## 1. Introduction

[2] The quality of atmospheric radar observations is limited by slow update times, clutter contamination, and range velocity ambiguities, among other factors [*Battan*, 1973; *National Research Council*, 2002; *Doviak and Zrnić*, 1993; *Friedrich et al.*, 2006]. Imaging radars consist of a widebeam transmitter and many independent receiving subarrays, and can mitigate some of these problems by illuminating a large field of view and using adaptive beamforming to scrutinize the structures and dynamics within this region. These radars have been applied to investigate clear-air turbulence and precipitation [*Mead et al.*, 1998; *Palmer et al.*, 1993, 2005], plasma irregularities in the equatorial *F* region [*Hysell*, 1996], structures of the polar mesosphere summer echoes [*Yu et al.*, 2001], airborne clutter sources in the boundary layer [*Cheong et al.*, 2006], and phenomena in the stratosphere-troposphere (ST) layers [*Hélal et al.*, 2009], among others [*Fukao*, 2007].

[3] Currently, the observation of mean signal power using an imaging radar is typically obtained by subtracting the noise power from the covariance function at lag zero. Classic techniques used in obtaining the mean signal power include signal statistics separation [*Hildebrand and Sekhon*, 1974], spectral thresholding [*Gordon*, 1997], spurious spectral peaks averaging [*Marple*, 1987], and matched filters [*Haykin*, 1996]. The mean signal power is unfortunately biased if the noise power is unsuccessfully estimated. A natural incentive exists to obtain the signal power in a way that bypasses the need for estimating the noise power. In this paper, a multilag (ML) correlation technique is introduced that satisfies this requirement by exploiting the temporal correlation difference between the scattered atmospheric signal and the receiver noise. Previously, higher-order correlation values have been used for similar purposes to retrieve velocity [*Strauch et al.*, 1978], spectrum width [*Srivastava and Jameson*, 1979], linear depolarization ratio [*Melnikov*, 2006; *Hubbert et al.*, 2003], and cross-polarization ratio [*Melnikov*, 2006].

[4] In the next section, the derivation of the ML technique and simulation of its performance for a Gaussian spectrum time series signal are presented. In section 3, the ML technique is applied to real data and the results are compared to those obtained using conventional Fourier and adaptive Capon beamforming. In section 4, conclusions are presented.