摘要
本文以相关性噪声中的频率估计为信号模型,在理论上推导、解释了影响MUSIC算法的参数估计的偏差和分辨力的各因素及具体关系。清楚的表明在很宽的范围内,估计偏差与信噪比、噪声的相关性强弱无必然联系,它主要是因为计算时的窗效应与被估计参数附近的噪声谱密度的起伏引起的。但噪声相关性很强时,将会引起错误的估计,不过这种错误可通过增大序列长度来消除。在多个信号的情况下,在相当宽的范围内,MUSIC算法的分辨力与信噪比、噪声的相关性强弱无必然联系,但它会随协方差矩阵估计子的维数的增大而增强。第一次全面、深入地揭示了影响相关性噪声中MUSIC算法的参数估计偏差和MUSIC算法的分辨能力的各因素及关系。仿真结果也强有力的证明了这些结论。
The factors affected the MUSIC（ Multi-Signal Classification） algorithm are analyzed in this paper on the basis of the model of frequency estimation in correlated noise theoretically. It is pointed out that estimation bias is independent from the SNR （signal-noise-ratio） and the correlativity of the noise in wide range and that it mainly results from the windowing effect and undulation of the power density spectrum of the noise near parameter estimated. However, severely correlated noise would result in the mistake in estimation, and this kind of mistake could be corrected by lengthening the sample sequence. In multi-signal case, the resolution ability is also unrelated with the SNR and correlativity of the noise in wide extent, and it can be improved by enlarging the dimension of the covariance matrix. Firstly, all these factors and relations are so deeply described. And the results are powerfully supported in the simulations.
出处
《电子测量与仪器学报》
CSCD
2005年第4期6-11,共6页
Journal of Electronic Measurement and Instrumentation
基金
本课题得到国家自然科学基金的资助(项目编号:60072027)。
