Frequency-Domain Convolutive Bounded Component Analysis Algorithm for the Blind Separation of Dependent Sources

Aiming at the problem of dependent source separation in complex mechanical systems, the highly universal frequency-domain convolutive bounded component analysis (FDCBCA) algorithm suitable for dependent sources and independent sources is proposed. First, the time-domain convolutive model of blind source separation (BSS) is converted into the instantaneous one in each frequency bin by the way of short-time Fourier transform (STFT). Then, a complex bounded component analysis (CBCA) algorithm is built by extending the boundary minimization bounded component analysis (BMBCA) algorithm proposed in our previous research to complex domain, which is adopted to separate the complex subsignals in different frequency bins. Next, borrowing the boundary minimization criterion of bounded component analysis (BCA), the scale alignment and permutation alignment methods for dependent sources are innovatively proposed. Finally, the time-domain separated signals are recovered by the inverse STFT (ISTFT). The high-accuracy separation performance of the FDCBCA algorithm is verified by numerical simulation and excitation experiments of aluminum honeycomb panel cabin structure. In short, this article provides a vibration source separation algorithm with high precision and strong universality, which can provide the reliable foundation for the identification of vibration sources in complex mechanical systems.