Abstract: A classical Blind Source Separation（BSS） algorithm based on Time-Frequency Distribution（TFD） was proposed by Belouchrani, A and Amin, M.G in 1998. When noise is present, however, this algorithm’ performance will decrease. This paper introduces a new BSS approach for source noise, which is achieved by firstly calculating the Wigner TFD matrix of observed signals, and uses the Hough transform to convert the signals detection to find the local peak values in the parameter space through considering the TFD as an image, followed by joint diagonalization of a combined set of matrix selected by auto-term theory, to estimate the source signals. This method extends the BSS constraints, and increases the robustness by spreading the noise power while localizing the source energy in the time-frequency domain.

Key words: source noise, Wigner distribution, Hough transform, blind source separation

摘要： 1998年，Belouchrani，A和Amin，M.G基于时频分布提出了一种经典的时频盲源算法，不足是当有噪声存在时，性能会下降。主要考虑源噪声的盲源分离问题，以Wigner分布计算观测信号的时频阵并将其看做图像，利用Hough变换将信号检测转换为在参数空间寻找局部极大值的问题，运用自项点理论选择合适的矩阵进行联合近似对角化估计源信号。该方法扩展了盲源分离的限制条件，且通过把噪声能量扩展到整个参数平面而只选择信号能量占主导的时频点，对噪声具有一定的抑制能力。

关键词: 源噪声, Wigner分布, Hough变换, 盲源分离