Research On Ultra-high-order Convergence Blind Signal Separation Technology Based On FastICA Algorithm

Blind Signal Separation technology is to separate the source signals from received mixed signals when the source signal information can not be obtained directly and the mixing mode of the source signal can not be fully known.As a new signal processing technology developed rapidly in the past two decades,it has been gradually applied to many fields,such as speech signal processing,bio-simulation,information communication system,underwater acoustic signal processing,image recognition and processing,fluid motion data analysis,mechanical vibration signal,etc.It has very important practical value and has become one of the research hotspots in the field of signal and information processing.Firstly,the related theories of Blind Signal Separation(BSS)technology are systematically expounded,such as the hypothesis,prior knowledge,fuzzy characteristics and hybrid mathematical model of BSS technology.And the independent component analysis(ICA)technology,which has a wide range of applications,is emphatically discussed.In addition,taking ICA as an example,the selection of the two most important parts of BSS technology,objective function and optimization function,is discussed.Finally,two evaluation indexes of BSS technology are introduced as evaluation criteria of algorithm performance.Secondly,the pre-whitening FastICA algorithm based on negative entropy is studied.The Newton iteration formula used in the algorithm is modified,and the ultra-high-order convergent iteration formula of 21th-order convergence is obtained,and the convergence of 21th-order iteration formula is proved.The simulation results show that the 21th-order convergent ultra-high-order iteration formula has good convergence speed and convergence performance.Thirdly,the traditional negative-entropy-based pre-whitening FastICA algorithm is improved,and the ultra-high-order pre-whitening FastICA algorithm based on negative-entropy is obtained.The speech signal database is built,and the separation performance of the second,fifth,seventh and fifteenth order convergent FastICA algorithm and 21th-order ultra-high-order convergent improved FastICA algorithm are simulated by means of average iteration times,average separation time,average similarity coefficient matrix and average signal-to-noise ratio.The experimental results show that the pre-whitening ultra-high-order FastICA algorithm based on negative entropy proposed in this thesis is no longer easy to fall into the dilemma of local convergence or non-convergence;the average number of iterations is less,the separation speed is faster,the separation signal is closer to the source signal,and the separation performance is better.Finally,the content of this thesis is summarized,and points out that the ultra-high-order FastICA algorithm needs to be further studied.The future research prospects of blind signal separation technology are sorted out and analyzed,which provides direction for further blind signal processing and separation.