Capon and APES spectrum estimation for real-valued signals

This paper considers the problem of estimating the spectrum of real-valued signals. We propose real-valued versions of the Capon and the APES spectral estimators. The estimators are derived as members of the Matched-Filterbank (MAFI) estimator class as introduced in [1]. Furthermore, we show that the real-valued estimators will be unbiased, whereas the complex-valued estimators will have a (slight) bias for real-valued data. Finally, we conclude the paper with a numerical example illustrating the performance of the proposed estimators. 1. INTRODUCTION In the filterbank approach to spectral estimation, the amplitude of the spectrum is estimated by passing the signal through a narrowband filter, h! , with varying center frequency ! (see, e.g., [2]). Here, and in the following, the subscript ! is used to indicate a parameter's dependence on the filter's center frequency. Let fy(t); t = 1; : : : ; Ng denote the available (stationary) data sample of which the spectrum is to be estimated,...