Deep-Learning Based Channel Estimation for OFDM Wireless Communications

Abstract

Multi-carrier technique is a backbone for modern commercial networks. However, the performances of multi-carrier systems in general depend greatly on the qualities of acquired channel state information (CSI). In this paper, we propose a novel deep-learning based processing pipeline to estimate CSI for payload time-frequency resource elements. The proposed pipeline contains two cascaded subblocks, namely, an initial denoise network (IDN), and a resolution enhancement network (REN). In brief, IDN applies a novel two-steps denoising structure while REN consists of pure fully-connected layers. Compared to existing works, our proposed processing architecture is more robust under lower signal-to-noise scenarios and delivers generally a significant gain.