Based On Blind Source Separation Is Close To Rail Train Voice Warning Signal

Train approach warning is an important part of the field of railway safety. It is also an important indicator of safety operation and the safety of railway construction workers, so to be able to timely and accurate identification train close to become very important. Therefore, this subject is based on the train rails of the waist to rail voice train close to warning research, which can effectively extend the distance of the warning, improve security and accuracy.The sound signal to train the key problem is close to the early warning of effective separation voice out of the train. Because of the acquisition of the train and the source signal source quantity is unknown. In such circumstances, the voice signal collection to train as research object, take blind source separation technology to the signal for research. Mainly discuss:improve second-order statistics blind source separation method and multi-source signal blind signal extraction algorithm.Through analysis of existing second-order statistics of the blind source separation algorithm, based on the actual train approach warning an improved second-order statistics blind source separation algorithm, that with different delay the covariance matrix of the average delay instead of joint matrix and get mixed matrix of average characteristic structure, thus it greatly simplifies the algorithm, At the same time in the process of the diagonalization of unit delay variance of matrix into albino calculation, reduce the noise on the influence of the separation results.In the source signal more circumstances, in order to speed up the separation speed, we discussed trimmers number extraction algorithm. It is a signal from the observation extract a source signal, the extraction main point two steps:at first, according to the statistic characteristics of signal to construct a extraction signal standard, accordance with the standard gradient method to extract a signal; then is a reduced order process, which namely the removal of the extracted source signals from observed signals, making the observed signal is no longer included in the newly extracted signal. Cycle until two steps to extract the signal that we need, this algorithm is simple, fast, separation, high efficiency.