Research On Analysis And Recognition For Multicomponoent Signal In Time-Frequency Domain

With the rapid development and wide application of the different devices over the past decades, a lot of new problems for the more popular case of many signals in the same channel have been presented on signal processing since more and more electromagnetic signals with different types and forms have been collected. The analysis and recognition problems for multicomponent signals in time and frequency domain, which is an extremely underdetermined problem, have been studied by more and more researchers because of its wide application background and great practical significance. Based on different types of signals, including the natural signals such as biomedical signals, mechanical vibration signals and audio signals, etc, and the artificial signals such as wireless communication signals and radar, etc, the researchers define the deterministic model, non-stationary model, nonlinear time series model and state space model, and propose various methods like time-frequency transform, matrix factorization and parameter estimation reconstruction. In the end, all of these studies have been successfully applied in specific applications, while they failed in the more complex situations or applications. Based on the previous research results on the analysis and recognition problem for multicomponent signals in time and frequency domain, we provide further research on this topic and get some new ideas and algorithms in this thesis. The main contributions and contents are summarized as follows.Firstly, the blind analysis and recognition problem for nonstationary multicomponent linear-frequency-modulated(LFM) signals is studied, and a new method based on the fractional Fourier spectral kurtosis(FRFSK) is proposed for the problem of weaker component identification failures under the condition of lower SNR. Circularity of the fractional Fourier transform(FRFT) is discussed in the proposed method. Then, based on the spectral kurtosis(SK) expressed in the fractional Fourier domain, the definition and characteristic of fractional Fourier spectral kurtosis(FRFSK) is proposed to detect the multicomponent LFM signals. Theoretical analysis and simulation result corroborate that the proposed method performs better than the other methods under lower SNR and weaker components with interference by strong signals.Secondly, the blind analysis and recognition problem for nonstationary multicomponent NLFM signals is studied, and a new ASTFRFT method based on spectral kurtosis is proposed for the problem of closely placed signal components identification failures. Based on the FRFSK weigthed by a window, the new definition of STFRFT is proposed. The spectrum proability density and non-circularity of the rotated Guassian window are derived to rectify ASTFRFT. Based on the circularity of the rectified ASTFRFT, FRFSK is applied to select the parameters of the optimum window. Then, the ASTFRFT based on FRFSK is proposed. Simulations corroborate that the proposed method performs better than the other methods for distinguishing the closed NLFM signals. In addition, it has shown that the FRFSK of a rectified ASTFRFT coefficient for Gaussian noise equals 0. This property has been utilized to detect the signal drowned in the noise based on time-frequency segmentation. The effectiveness of this method is evaluated via simulations.Thirdly, the blind analysis and recognition problem for practical signals, in which most of them are nonlinear, nonstationary and even chaotic in many fields, is studied, and a new method based on high order statistics-single channel ICA(HOS-SCICA) is proposed for the problem of nonstational and nonlinear time series identification failures. To solve this problem, a three-step method is provided in this paper. In the first step, the measured signal which is assumed to be piecewise higher order stationary is introduced and divided into a series of higher order stationary segments by applying a modified BG segmentation algorithm. Then, by the appropriate parameters selected, the state space is reconstructed effectively and the segment is transformed into a pseudo multiple input multiple output(MIMO) mode, which is modeled as multipath ILM using a method of coordinate transformations based on the high order statistics(HOS). In the last step, ICA is performed on the pseudo-MIMO data to decompose the single channel recording into its underlying independent components(ICs) and the interested ICs are then extracted. The effectiveness of the HOS-SCICA method is validated with a measured data throughout experiments in the end. Also, the proposed method is proved to be more robust than SCICA under different SNR and/or embedding dimension via explicit formulae and simulations.Finally, the blind separation and recognition problem of multiple digital phase modulation signals received by a single sensor is studied, and a new BSS method based on HOS-singular value decomposition(HOS-SVD) is proposed for the problem of multiple digital phase modulation signals identification failures under the condition of different rates. Based on the different rates and waveforms of communication signals, the problem can be transformed into the blind separation of period signals with phase changing by utilizing oversampling. Then, the reference waveforms of signal sources are estimated by HOS-SVD. The independence of the real and image parts of communication signals is analyzed and the lists of the optimum rotating angles for various modulations are set up. Based on the analyses, the symbol sequences are directly estimated by applying ICA to the matrix, which is a projection of the mixed signals to a certain subspace. In the end, the symbol sequences of the sources are obtained by removing the redundant phases and rotating the optimal angle. Simulations show that the proposed method can solve the blind recognition problem of multiple digital phase modulation signals and is robust to the impact of noise and non-equal power in some degree.