系统工程与电子技术 ›› 2022, Vol. 44 ›› Issue (1): 70-75.doi: 10.12305/j.issn.1001-506X.2022.01.10

• 传感器与信号处理 • 上一篇    下一篇

基于深度学习的捷变相参雷达1-Bit块稀疏重构

付蓉, 黄天耀*, 刘一民   

  1. 清华大学电子工程系, 北京 100084
  • 收稿日期:2020-12-31 出版日期:2022-01-01 发布日期:2022-01-19
  • 通讯作者: 黄天耀
  • 作者简介:付蓉(1994—), 女, 博士研究生, 主要研究方向为雷达信号处理、压缩感知、深度学习|黄天耀(1989—), 男, 助理研究员, 博士, 主要研究方向为雷达信号处理、波形设计和压缩感知|刘一民(1983—), 男, 副教授, 博士, 主要研究方向为雷达理论、统计信号处理、压缩感知及其在雷达中的应用、频谱感知和智能交通系统
  • 基金资助:
    国家自然科学基金(61801258)

DNN based 1-bit block sparse recovery in frequency agile coherent radar

Rong FU, Tianyao HUANG*, Yimin LIU   

  1. Department of Electronic Engineering, Tsinghua University, Beijing 100084, China
  • Received:2020-12-31 Online:2022-01-01 Published:2022-01-19
  • Contact: Tianyao HUANG

摘要:

近年来, 量化压缩感知理论在雷达目标参数估计问题中得到了广泛应用, 其主要思想是对采样回波数据进行量化, 并将雷达观测模型建模为欠定方程, 再利用压缩感知理论对稀疏目标信号进行恢复, 降低回波数据的位宽, 达到简化系统、提升效率的目的。本文建立了捷变相参雷达信号的块稀疏压缩感知模型, 并提出一种基于深度学习的1 Bit块稀疏重建网络B-BAdaLISTA, 该重建网络与传统1-Bit硬判决迭代算法比较, 在1 Bit采样量化下具有相近的重构性能和更快的收敛速度, 同时将块稀疏的结构特征融入到网络结构中, 显著提高了雷达目标参数的恢复质量。通过仿真实验定量分析了B-BAdaLISTA重建网络在无噪、有噪条件下的恢复能力, 验证了算法的有效性。

关键词: 捷变相参雷达, 块稀疏, 1-Bit量化, 深度学习

Abstract:

In recent years, the theory of quantized compressed sensing has received extensive attention and development in the problem of radar target parameter estimation. Its main idea is to quantify the sampled radar echo and model it as an underdetermined equation, then target signal recovery can be solved via quantized compressed sensing algorithms. As sampling data is quantized, the bit width is greatly reduced thus simplifying the system and improving efficiency. This paper formulates the parameter estimation problem for frequency agile coherent radars as an underestimated problem, and proposes a 1-Bit block-sparse reconstruction network based on deep learning, namely B-BAdaLISTA. Compared with the traditional binary iterative hard thresholding algorithm, this reconstruction network has similar reconstruction performance and faster recovery speed. At the same time, the block-sparse structure is integrated into the network structure, which greatly improves the quality of the recovery of target parameters. The simulation experiments verify the recovery performance of the proposed B-BAdaLISTA network under both noiseless and noisy cases.

Key words: frequency agile coherent radar, block sparse, 1-Bit quantization, deep learning

中图分类号: