Separation of Mixed Signals Using DWT Based Overcomplete ICA Estimation

In this paper, over complete independent component analysis (Over complete ICA) is solved using discrete wavelet transform based parallel architecture, which is a combined system consisting of two sub-over complete ICA. One process takes the high-frequency wavelet part of observasions as it's inputs and the other process takes the low-frequency part, then the final results are generated by merged their results. The proposed method utilizes the full observation information compared to the existing over complete ICA algorithms, but the effective input length of the two parallel process is halved. Therefore a method is provided for over complete ICA problems and the experimental results in this paper indicates it's good performance for separating the mixed speech signals.