Research On Quantization Compressive Sensing

Based on the sparse feature and compressibility of signals, compressive sensing (CS) isproposed to accurately reconstruct the unscanned sparse signal with few sampling data, inwhich the sampling rate is much smaller than the Nyquist sampling theory. Since this theoryis proposed, it draws widely attention in academics and industries.In recent years, the research about CS is mainly focused on the sparsity of signals, theconstruction of the unrelated observation matrix, and the signal reconstruction algorithm. Thequantization of CS, which can be used to enhance the compression ratio, has got less attention.Trellis coded quantization (TCQ) is an efficient method for realizing signal quantization. TCQuses the convolution coding and signal space expansion to increase the Euclidean distancebetween signals, and uses Viterbi algorithm to search the best path. This paper is dedicated tothe research of quantization in compressed sensing based on TCQ. The main contents of thispaper are summarized as follows:The combination of TCQ and CS theory is detailed in this paper. The TCQ is used toquantify the sampling signals in CS. Based on the CS theory, the sampling observation isobtained by using the Gaussian random observation matrix at the encoder. The samplingobservation is quantified with TCQ and then is transferred to the decoder. The decoderreconstructs the signal with orthogonal matching pursuit algorithm.The TCQ is used to quantify the sparse signals obtained by CS at the encoder and then isused to quantify inversely at the decoder. The simulation results using different images showthat the implementation with TCQ increases the peak signal to noise ratio (PSNR) by3.75%than the implementation without TCQ, and decreases the running time by15.74%. Theexperimental results testing the same image under different compression rates show theimplementation with TCQ increases the PSNR by1.34%and decreases the time by14.6%.The simulation results show the TCQ could improve the compression performance and shortthe running time at the same time.