Frequency domain convolution blind source separation multi-band and multi-centroid clustering sorting method

Abstract

The present invention discloses a frequency domain convolution blind source separation multi-band and multi-centroid clustering sorting method. The method comprises the steps of: (1) performing overlapping grouping of full-band signals, wherein the whole frequency band is uniformly divided into a plurality of sub bands, and overlapping frequency bands are arranged between the sub bands; (2) in each sub band, performing multi-centroid clustering and single-centroid clustering in order; (3) finding out frequency bands with failed sorting, performing resorting of the band signals to maximize thesum of coefficients of association of a separative signal amplitude envelope and a clustering center at the frequency band; and (4) traversing all the sub bands in order to complete sorting of the full-band signals. The frequency domain convolution blind source separation multi-band and multi-centroid clustering sorting method employs an overlapping sub-band multi-centroid clustering method to improve the clustering center, namely the precision of reference signals, reduce the probability of wrong arrangement among the sub bands. The simulation experiment shows that the method is good in accuracy, and can keep a good separation performance for the change of the short-time Fourier transform window in a certain range.

Description

technical field [0001] The invention belongs to the field of mechanical vibration signal and acoustic radiation signal processing, in particular to a frequency-domain convolution blind source separation sub-frequency band multi-centroid clustering and sorting method. Background technique [0002] In a complex mechanical system and multi-source coupled sound field environment, the signals measured by the sensors installed on the mechanical system are often the result of the complex mixing process of each vibration source signal, causing multiple excitation sources to interfere with each other, which gives vibration The identification of noise sources presents difficulties. [0003] Blind source separation is a method of estimating the original signal from the observed mixed signal when the source signal and the mixing process are unknown, which provides a good solution for the separation and identification of noise sources. Among them, instantaneous mixing is the simplest mo...

AI Technical Summary

Problems solved by technology

However, although the time series between signal frequency bands has a high correlation, as the inter-frequency span increases, this correlation will become worse, and even the time series correlation of the same source is smaller than that of different sources Case

Therefore, in some cases it is difficult to find a global reference to make the ordering consistent across the entire frequency band, and as a result, ordering errors still occur in some frequency bands, or even the entire frequency band

Images