基于压缩感知的重构算法与语音压缩分析-reconstruction algorithm and speech compression analysis based on compressive sensing.docxVIP

AbstractThe conventional Nyquist sampling theorem states that signal can be reconstructed without distortion only when the sampling frequency is greater than or equal to two times the highest frequency of the signal. With the improvement of living standards and the development of information technology, the amount and quality of information block increasing, resulting in the signal sampling rate and processing speed are getting higher and higher. The theory of compressed sensing proposed in recent years becomes a hot idea in the field of signal processing. It completes signal compression and sampling at the same time so that the compressed sampling rate can be much less than the Nyquist sampling rate. It makes use of the sparsity of the signal to sample at lower rate and reconstruct the signal almost without distortion, which greatly reducesthe signal sampling rate，data storage and transmission costs. The new sampling method beginsto break the traditional sampling “bottleneck”.The compressed sensing theory and traditional recovery algorithm is researched. An improved regularized Newton algorithm is proposed .The experiments shows that reconstruction possibility and the convergence speed is better than other similar algorithms.In order to compress the speech based on compressed sensing, the character of speech is studied at first. Simulation results demonstrate that speech signal is sparse in DCT domain, so an adaptive compressed method is proposed and experiments results shows that the performance of recovered speech based on the method above has good compression rate and reconstructional possibility. This Thesis also compares the traditional DCT compression method and the CS compression, the natural noise robustness is also analyzed at the same time.It is necessary to encode the speech before transmission. Two kinds of traditional encoding method are discussed in this thesis: Plus Coding Modulation and Linear Prediction Coding. Some experiments based on others ide