Research On Intermediate Frequency Data Compression And Parallelization Based On Wavelet Transform

Data compression technology has become increasing mature in a wide range of applications, such as image, voice, seismic signal, etc. With the development of radar technology, the Intermediate Frequency(IF) data amount is ever increasing, thus it is necessary to realize the fast compression of IF data. In this thesis, the IF data fast compression algorithm based on lifting discrete wavelet transform(lifting DWT) is proposed, which combines the characteristics of IF data. The main contents of this thesis contains the selection of wavelet function and wavelet decomposition series, the research on quantitative approaches and encoding methods, the implementation of parallelism based on Graphics Processing Unit(GPU). In this thesis, firstly, the db4 wavelet is chosen to carry on the 2 level lifting DWT. Then, the wavelet coefficients are segment quantized. Finally, the quantized coefficients are first processed with run length encoding and range encoding. In the end, the parallelization research and implementation of wavelet transform and quantization in the algorithm are carried out. The main features of this thesis are as follows:(1) Data characteristics: The input IF data format is in binary format, which is difficult for compression and need to be read into the system in integer format. The processed IF data is visible and intuitive in the data image, in which the envelopes of useful information behaves like pulses. The useful information is stored in the pulse and the slots outside the pulses contain no useful information. Therefore, the processed IF data is convenient for compression. To obtain a greater compression ratio on this basis, it is necessary to transform the data, and the energy will be concentrated in a few coefficients.(2) Compression ratio characteristics: a) Compared with the direct compression of IF data, the compression based on wavelet transform can achieve a greater compression ratio with the same energy loss. With the increase of the energy loss, the advantage of IF data compression based on wavelet transform is more obvious. b) As for quantization method, according to the characteristics of wavelet coefficients, non-uniform quantization is adopted, and different compression ratios can be obtained by inputting different quantization parameters.(3) Compression speed characteristics: According to the characteristics of the IF data, this thesis puts forward the realization of data parallel division with the integrity guarantee of data. Then, the IF data compression based on wavelet transform using the cooperative CPU-GPU method is realized. Compared with the serial implementation, the compression speed of parallel implementation is improved when the data amount is huge.The IF data compression algorithm based on wavelet transform designed in this thesis has been tested. The results show that the proposed algorithm is beneficial for the IF data compression. The compression ratio is up to 20 at 5% energy loss, meanwhile, the parallel compression speed increases by 2.8 times than that of serial compression when the size of input file is 64 MB.