A Model-Driven DL Algorithm for PAPR Reduction in OFDM System

Deep learning (DL) has dramatically improved the peak-to-average power ratio (PAPR) performance. However, the high computational complexity and excessive training data constitute a significant hurdle. In this letter, a model-driven deep learning algorithm is proposed for PAPR reduction in orthogonal frequency division multiplexing (OFDM) system. Precisely, an iterative peak-canceling signal generation scheme is unfolded as a layer structure of the DL network. The scheme falls into the category of tone reservation technique. A set of trainable parameters, which optimizes the clipping threshold and weights time-domain kernel function, has been designed and introduced into the iterative scheme. Compared with the existing approaches, the simulation results demonstrate that the proposed algorithm achieves comparable PAPR performance with low complexity and training costs.

I. Introduction

Orthogonal frequency division multiplexing (OFDM) technique has been widely deployed in high-speed data transmission systems, primarily for its good resistance to multipath fading and high spectral efficiency. However, the inherent high peak-to-average power ratio (PAPR) problem is still one main defect. Various PAPR reduction schemes have been proposed [1]–[4]. Nevertheless, reducing complexity has always been a tricky task.