A Fixed Point Implementation of ICA and its Application in Acoustic Denoising and Multiple Vehicle Classification

Independent Component Analysis has been recently used to solve Blind Source Separation-style problems. In this paper, the implementation of a fixed-point ICA algorithm similar to FastICA is described, and its performance is evaluated. The algorithm is used to perform denoising and source separation on signals recorded by a distributed wireless sensor network performing vehicle tracking in a military field. Both noise and the presence of multiple vehicles affect the performance of typical surveillance tasks such as detection, classification, tracking and localization. The algorithm is shown to perform well in artificial test cases; however, it performs poorly for realworld scenarios, and some possible reasons are offered for this behavior.