In this paper, a novel image-based Deep Learning (DL) approach for channel estimation for future wireless communications is proposed. The time–frequency response of the fast-fading wireless channel is represented as a 2D image to estimate the unknown values of the channel response using known values at the pilot locations. With given images, both image Super-Resolution (SR) and image Denoising Network (DnN), termed as Super-Resolution and Denoising Network (SRDnN), are combined to estimate the wireless channel. To show the effectiveness, the proposed SRDnN is applied to Massive Multiple-Input Multiple-Output (mMIMO) with Non-Orthogonal Multiple Access (NOMA). The enhanced performances of SRDnN are quantified in terms of Mean Square Error (MSE) and Symbol Error Rate (SER). In addition, the influence of pilot numbers on SRDnN performance for next-generation mMIMO-NOMA networks is presented. The simulation results demonstrate that SRDnN is comparable to the level of Maximum Likelihood (ML) detection for both with and without complete Channel State Information (CSI) at the receiver with less number of pilots.
Read full abstract