Use of Decorrelation Procedure for Source and Echo Suppression

The so-called semi-blind source separation (SBSS) is an extension of BSS when some signals can be directly and individually measured. SBSS can be used to implement multi-channel acoustic cancellation (MCAEC) without the need for double-talk detection. However, traditional MCAEC approaches based on least mean-square (LMS) lead to a non-unique solution due to correlated sources formed when a single far-end signal is captured by multiple microphones and played out through the near-end loudspeakers. The same nonuniqueness problem can be also expected during SBSS. In this study, the effect of using a decorrelation procedure for mitigating the nonuniqueness problem by applying a nonlinearity or a noise to the far-end reference signals is measured when BSS and stereophonic AEC (SAEC) are simultaneously implemented through SBSS. The simulation results show that the benefit, in terms of the misalignment and the echo return loss enhancement (ERLE), of using a decorrelation procedure can be significant for SBSS while not interfering with the source separation performance at all.