Abstract: Blind source separation (BSS) based on spatial time frequency distribution can separate signals with different time frequency distributions. However the cross terms in the time frequency distribution affect the BSS performance, and different time frequency distributions have different abilities in reducing cross terms. While most current BSS methods are based on fixed kernel functions, we propose BSS based on adaptive time frequency kernel function to improve the performance of resisting cross terms. The proposed technique has strong ability of removing cross terms and is immune to noise. Computer simulations show that the performance of the proposed algorithm is better than that of those using Cohen class time frequency distributions in environments with or without noise.

Key words: adaptive time frequency distribution, spatial time frequency distribution, blind source separation