Research On Lossless Compression Of Fullpulse Radar Data

In today’s era of big data, data compression research and development has become more sophisticated, and applied in all aspects of life. The development of electronic surveillance and radar system has led to a increasing amount of the fullpulse data, and the common compression algorithm can not be good for the fullpulse data, therefore, a research on the fullpulse data lossless compression algorithm is imperative.The author designed a clustering compression system based on clustering analysis, which can fulfill the lossless compression of fullpulse data.This paper studies several key technologies of cluster compression system: Firstly, analyze some of the representative clustering algorithm to test and compare the results, choose K-means clutering because of its high-speed and flexibility; Secondly, encoding the data after the clustering process, analysis and comparison of commonly used lossless compression encoding, including the compression method for numerical and text, choose range of coding; then, combine the clustering and coding forming a system, and use some method to deal with the defect of K-means algorithm; finally, try to conduct the entire K-means algorithm parallel computing on GPU based on data block, and obtain a good results.The characteristics of this paper are as follows:(1) The data structure is relatively new, full pulse data is a new type of radar data format, each data object has five parameters, time of arrival(TOA), pulse width(PW), amplitude(PA), carry frequency(CF) and direction of arrival(DOA). The data comes from the UAV reconnaissance, some of the data pulse will correspond to the same goal. Thus between these data there is a strong correlation, we will take advantage of this correlation to accomplish lossless compression of fullpulse data.(2) Clustering algorithm is mainly used for signal sorting areas, the main function is to identify the internal data structure, and divide data collection into different classes or object clusters. In this paper, the clustering algorithm is used in the field of data compression, the strong correlation in a cluster is performed by the difference between the center value and the rest of the data points, use this difference for further process. And this research expand the application of clustering algorithm.(3) Try K-means clustering on the GPGPU parallel computing. Heterogeneous computing platform is one of the hot topic in recent years, GPGPU is heterogeneous platforms with logic control functions of CPU and powerful floating-point computing power of GPU. In this paper, we achieved good results for K-means clustering parallelization.This designed clustering-compression system has been tested and gained a good result for fullpulse data. For a 152 MB of fullpulse data compression rate can reach 48.1%; at the same time, when the input data size is 2000 KB, parallel K-means clustering is faster than serial for 1.65 times.