Realization Of False Alarm And Target Cohesion Algorithm Based On TMS320C6678

CFAR and target cohesion are both key algorithms in radar signal processing to improve the performance of detection and accuracy. Modern radar signal processing systems are generally based on FPGA+DSP architecture. In real-time systems, DSP must equipped with powerful computing capability and large data transmission bandwidth. Combined with "comprehensive XXX radar" project, a plan to realize CFAR and target cohesion algorithms on TMS320C6678 is proposed.First this paper analyzes the basic principle.of the GO_CFAR, the clutter map CFAR, the traditional target cohesion and the connected domain target cohesion based on the contour tracking of binary image. The simulation of each algorithm module provides theoretical support for the realization on hardware. Then we resrach the three interfaces technology of TMS320C6678, namely, SRIO, GPIO and Gigabit Ethernet. We design the software of each interface and debug them. The debug of the three interfaces verifies the correctness and reliability of transmission, so the problem of data transmission between DSP and FPGA or host computer is solved. The next we proposes an implementation of three channel CFAR and target cohesion algorithms due to four cores:core 0 is used to receive and send data, and core 1～core 3 are used to run CFAR and target cohesion of every channel. The CFAR algorithm is optimized by linear assembly, the data transpose is optimized by EDMA3, and the connected domain target cohesion is adopted. Because EDMA3 transpose can run without CPU intervention, it can operate the next row with CFAR while transposing, which greatly reduces the time of CFAR processing. Connected domain target cohesion can not only solve the spurious targets problem of traditional cohesion, but also has a faster realization because it scans the matrix after CFAR only once. The inter-core communication and the cache consistency are also designed. Final the paper discusses the commissioning of each algorithm module on TMS320C6678. Comparing to the results of MATLAB, the correctness of algorithms is verified. We analyze the resources and time of the whole algorithms, the total processing can be finished in 198ms, less than a CPI 215ms, so we can conclude that the real time property of the algorithm implementation is attained.