Abstract: Multiple signal classification (MUSIC) and fast independent component analysis are jointly applied to separation and waveform reconstruction of multiple spatio-temporal source signals. The MUSIC is used for identifying the noise subspace in an observation with second order statistics and search for multisource location parameters that orthogonalize the steering vector and the noise subspace. Thus, directions of arrival of multiple sources are estimated. The FastICA based on fixed-point iteration is used for estimating the sensor gain pattern and the unknown mixing matrix of source signals, by minimizing the mutual information, a measure constructed by all statistics higher than second order. Multiple unknown spatio-temporal sources are separated and their waveforms are reconstructed. Experimental results indicate that the newly proposed method enables effective identification of complex valued spatio-temporal mixing matrix, precise separation and reconstruction of mixed sources from different directions, and establishing a solid base for some further applications such as weak signal detection and feature extraction in fault diagnosis.

Key words: Direction of arrival, Fast independent component analysis, Multiple signal classification, Steering vectors, Droplet impingement, Numerical simulation, Stationary pileup, weld pool dynamics, Key words: Double-electrode micro plasma arc welding