RecNet: Deep Learning-Based OFDM Receiver with Semi-Blind Channel Estimation

Abstract

This paper proposes a novel deep learning-based system for channel estimation and signal detection in an orthogonal frequency-division multiplexing (OFDM) system. Different from viewing the whole OFDM system as a black box in the literature, the proposed receiving system is divided into two low-complexity neural networks (NNs). The first NN is designed for semi-blind channel estimation and the second NN recovers the original signals based on the channel state information (CSI) obtained from the first network. Simulation results show that this system offers a competitive accuracy for channel estimation. Specifically, our simulations show better robustness when a very small number of pilots are used compared with traditional channel estimation methods. With the help of the estimated CSI, the NN for signal detection converges within much fewer epochs than data-driven solutions. Due to fast convergence and small size of the NNs, this system can be trained within a short time to adapt to different channel models. Compared with the traditional schemes, our whole system has a better performance under low signal-to-noise ratio (SNR).