Deep Learning Based Channel Covariance Matrix Estimation With User Location and Scene Images

Channel covariance matrix (CCM) is one critical parameter for designing the communications systems. In this paper, a novel framework of deep learning (DL) based CCM estimation proposed that exploits perception transmission environment without any channel sample or pilot signals. Specifically, as affected by user's movement, we design neural network (DNN) to predict from user location and speed, corresponding method named ULCCME. A denoising further developed reduce positioning error improve robustness For cases when information not available, propose an interesting way uses environmental 3D images CCM, SICCME. Simulation results show both methods are effective will benefit subsequent estimation.