinkage functions. To overcome this limitation, a Learned Approximate Message passing network is first explored. However, the performance of the Learned Approximate Message passing network degrades for both non-i.i.d. and i.i.d. Gaussian matrices. Hence, a Learned Vector Approximate Message Passing algorithm is proposed to improve channel estimation accuracy for both matrix types. This paper presents the performance of the Learned Vector Approximate Message Passing network-based channel estimation for Reconfigurable Intelligent Surfaces assisted mmWave Multi-User Multiple-Input Multiple-Output Systems Systems, comparing it with the existing algorithms algorithms. Additionally, the impact of different shrinkage functions, such as Gaussian Mixture, Bernoulli-Gaussian, and Soft Threshold, are also analyzed within the proposed network.