A Study of ICA Based DOA Estimation with Respect to Permutation Ambiguity, Scaling Ambiguity and Sensor Gain Mismatch

Recently, direction-of-arrival (DOA) and position estimation for acoustic signals have been studied intensively and many different algorithms have been proposed. Among different approaches for multiple sources, independent component analysis (ICA) based methods have drawn much attention. In this paper, we study the effects of permutation ambiguity, source scaling ambiguity and sensor gain mismatch on source localization based on an estimate of the channel matrix or its inverse. We show that the source scaling ambiguity can be removed from the estimate of the inverse channel by proper normalization, but not the sensor gain mismatch. We evaluate the influence of source scaling ambiguity and sensor gain mismatch on two frequency domain ICA based localization algorithms, the averaged directivity pattern (ADP) and the state coherence transform (SCT) using simulations. We show that the original ADP is very sensitive to source scaling ambiguity and sensor gain mismatch. In contrast, SCT is completely insensitive to these effects and shows a superior localization performance.