Underdetermined aliasing matrix estimation method based on neural network and genetic algorithm

Abstract

The invention discloses an underdetermined aliasing matrix estimation method based on a neural network and a genetic algorithm. The method comprises the following steps: carrying out short-time Fourier transform on observation signals; calculating a tangent value of each observation signal obtained after transform at each moment; constructing a competitive neural network, and with the tangent values being as input, carrying out preliminary clustering on the input tangent values; carrying out optimization on the preliminary clustering result through the genetic algorithm; and according to the optimized clustering result, calculating an estimated value of an aliasing matrix through arctangent. The method can carry out accurate separation on underdetermined blind source signals under the conditions of low-dimensional observation and unstable mixed signals; by converting the high-dimensional solution parameters into the one-dimensional tangent value, computation complexity is reduced; and through clustering analysis with the neural network and the genetic algorithm being combined, the clustering effect is more obvious, and solution accuracy of the aliasing matrix is improved.

Description

technical field [0001] The invention relates to an underdetermined aliasing matrix estimation method based on a neural network and a genetic algorithm, and belongs to the technical field of blind signal processing. Background technique [0002] Blind source separation (BSS) is an important modern multimedia technology, a powerful tool for data analysis and signal processing, and has a wide range of applications in many fields, such as speech recognition, image recognition, data mining, wireless communication and biology Wait. In the actual environment, the speech signal will be interfered by noise or other speech, so the blind source separation technology of speech signal has been widely concerned as an effective signal separation technology. Blind source separation of speech signals refers to the process of separating each sound source signal from the observed speech signal when the theoretical model of the signal and the sound source signal cannot be accurately known. In...

AI Technical Summary

Problems solved by technology

However, due to the non-stationarity of speech signals and the small number of observations, how to improve the separation effect of multi-channel mixed speech signals is still a subject that needs to be studied.

Images