Blind Source Separation using Dynamic Mode Decomposition

This thesis demonstrates a novel approach to performing blind source separation through the use of Dynamic Mode Decomposition (DMD). Blind source separation is the act of separating mixture signals into accurate estimations of the source signals used to create the mixture signals. The biggest challenge in blind source separation is being able to accurately estimate our unmixing matrix. This is challenging because the mixing matrix as well as the source signals used to compose the mixture signals are unknown. All that is available are our mixture signals. The most popular approach to solving such a problem, is through the use of Independent Component Analysis (ICA). ICA is very efficient and effective when performing blind source separation. However, there is one main downside to this approach. The main downside to ICA is that the algorithm is inconsistent with its result. ICA generally performs very well, however this algorithm has a chance to fail and yield bad results. This thesis aims to produce a method that can not only out perform ICA, but also solve the inconsistency problem that exists. This work contributes to the field by developing a novel algorithm that is able to alleviate the inconsistencies present within ICA. DMD is able to solve this problem. This is because the unmixing matrix used to perform the the separation will always be the same. This then allows for the separation algorithm to be consistently the same.