Research On Chaotic Masking And Blind Extraction Algorithm Of Speech Signal Under Multipath Channel

The diversified application of contemporary communication technology enabled people to obtain valuable information in a timely and multi-channel manner.As the main form of human communication,speech information is widely used for its strong real-time and easy recognition.On the one hand,while the speech signal is easy to identify and read during the transmission process of the communication system,it is also more prone to information leakage and malicious tampering.On the other hand,speech information was inevitably affected by multipath transmission effects during both sending and receiving,i.e.,speech was sent out and needed to reach the receiving end via multiple transmission paths,so that the observed signal was a convolutional mixture of multi-channel speech signals with attenuation,delay and phase effects.The mixing model of such signals was more complex,which made the separation and extraction of the source speech signal more difficult.Based on the above background,this paper studied the secure communication technology of speech information,adopted a chaotic system with noise-like characteristics as the transmission carrier of speech information,and used chaotic masking technology to process the speech information confidentially to ensure its secure communication in unknown channels.At the same time,the blind source separation technology is used to separate and extract complex convolutional mixed signals at the receiving end.Because the mixed speech signal adopted the time-domain convolutional method to separate and extract the source signal,the calculation amount was very large,the algorithm complexity was high,and the convergence speed was slow.Therefore,this article first transformed the time-domain convolutional signal into a frequency-domain product signal by short-time Fourier transform,and successfully changed the convolutional blind source separation into the instantaneous blind source separation at each frequency point in the frequency domain.And the several traditional convolutional blind separation algorithms was analyzed and compared.Although the frequency-domain convolutional blind separation algorithm avoided the tedious convolutional calculation process and improved the convergence speed,it suffered from the problem of uncertainty in the amplitude and ranking of each frequency component,resulting in inaccurate extraction of the source speech information.Using this as an entry point,this thesis proposed a cross-correlation permutation algorithm based on a spectral peak search counter by analyzing the energy correlation of adjacent frequency points.Simulation results show that the proposed algorithm can effectively improve the separation performance of blind separation algorithm with low complexity and computational complexity.