Design,Implementation,Verification And Optimization Of DSP Coprocessor Based On FPGA Platform

In the various control and graphics image processing algorithms involved in aerospace systems,there are a large number of transcendental functions and matrix multiplication calculations,and their computing performance often plays a decisive role in the performance of upper-layer applications.Field Programmable Gate Array(FPGA)has reconfigurable flexibility and high energy efficiency ratio,and has unique advantages in the field of accelerating transcendental functions and matrix multiplication.Aiming at the computing requirements of transcendental functions and matrix multiplication in aerospace control system,this topic designs and implements a dedicated coprocessor based on the DSP-FPGA heterogeneous information processing platform,and conducts a comprehensive verification and optimization of it.The main work of this paper is as follows:1.Based on the acceleration requirements for transcendental functions and matrix multiplication,this paper designs and implements a coprocessor structure with high-lowspeed separation and multi-stage pipeline buffering.The coprocessor is mainly composed of two parts: transcendental function acceleration subsystem and matrix multiplication acceleration subsystem group,and has the characteristics of efficient concurrent and highthroughput computing.2.The verification and synthesis of the DSP coprocessor based on the FPGA platform are completed.In this paper,a detailed module-level and system-level verification of the designed coprocessor is carried out,and its comprehensive results and performance are analyzed.Among them,the module-level verification mainly verifies each function calculation module of the transcendental function acceleration unit and the matrix multiplication acceleration unit module.System-level verification verifies the function of the entire coprocessor system based on the coprocessor deployment platform.The experimental results show that the transcendental function calculation in the coprocessor obtains a speedup of 10.4-31.1 times,and the matrix multiplication calculation obtains a speedup of 127 times.3.The optimization of DSP coprocessor based on FPGA platform is completed.The operation process of transcendental function and matrix multiplication in the coprocessor is deeply analyzed,and the structure of the coprocessor is further optimized for the performance bottleneck.Finally,for specific application scenarios,from modules to the whole,concurrency optimization strategies to improve throughput are proposed.