A Modified Source Number Estimation Method based on Wavelet Analysis and Singular Value Decomposition
Chuanchuan Wang, Yonghu Zeng, Liandong Wang · 2018
In underdetermined blind source separation (BSS) algorithms, source number estimation is a precondition and a difficulty. An improved source number estimation method based on wavelet analysis and singular value decomposition is pro-posed. First, the whitened observed signals are decomposed by wavelet analysis method. Next, the decomposed wavelet coefficients are reconstructed separately. Low frequency wavelet coefficients, the highest frequency wavelet coefficients, and original observed signals are chosen to form a new observed signals matrix. Then, covariance matrix of this new signal matrix is calculated, and eigenvalues of covariance matrix are calculated. Based on the obtained eigenvalues, a new calculation formula of judging threshold for source signals and noise is proposed. The proposed algorithm is validated by simulation.