Fanaticism, Separation Algorithm Based On Time-frequency Analysis Is Studied

Blind signal separation is also called as blind source separation (BSS, Blind Signal/Source Separation), refers to the case of using information only from observed mixingsignals to estimate the source signal. Blind signal mixing can be divided into linearinstantaneous mixing, convolutive mixing，and non-linear mixing. Or it can be dividedinto over-determinned, exact-determined and under-determined blind signal separationaccording to the number of source signals and observed signals. Nonlinear andunder-determined situation is more complicated. Convolutive mixing takes into accountthe multipath and time delay and is closer to the actual communication systems, so it isone of the hot spots of blind signal separation.Frequency hopping is widely used in military and civilian communication systems,for its good performance of anti-jamming, LPI (Low Probability of Intercept) andnetworking capability. For Frequency hopping communication systems, the estimationof the hopping signal parameters is very important.Based on the above background and using time-frequency analysis method, thedissertation proposes a sorting algorithm for convolutive blind speech signals separationbased on third-order cumulant. Methods of blind signal separation and parameterestimation for the frequency hopping communication systems are presented based ontime-frequency analysis. This dissertation is divided into six chapters, and chapter III tochapter IV are the main work. The structure is organized as follows.The first chapter is an introduction, and introduces the research background andsignificance of convolutive blind signal separation. In addition, the development historyand application of frequency hopping communications are slao introduced briefly.The second chapter is the basic theory, and briefly analyzes the principle ofindependent component analysis and time-frequency analysis.Chapter three studies convolutive blind source separation for speech signals usingshort-time Fourier transform to change convolution of time domain into the product offrequency domain, so that linear instantaneous blind signal separation is performed ineach sub-band. Compared with traditional time-domain algorithms, the new method greatly reduces the computational complexity, but this will lead to uncertainty of theseparation order and magnitude. This chapter proposes a sorting algorithm in frequencydomain based on the third-order cumulant, and calculates three-order cumulant in eachsub-band. This can solve the disorder problems of the separated signals, and simulationresults show that the algorithm is feasible.Chapter four concerns frequency hopping communication systems. As most of theexisting blind source separation algorithms are fit for traditional speech signals, and fewresearchers study blind signal separation in frequency hopping communication systems.Combining with sparsity of frequency hopping signals in time and frequency domain, anew blind signal separation algorithm for FH communication systems is proposed.Compared with the traditional joint diagonalization algorithm, the new algorithmachieves better performance.Chapter five considers a single frequency-hopping signal, and proposes anestimation method to further improve the estimation accuracy of frequency hoppingsignal parameters based on the smoothed Winger-Ville Distribution. After estimating thehopping cycle and the transition moment, the new algorithm combines thecharacteristics of frequency hopping signals, and change the window length of ofsmoothed Winger-Ville Distribution. The simulation results show that the new algorithmhas better estimation performance, but increases the complexity to a certain degree.Chapter six is a summary of the whole dissertation and the future researchdirections are pointed out.