New Optimization Method Based on Neural Networks for Designing Radar Waveforms With Good Correlation Properties

Meng Xia,Shichuan Chen,Xiaoniu Yang

doi:10.1109/access.2021.3092006

Abstract

Owing to advances in the overall performance and anti-interception capability of radars, the designs of radar waveforms with good correlation properties have been a concern for researchers. In this paper, we propose a novel method based on convolutional neural networks (CNNs) for designing single or multiple unimodular sequences with good auto- and cross-correlation or weighted correlation properties. The framework of the neural networks for sequence optimization is constructed using group convolution and identity mapping, and three different loss functions are presented using different optimization objectives. To illustrate the performance of the proposed method, we present numerous examples, including the design of sequences with low autocorrelation sidelobes in a specified lag interval and a sequence set with good auto- and cross-correlation properties. Moreover, an analysis of the simulations shows that the sequences designed through our method demonstrate better correlation properties than classic algorithms.

Highlights

Sequences with good autocorrelation properties have important applications in wireless communication, radar, sonar, and other fields
THE convolutional neural networks (CNNs) MODEL Considering the successful application of deep learning in speech recognition, natural language processing, target tracking, and machine translation [29]–[33], we introduce deep neural networks into sequence design and propose a sequence optimization method based on deep learning
In this paper, we have presented a radar waveform optimization method based on CNNs

Summary

INTRODUCTION

Based on ResNeXt, we propose a novel sequence optimization method by leveraging group convolution and residual units. The input of the neural networks is phase data in a uniform distribution from 0 to 2π, and its length is equal to the length of a sequence to be optimized. According to the phase distribution of the sequences after optimization, the novel activation function applied in this paper is. We apply neural networks to phase-coded sequence optimization to obtain a sequence or sequence set with low peak sidelobe levels and cross-correlations. The back-propagation algorithm is employed in the neural networks to minimize the loss function, and the optimized phase data are obtained through multiple iterations. It should be noted that according to different optimization objectives, only one loss function is selected for the neural networks at a time

SIGNAL MODEL

NUMERICAL SIMULATION

Findings

CONCLUSION