Efficient Implementation Of SAR Imaging Algorithm Based On TX2i

With the increasing maturity of synthetic aperture radar technology,it has high practical value and a wide range of application scenarios in national defense and civil fields.In recent years,SAR is developing in the direction of ultra-high resolution,multi-polarization,multi-band and multi-mode coordination.At the same time,in order to be able to produce SAR images with both high quality and very high resolution,the design of SAR imaging algorithms begin to become more complex,and the corresponding algorithm processing modules are also increasing.In addition,in order to be able to study the characteristics of a specific area scene more comprehensively,the SAR imaging system needs to acquire more abundant echo signal data.The complex algorithm design and massive echo data make it difficult for traditional computing platforms to meet the real-time requirements of imaging,while the traditional plug-in GPUs consume too much power and it is very difficult to achieve on-board processing.The industrial-grade TX2 i module of NVIDIA’s Jetson series has the advantages of low power consumption,high performance,ease of development and integration,which make it very suitable for accelerating airborne SAR real-time imaging processing.Therefore,this paper proposes an efficient implementation method of SAR imaging algorithm based on TX2 i platform,and significantly reduces the requirement of GPU memory for the algorithm.Aiming at the fact that the real flight trajectory of airborne synthetic aperture radar is very complex,traditional SAR imaging algorithms are often difficult to produce high-quality focused images.This paper studies the range-Doppler algorithm in stripe mode and the polar format algorithm in spotlight mode,and according to the characteristics of the actual flight trajectory of SAR in different working modes,an improved design of the corresponding imaging algorithm is carried out.Aiming at the difficulty of implementing SAR imaging algorithm on TX2 i platform,this paper studies the principle of GPU parallel programming,and then some key points of CUDA program performance optimization are summarized on this basis.And based on the radar signal processing hardware platform used by a research institute,a real-time signal processing software with Tegra X2 architecture as the core processing unit is designed.At the same time,considering the characteristics of GPU software programming and the characteristics of the TX2 i hardware platform,the strip SAR imaging function with a resolution of 0.3m×0.3m and the spotlight SAR imaging function with a resolution of0.05m×0.05 m were successfully realized.In terms of program design,in order to solve the problem of real-time imaging,shared memory,texture and surface cache,registers and other storage resources are used reasonably,and use the advantages of GPU multi-threading and multi-computing cores to process echo data in parallel,and reasonably give each the kernel function allocates grid and thread block sizes to make full use of GPU resources,thereby improving program execution performance;In order to solve the problem of insufficient storage resources of the embedded GPU,this paper implements the fast Fourier transform and matrix in-situ transposition that are frequently used in the algorithm,which greatly saves the memory space,and further reduces the execution time of the program by integrating the operations before and after the FFT algorithm.A series of methods such as memory reuse and in-situ writing are used to reduce the use of storage space,and the problems of time-consuming data transmission and memory occupation are solved by means of unified memory and zero-copy technology,so that the limited GPU resources on the TX2 i can be more fully utilized.