Nonparametric statistics for subspace based frequency estimation

The paper introduces new subspace based frequency estimation methods. The techniques are based on estimating the noise or signal subspace from the sample spatial sign autocovariance matrix. The theoretical motivation for the techniques is shown under the white Gaussian noise assumption. A simulation study is performed to demonstrate the robust performance of the algorithms both in Gaussian and non-Gaussian noise. The results imply that when the noise is Gaussian, the proposed methods have similar good performance as the standard subspace methods (MUSIC, ESPRIT). When the noise is heavy-tailed, the proposed methods outperform the standard subspace techniques.