Research On Blind Source Seperation Of Communication Signal Based On Sparse Representation

Blind source separation is defined as restoring the source only by observing signal in the condition that the priori knowledge of source signals and transmission channel are rare or even unknown. In recent years, the research of blind source separation like naphthaleneand and it is widely used in wireless data communication, image processing, speech signal processing and biomedical signal processing et al. Therefore, Blind source separation become one of the important research direction in the field of information processing in modern life.Signal sparse representation theory is a new signal analysis method, its purpose is to obtain the essential characteristics of signal by using as few atoms as possible, and then get the concise representation of the signal. The study found that it is also good for revealing time-varying properties of non-stationary signal. Therefore, based on the investigation of the current research of blind source separation, this paper start with signal transformation and signal compression with the main line of sparse feature. First, considering the time and frequency domain aliasing of communication signal, we extend the cyclic spectrum analysis method to lower order domain in a narrow band pulse noise environment, and structure separation characteristic parameters through studying the spectral domain features; Secondly, through the contrast of mathmatic model of compression sensing and blind source separation, we build a cascade dictionary and realize signal sparse representation combining with orthogonal matching pursuit OMP algorithm, thus realize the blind source separation. In this paper, the main work includes the following:(1) Based on the classical model of blind source separation, first of all, this paper discusses the similarity of blind source separation and compression perception model on the basis of familiar with compressed sensing theory, and then puts forward the source signal recovery algorithm and the idea of over-complete dictionary into signal separation. Then, the article analysis and indict the random observation matrix and the deterministic matrix, learning dictionary and combined dictionary, and norm optimization algorithm, and finally, puts forward some problems to study.(2) Based on the cyclostationary theory, this paper discusses low-order cyclic spectrum theory in the environment of impulse noise. With covariant and low-order cyclic spectrum density as the basic theory, probes into the low-order FSM algorithm, which provides a new way for signal processing under pulse noise environment.(3) Jointing Fourier and wavelet structure combined dictionary, and combining the orthogonal matching pursuit algorithms, this paper discusses the signal sparse representation with combined dictionary dictionary. Compared to the traditional orthogonal complete basis dictionary, combined dictionary has a good adaptability, and has a more reliable express for signal, all these works lays a good foundation for the next step of blind source separation.