Title:Adaptive Multi-User Channel Estimation Based on Contrastive Feature Learning

Abstract:Correlation exploitation is essential for efficient multi-user channel estimation (MUCE) in massive MIMO systems. However, the existing works either rely on presumed strong correlation or learn the correlation through large amount of labeled data, which are difficult to acquire in a real system. In this paper, we propose an adaptive MUCE algorithm based on contrastive feature learning. The contrastive learning (CL) is used to automatically learn the similarity within channels by extracting the channel state information (CSI) features based on location information. The similar features will be fed into the downstream network to explore the strong correlations among CSI features to improve the MUCE performance with a small number of labeled data. Simulation results show that the contrastive feature learning can enhance the overall MUCE performance with high training efficiency.