A robust source number estimation method based on few snapshots
Xiaoyu Zheng,Jianwei Wang,Yechao Bai
DOI: https://doi.org/10.13232/j.cnki.jnju.2016.06.019
2016-01-01
Abstract:In the array signal processing of radar,signal source number estimation is a basis of spatial spectrum analysis.The traditional source number estimators are mostly based on the assumption that the number of the array elements is fixed and the number of sampling snapshots tends to infinite.Those traditional methods perform well under the condition that the number of snapshots is extremely higher than the number of array elements.However, in the arrays with larger number of elements and few sampling snapshots,the performance of those traditional source number estimation methods declines,even cannot work properly in practical applications.Results from large dimensional random matrix theory are leveraged in this paper to solve this problem.Large dimensional random matrix theory adopts a new asymptotic regime,i.e.,the number of array elements tending to infinite,the number of snapshots tending to infinite,with the ratio between them tending to a constant larger than zero.The results deduced from the new regime handle the situation when the number of array elements and the number of snapshots are of the similar magnitude.Under the certain condition of higher sampling dimension and limited snapshot number,firstly we try to use the matrix weighted algorithm to optimize the covariance matrix of the signal space to reduce the mean square error(MSE)of the sample covariance matrix versus the real signal covariance matrix,and then we apply a source number estimation method using theory of the random matrix to predict the number of signal source number. Through those work,we get a relatively stable source number estimation.By the way of some related simulations on Matlab,we can find that,under the conditions with limited snapshots,the method in this paper can efficiently and stably increase the accuracy of the source number estimation.