Kalman Filter-based and Semblance-based methods for Blind Source Separation
Blind Source Separation (BSS) is a well-known problem in signal processing and stillreceives attention from the scientific community, given the applicability in many differentareas. The Kalman Filter, a classic estimation tool, has been explored to the BSS problemas an additional step, or a parameter estimator approach. This work presents a theoreticalbackground overview of the BSS problem formulation and the application of the Kalmanfilter in source separation as a parameter estimator in two different approaches; joint and dualparameter estimation. These approaches are evaluated in different scenarios, with analysis overthe initialization details and algorithm parameters variation, showing simulation results andperformance comparison, evaluated by SIR and MER. Results showed that both approaches canperform separation in a two-source-two-mixture scenario but may not converge for an increasednumber of sources. Moreover, it is presented a beamformer algorithm for source separation basedon the semblance coherence function. The algorithm evaluation for artificially mixed signals wasobtained using an objective intelligibility metric (STOI) throughout Monte Carlo simulations intwo scenarios and different SNR levels. Results are compared with classic techniques: GSS andDelay-and-Sum, where the proposed algorithm achieves the best performance under no influenceof additive noise.