Over-complete Blind Source Separation for Nonlinesr Convolutive Mixtures

针对OC-BSS提出了一种反馈方法及其学习算法。通过使用超过一半的信号源的传感器，至少一个输出可以分离单个信号源。该输出被反馈到分离块的输入，并从观测值中减去，以减少等效信号源的数量。提出了两种反馈方法，即基于观测信号直方图的直接减法和基于上述分离信号直方图的样本消除法。在后一过程中，通过频谱抑制技术进一步抑制信号失真。与传统方法相比，该方法可以提高信干比6~10dB。对盲源分离系统中的信号分离和信号失真进行了理论分析。前馈盲源分离系统在解空间中具有一定的自由度。因此，很可能会出现信号失真。其次，给出了多通道FF-BSS系统完全分离和无失真的条件。该条件作为无失真约束被并入学习算法中。对于采用所提出的无失真约束的传统方法和新的学习算法，使用语音信号和平稳有色信号进行了计算机仿真。该方法在保持较高分离性能的同时，能显著抑制信号失真。

A feedback approach and its learning algorithm are proposed for the OC-BSS. By using the sensors more than a half of the sources, at least one output can separate a single signal source. This output is fed back to the inputs of the separation block, and is subtracted from the observations, in order to reduce the number of equivalent signal sources. Two kinds of feedback methods are proposed, which are direct subtraction and sample elimination based on histogram of the observed signals and the separated signal above. In the latter process, signal distortion is further suppressed by the spectral suppression technique. The proposed method can improve a signal to interference ratio by 6〜10 dB compared to the conventional methods.Source separation and signal distortion are theoretically analyzed in blind source separation (BSS) systems implemented in both the time and the frequency domains. Feedforward (FF-) BSS systems have some degree of freedom in the solution space. Therefore, signal distortion is likely to occur. Next, a condition for complete separation and distortion free is derived for multi-channel FF-BSS systems. This condition is incorporated in learning algorithms as a distortion free constraint. Computer simulations using speech signals and stationary colored signals are performed for conventional methods and the new learning algorithms employing the proposed distortion free constraint. The proposed method can drastically suppress signal distortion, while maintaining a high separation performance.