基于信息论准则的信源数估计改进算法研究

MDL准则和AIC准则是基于信息论准则的信号源数目估计算法。针对其只适用于白噪声背景而不适用于色噪声背景的限制提出了两点改进,首先根据信号协方差矩阵特征值大小分布特点提出一种减小色噪声背景下信号协方差矩阵噪声特征值发散程度的特征值校正算法。该算法可以明显改善色噪声背景下噪声特征值的特性,使其接近白噪声条件下特征值的特性,利用校正后的特征值,MDL准则和AIC准则可以进行色噪声背景下的信源数目估计。然后为了提高检测精度,采用了基于MDL准则和AIC准则函数比值的估计方法。理论分析和仿真结果表明,改进后的MDL准则和AIC准则可以有效估计色噪声背景下的信号源数目,并且性能优于适用于色噪声背景的盖尔圆算法。

Optimizing Algorithm in Determining the Number of Signals Based on Information Criterion

The MDL criterion and the AIC criterion estimate the number of signals based on the information criterion.For their application to colored noise,the paper analyzed the effect of colored noise to the eigenvalues of the covariance matrix.Then a method to eliminate the inequality of the eigenvalues caused by colored noise is proposed.Using the corrected eigenvalues,the information criterion can estimate the number of sources in colored noise.In order to improve the accuracy of estimation,a function ratio estimation method based on MDL criterion and the AIC criterion was presented also.Simulation results show that the MDL criterion and the AIC criterion,using the corrected eigenvalues and ratio method,outperform the Gerschgorin disk method.