Research On Multi-core Parallel Optimization Of Bottom Database Based On Missile-borne Synthetic Aperture Radar

Missile-borne synthetic aperture radar real-time signal processing has always been the hot research topic in the area of national defense among all countries around the world.The development of the imaging algorithm in missile and the transplantation of the software code to signal processing system directly influence the accuracy of image matching and positioning after imaging process and real-time performance of applications on missile,and subsequently influence the guidance accuracy of the precision guided weapons.The previous high resolution missile-borne synthetic aperture radar imaging algorithms have some problems such as large computational complexity and hard to realize in engineering application,which are not suitable for the condition on missile that has a high requirement on real-time performance.Based on above background,combined with a project cooperation with a research institution,this thesis presents some improvements towards the poor real-time performance of the previous imaging algorithms in the following aspects.1.This thesis presents a big-squint-high-resolution sub-aperture frequency domain filtering approach which is suitable for high-speed moving platforms.To deal with the problems of high moving speed of the platform and the high requirement of the real-time performance,this algorithm uses sub-aperture data instead of the full aperture data to enhance the real-time performance with the cost of degrading the resolution properly.At the same time,this algorithm uses the frequency domain phase filtering method to achieve the goal of widely imaging in large azimuth scene during missile's decent trajectory.2.With regard to the complexity of signal processing system and the difficulty of task synchronization and distribution in signal processing chips,this thesis makes full use of hardware resource of the signal processing system.Between the different cores of the multicore DSP,we choose notify and Inter-Processor communication(IPC)interrupt based on operating system as synchronous mechanism and core Pac communication measure individually.Between different DSPs in a single board and different boards,we use effective data flow synchronous structure to arrange algorithm modules into different processing units reasonably.When the pipeline flows,we can get parallel processing mode effectively.3.In order to deal with the problem of inefficient code transplantation,we figure out a lot of methods from top to bottom to optimize deeply.For example,the algorithm level,the frequent-signal-processing-module level,and the code level.In the bottom level,we simulate a large amount of experiments in terms of compiler,assembly keywords,inline functions and liner assembly language.After comparing the different results statistically,we get the real-time performance of different optimizing methods.Finally we improve the bottom database of imaging algorithm by forementioned approaches,analyze the performance of the updated imaging algorithm and process the real data form field tests.The results show the correctness and validity of our proposed methods.