Abstract:

Nonparametric spectral estimation methods are analyzed. It shows they are all weighted least square(WLS) estimation and the only difference is how to build a weighted matrix from generalized noise covariance. Based on the same idea, an adaptive iterative nonparametric spectral estimation method is proposed. It can simultaneously estimate the signal spectrum and measurement noise. During each iteration process, the generalized noise covariance matrix is estimated from the last time estimative result. And it can approach the real noise covariance matrix iteratively. Analyses and numerical simulations validate that the proposed method has the characteristics of high resolution with low spectrum leakage, the enhanced invertibility of signal covariance matrixes and a free choice of spectrum range.