高斯噪声中的参数盲估计

一般的盲信号处理方法常忽略噪声的影响,而实际问题中噪声的影响是存在的。本文主要讨论了在协方差矩阵未知的加性高斯噪声中混合系数的盲估计问题。以最大似然估计为基础,本文提出一种求解参数的最优化算法,并给出了混合矩阵和协方差矩阵的计算式。采用高斯混合模型(GMM)来逼近源信号的概率密度函数,简化了算法中的积分,导出了一种实用的期望最大算法(EM)算法迭代式。计算机仿真结果表明,算法不仅能稳定收敛,而且在低信噪比下的性能也很好。

Blind estimation of parameters in Gaussian noise

In general, some methods of blind signal processing ignore noise, but in practice, noise affects the performance of algorithms, especially seriously in some areas. This paper provides solutions to the problem that mixing matrix is estimated blindly in Gaussian noise with unknown covariance. Based on Maximum Likelihood estimation, the equations are given for solving the mixing matrix and covariance matrix. Gaussian Mixture Model (GMM) is used to approximate the pdf of sources and results in a practical Expectation Maximum (EM) algorithm. Computer simulation shows that this algorithm is convergent and has good performance in low SNR.