A fast algorithm for the separation of dependent sources based on bounded component analysis

In order to separate the mixture of bounded signals faster and more accurately, this paper proposes a Bounded Component Analysis (BCA) algorithm based on the detection and removing of steady state learning curve oscillations (DRSSLCO-BCA). In this algorithm, the assumption of statistic independence is replaced by the source boundedness side information with a weaker assumption so that it can separate both independent and dependent sources up to permutation, scaling and phase ambiguities. The algorithm detects the oscillations of the steady state learning curve by the correlation extent at the steady state and uses the iteration of variable step to remove the oscillations of the steady state learning curve. This leads to a faster computation speed and better separation quality. Computer simulations show that the proposed DRSSLCO-BCA algorithm can separate both the dependent and independent sources efficiently with superior performance of computational cost and the separation quality in comparison with the existing BCA algorithms obviously in the noisy or noiseless case.