Robust Deep Learning-Based End-to-End Receiver for OFDM System With Non-Linear Distortion

In this letter, we propose a deep learning-based receiver named one-dimensional transmit and recovery net (1D-TRNet), which is robust to non-linear clipping distortion for orthogonal frequency-division multiplexing (OFDM) systems. The proposed scheme uses a 1D U-shape structure with multiple gated recurrent units (GRUs). The 1D-convolution kernel can extract both the low-level (local) and high-level (global) features in the non-linearly distorted OFDM signals. Simulation results show that the proposed 1D-TRNet receiver achieves significant bit-error rate (BER) performance gain over the traditional OFDM receivers. Moreover, the proposed 1D-TRNet receiver outperforms the state-of-the-art generalized approximate message passing (GAMP) receiver under the same computational complexity constraint.

I. Introduction

MUlti-carrier techniques such as orthogonal frequency-division multiplexing (OFDM) systems are widely used in wireless communications [1], [2]. Compared with single-carrier techniques, multi-carrier techniques have higher spectrum efficiency, and are more resistant to deep fading effect. However, the peak-to-average power ratio (PAPR) is an inherent problem in the multi-carrier system [3]. The design of high power amplifier (HPA) in the transmitter can reduce the PAPR at the expense of lower power efficiency [3], [4].