微波成像稀疏信号重构的FPGA加速方法研究

Microwave imaging has a great demand and widely used in military and civilian fields. Sparse signal reconstruction theory provides a new technical approach to the high-resolution microwave imaging. With the constant improvement of the reconstruction algorithm, the robust real-time implementation of the algorithm is an urgent problem to solve. Compared with the traditional signal processing algorithms, the computational complexity of sparse reconstruction algorithm is increased dramatically, which bring great difficulties to real-time processing. To solve this problem, this project carries out FPGA acceleration research on sparse signal reconstruction for microwave imaging, the main contents include: 1) the storage and computing model of large dictionary matrix and pipelined parallel structure; 2) the computing strategy selection method and circuit implementation of the adaptive signal reconstruction combining prior information mining; 3) multiplexing circuit design for a variety of algorithms based on a unified framework. We strive to achieve a breakthrough on hardware structure mapping theory and FPGA acceleration of sparse optimization algorithm for microwave imaging.This project would lay the theoretical and technical foundation for VLSI circuit design for sparse signal reconstruction and promote the application of the sparse optimization algorithms in microwave imaging system.

微波成像在军用和民用领域有着重大需求和广泛应用。稀疏信号重构理论给高分辨微波成像提供了新的技术途径。随着重构算法的不断完善，算法稳健的实时实现成为迫切需要解决的问题。和传统的微波成像信号处理算法相比，稀疏重构算法计算复杂度急剧增加，这给实时处理带来了极大的困难。针对该问题，本课题开展微波成像稀疏信号重构的FPGA加速研究，主要研究内容：1）大字典矩阵存储计算模型和全流水线并行结构；2）结合先验信息挖掘的自适应信号重构计算策略选择和电路实现方法；3）基于统一框架的混合算法复用电路设计。本课题力争在微波成像稀疏优化算法硬件结构映射理论和FPGA加速技术上有所创新，为稀疏信号重构的VLSI电路设计奠定一定的理论和技术基础，促进稀疏优化算法在微波成像系统中的应用。

本项目在微波成像稀疏重构理论、稀疏重构算法FPGA硬件加速方面取得一些成果。微波成像稀疏重构成果: ①提出了基于量子关联原理的相控阵雷达三维超分辨成像体制，为前视成像提供了新的技术途径。②建立前视关联成像模型，引入稀疏重构理论，对二维成像切片逐点计算成像，突破常规实孔径阵列雷达方位分辨力的限制，用更少的回波采样实现超分辨三维成像。③发明了利用短孔径有限数据实现高分辨逆合成孔径雷达成像的方法。将逆合成孔径雷达成像转换为利用正交基重构稀疏信号的问题，然后利用稀疏恢复技术进行目标像的高分辨优化重建。稀疏重构算法FPGA硬件加速成果: ①研制了微波成像稀疏重构硬件加速平台，为算法加速提供了基础。②研究了字典支撑集正交化以及基于共轭梯度实现最小二乘的计算方法，提升了稀疏重构速度。③基于字典先验信息，研究了基于快速傅里叶变换的加速重构电路设计方法，在简化稀疏基构造的同时降低了计算复杂度。研究成果通过了实测数据以及电路仿真综合验证，具有重要的科学意义和广泛的应用前景。