An Adaptive VMD Method for Single-Channel Underdetermined Blind Source Separation Guided by K-Means Clustering

For single-channel underdetermined blind source separation, an adaptive VMD method guided by k-means clustering is proposed. The method exploits differences in higher-order cumulants to design a cyclic iterative strategy for adaptive optimization of the mode number K, and completes modal reconstruction through a second clustering. Simulations show that when the SNR exceeds 0 dB, the unique correspondence matching rate (UCMR) remains above 94% and the average correlation coefficient exceeds 0.85. Comparative experiments confirm that the method overcomes the dependence on linear mixing models, achieves balanced fidelity between the two reconstructed signals, and exhibits superior robustness.