IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 55, NO. 4, APRIL 2007 1223 Adaptive Radar Detection of Distributed Targets in Homogeneous and Partially Homogeneous Noise Plus Subspace Interference Francesco Bandiera, Member, IEEE, Antonio De Maio, Member, IEEE, Antonio Stefano Greco, and Giuseppe Ricci, Member, IEEE Abstract—This paper addresses adaptive radar detection of distributed targets in noise plus interference assumed to belong to a known or unknown subspace of the observables. At the design stage we resort to either the GLRT or the so-called two-step GLRT-based design procedure and assume that a set of noise-only data is available (the so-called secondary data). Detection algo- rithms have been derived modeling noise vectors, corresponding to different range cells, as independent, zero-mean, complex normal ones, sharing either the same covariance matrix (homogeneous en- vironment) or the same covariance matrix up to possibly different (mean) power levels between primary data, i.e., range cells under test, and secondary ones (partially homogeneous environment). The performance assessment has been conducted by Monte Carlo simulation, also in comparison to previously proposed detection algorithms, and confirms the effectiveness of the newly proposed ones. Index Terms—Adaptive detection, distributed targets, general- ized-likelihood ratio test, interference rejection. I. INTRODUCTION A DAPTIVE radar detection of point-like and of multiple point-like or range-spread (in a word, distributed) tar- gets embedded in Gaussian or non-Gaussian disturbance has received an increasing attention from the radar community in recent years [1]–[11, and references therein]. More precisely, adaptive detection of distributed targets has been addressed in [1] and [2]; therein useful target echoes have been modeled as signals known up to multiplicative factors, possibly different from one range cell to another, namely supposed to belong to a one-dimensional (1–D) subspace of the observables. Noise is modeled in terms of independent, complex normal random vec- tors with a common covariance matrix up to possibly different power levels. Covariance matrices are unknown at the receiver Manuscript received September 28, 2005; revised April 4, 2006. This work was supported in part by the Ministero dell’Istruzione, dell’Università e della Ricerca under the National Research Project “Innovative Signal Processing Al- gorithms for Radar Target Detection and Tracking”. This work was presented in part at the 39th Asilomar Conference on Signals, Systems and Computers, Pa- cific Grove, CA, October 30–November 2, 2006, and in part at the International Conference on Acoustics, Speech and Signal Processing (ICASSP), Toulouse, France, May 14–19, 2006. The associate editor coordinating the review of this paper and approving it for publication was Prof. Leslie Collins. F. Bandiera, A. S. Greco, and G. Ricci are with the Dipartimento di Ingeg- neria dell’Innovazione, Università del Salento, I-73100 Lecce, Italy (e-mail: francesco.bandiera@unile.it; antoniostefano.greco@unile.it; giuseppe.ricci@ unile.it). A. De Maio is with the Dipartimento di Ingegneria Elettronica e delle Tele- comunicazioni, Università degli Studi di Napoli “Federico II,” I-80125 Napoli, Italy (e-mail: ademaio@unina.it). Digital Object Identifier 10.1109/TSP.2006.888065 and a set of noise-only additional data (the so-called secondary data) is available for estimation purposes. In [1] detectors based on the Generalized Likelihood Ratio Test (GLRT) and ad hoc decision schemes (relying on the two-step GLRT-based design procedure) have been proposed for the case that noise vectors share one and the same covariance matrix (homogeneous sce- nario) or the same covariance matrix up to possibly different power levels between primary data, i.e., range cells under test, and secondary ones (partially homogeneous scenario). Pro- posed detectors possess the constant false alarm rate (CFAR) property under the design assumptions. In [2] the two-step GLRT-based design procedure is adopted in order to address detection of target echoes in a heterogeneous scenario, namely for the more general case that noise returns share the same covariance matrix up to possibly different power levels from one cell to another. Remarkably, the proposed ad hoc detector guarantees the CFAR property with respect to the covariance matrices of noise returns (under the design assumptions). Detection of point-like targets, modeled as vectors con- strained to belong to a known subspace of the observables, in presence of interference and noise of unknown power has been considered in [3]; therein the interference subspace is known and linearly independent of the signal subspace. Adaptive subspace detection of point-like targets has been addressed in [4]. Detection of distributed targets, modeled in terms of vectors confined to a known subspace, and embedded in noise of unknown power plus deterministic interference, assumed to belong to an unknown subspace, has been considered in [5]. Finally, several detection algorithms are encompassed as special cases of the amazingly general framework and deriva- tions in [6]. All of the above papers assume that target echoes can be modeled in terms of deterministic signals; the case of targets modeled in terms of Gaussian signals, confined to a known subspace of the observables, has been dealt with in [7]–[10]. More precisely, [7]–[9] address adaptive detection in presence of Gaussian noise while [10] also considers the case of non-Gaussian disturbance, modeled as a compound-Gaussian process. Several works have also attacked clutter modeling and performance analysis of radar detection algorithms using measured data, see [11, and references therein]. In the following we address adaptive detection of distributed targets, within a given set of range cells (the primary data), in presence of noise, modeled in terms of complex normal random vectors with unknown covariance matrix, plus subspace inter- ference. A set of noise-only additional data (the secondary data) 1053-587X/$25.00 © 2007 IEEE