Multi-platform Parallel Acceleration Of SAR Imaging

Synthetic Aperture Radar（SAR）has many characteristics such as long-range action,strong penetration and multi-polarization,and has been widely used in various fields such as military,agriculture and remote sensing^[4],and has high research value.At present,SAR technology is becoming more and more mature,and the pursuit of high resolution in SAR imaging means that the amount of SAR echo signal data to be processed is also growing rapidly,especially in scenarios requiring high real-time performance such as assisted guidance,and the performance requirements of SAR signal processing systems are also increasing.Therefore,in addition to the optimization of SAR imaging algorithms to reduce the time complexity,the research on hardware-based parallelization and acceleration of SAR imaging is also of great significance.In current practical engineering applications,most of the cases use Central Processing Unit（CPU）as the platform for SAR signal processing,which is not only inefficient but also costly.In contrast,the Graphic Processing Unit（GPU）platform has excellent performance in SAR data processing due to its excellent floating-point operation performance.The Digital Signal Processor（DSP）platform has been widely used in digital signal processing field.However,in recent years,due to the limitations of manufacturing process,power consumption,and integration in the semiconductor field,the chip performance has encountered a bottleneck,and the architecture of single computing node will not be able to meet the requirements of certain applications.In this paper,we mainly focus on CPU and GPU for multi-node distributed parallel computing clusters,and use the FT-M3000 high-performance dedicated DSP chip to realize the acceleration scheme for real-time imaging of SAR echo data.The specific work mainly includes the following aspects.First,on the basis of in-depth analysis of frequency domain SAR imaging and time domain SAR imaging algorithms,their time complexity and parallelizable schemes are analyzed,and a single-node multi-core CPU and GPU platform is used to parallelize and accelerate SAR imaging,respectively.On this basis,a high-performance distributed multi-node parallel computing cluster framework with master-slave architecture+load balancing is designed,which includes a master node and several slave nodes,with the master node responsible for logic control,task allocation,load balancing,etc.and the slave nodes mainly responsible for computation.In this framework implementation,each compute node is implemented with GPU and parallel acceleration,thus realizing a dual parallel acceleration system with distributed memory as the first level acceleration model and shared memory as the second level acceleration model.Secondly,the acceleration of Fast Fourier Transform（FFT）and Inverse Fast Fourier Transform（IFFT）based on the FT-M3000 high-performance dedicated DSP chip was studied.Tests have shown that the FFT/IFFT operations take up about 90%of the imaging time in the frequency domain imaging process of the SAR imaging,so in addition to parallelizing the data,this paper also combines the resource characteristics of the FT-M3000 DSP to optimize the acceleration of the FFT/IFFT.The final results show that the acceleration effect of the proposed scheme in this paper is obvious,and the processing speed is greatly improved compared with the conventional FFT/IFFT implementation.Third,the acceleration of SAR signal processing based on the FT-M3000 high-performance dedicated DSP chip is studied.First,the system architecture and hardware resources of the FT-M3000 DSP chip are briefly introduced.Secondly,the hardware resource characteristics of FT-M3000 are combined with the parallelizable characteristics of SAR imaging algorithms to analyze and explore a suitable parallelization acceleration scheme.Finally,the parallelized acceleration system of SAR imaging based on FT-M3000 DSP is implemented and test work is carried out.Through the comparative investigation of the experimental results,the proposed parallel accelerated imaging scheme is proved to be correct in terms of imaging results and has a significant acceleration effect,which can meet the requirements for imaging speed and real-time in certain application scenarios.The research done in this paper is oriented to engineering applications,simulating actual application scenarios,and testing and verifying the proposed scheme using a large amount of data,and the results show that the work done in this paper is of high practical significance for the research of SAR imaging acceleration methods.