DEEP LEARNING COMPRESSED SENSING-BASED CHANNEL ESTIMATION FOR MASSIVE MISO SYSTEMS

With the deployment of large-scale antenna arrays, wireless channels have become sufficiently sparse that they can be analogized to an image. Consequently, the interdisciplinary integration of deep learning and channel estimation has emerged as a new research direction. In this paper, a novel channel image generation method is developed and a deep learning compressed sensing-based channel estimation network (CSCENet) is proposed for massive MIMO systems using a data-driven approach. Simulation results show that the proposed CSCENet can achieve a good performance at a large dynamically changing SNR range. Especially at low SNRs, considerable gains can be observed as compared to the benchmark channel estimation algorithm of linear minimum mean squared error (LMMSE).