The Study Of The Blind Source Separation Based On TIFROM Method

More recently, a few other concepts for BSS which are based on independent component analysis (ICA) have also been considered. Especially, several methods based on a time–frequency (TF) analysis of the signals have been reported. Three main classes emerge from these TF-BSS methods. The first one directly results from classical BSS approaches, as it consists of TF adaptations of previously developed joint-diagonalisation methods, with subsequent modifications. The second class includes several methods based on ratios of TF transforms of the observed signals. The third class is based on TF correlation or TF coherence parameters. All these methods are restricted to linear instantaneous mixtures, which applies to time-delayed mixtures but sets other restrictive conditions as stated above.First of all, thesis introduces the investigation background, history and development of the BSS. We introduces a lot of theoretic correlated to classical BSS methods and this paper method. Then this paper explains several classical BSS methods as Maximum Likehood Estimation, Joint Diagonaliation Estimation, Independent Component Analysis, and TF method. In addition, thesis introduces Time-Frequency analysis knowledge, which include STFT, gabor expansion ,Wavelet transform and wigner-ville distribution. In this paper, we propose a novel TF-BSS method which is inspired by our linear instantaneous TIFROM method, but suited to more general mixtures, involving time shifts. We thus avoid the restriction of the DUET method concerning the sparsity of the sources in the TF domain. Through computing the ratio of mixed signals TF transformation in one TF window, this method can estimate the elements of the mixing matrix. The key is how to find the area where only one source occurs.The LI-TIFROM which is restricted to linear instantaneous mixtures. The TF transform of the signals considered in that approach is the Short-Time Fourier Transform (STFT).This method avoids the effect of Cross-Terms, and the efficiency of thealgorithm has been improved by FFT. Then we derives the radios of STFTs of the mixed signals, finds TF analysis zones where a source occurs alone. we eventually obtain an ordered list of single-source TF analysis zones which then allows us to estimate the columns of B,.we eventually left-multiply the vector of mixed signals by the inverse of the estimated mixing matrix B, in order to obtain the extracted source signals.We derives the radios of STFTs of the mixed signals to find the source analysis zones which is occur alone. We proposed an improved version of this detection stage where we also consider the inverse ratios. We therefore order the TF zones, according to increasing values of the minimum among the averaged variances of both ratios and inverse ratios. Than analysis the TF ratio relationship betweenone of the mixing signals and all the others, single source zones exist precisely.However, AD mixtures correspond to the specific case of LI mixtures when all time shifts are zero. We can find that the AD mixtures become the LI mixtures at all frequencies when we transform the AD mixtures by FFT. We were inspired by the TF BSS approach that we previously developed for LI mixtures. We introduce two extensions of this approach to AD mixtures, called"AD-TIFROM-CT"and"AD-TIFROM-CF", depending whether they only use Constant-Time analysis zones or also Constant-Frequency zones, select analysis zone properlyinstead of the whole TF domain. Phase unwrapping will be a straight line in single source zone. Resolving the phase unwrapping of the mixing signals TF transformation also using the minimal variance method can find the single source zone precisely. Therefore, this method resoles the time-delayed problem effectively. In AD-TIFROM-CF, we use Constant-Frequency analysis zones and Constant-Time analysis zones, thus we find the analysis zone two times, the efficiency will not excellent. In AD-TIFROM-CT, we only uses constant-time (CT) analysis zones. The method which estimate the columns of B are alike. Using the parameters, we eventually derive the extracted sources from the observations, this method resoles the time-delayed problem effectively.The time-frequency ratio method has many special feature compared with those classical ICA methods. Unlike classical ICA methods, this method intrinsically well-suited to non-stationary signals and set no restrictions on the Gaussianity of the sources. This paper give several computer simulations to compare these method in some complex condition include LI mixtures, AD mixtures, non-stationary signals, and higher number of sources. The algorithms include TIFROM, FastICA. Experiment results approve that the proposed method can solve the difficulties better than others.The TIFROM method is a new member in the BSS area, its ability of practical application is more important. BSS as a new technology have a wide application area and foreground. This paper discusses its capability with speech signal. We find that they are show different performance when operated with the different parameter values. The experiment results present that the method separated the speech signal successively.