In this paper, we deal with the problem of adaptive detection of distributed targets embedded in colored noise modeled in terms of a compound-Gaussian process and without assuming that a set of secondary data is available. The covariance matrices of the data under test share a common structure while having different power levels. A Bayesian approach is proposed here, where the structure and possibly the power levels are assumed to be random, with appropriate distributions. Within this framework we propose GLRT-based and ad-hoc detectors. Some simulation studies are presented to illustrate the performances of the proposed algorithms. The analysis indicates that the Bayesian framework could be a viable means to alleviate the need for secondary data, a critical issue in heterogeneous scenarios.

This paper addresses the problem of detecting multiple point-like targets in the presence of steering vector mismatches and Gaussian disturbance with unknown covariance matrix. To this end, we first model the actual useful signal as a vector belonging to a proper cone whose axis coincides with the whitened direction of the nominal array response. Then we develop two robust adaptive detectors resorting to the two-step GLRT-based design procedure without assignment of a distinct set of secondary data. The performance assessment has been conducted by Monte Carlo simulation, also in comparison to previously proposed detectors, and confirms the effectiveness of the newly proposed ones. In the last part of the work, in order to restore the detection performance of the newly proposed detectors in the presence of a large number of range cells contaminated by useful signals, we consider two adaptive detectors which resort to the structure information of the disturbance covariance matrix, and show that the a-priori information on the covariance structure can lead to a noticeable performance improvement.

In this paper we consider the problem of detecting multiple point-like targets in the presence of steering vector mismatches and Gaussian disturbance with unknown covariance matrix. To this end, we first model the actual useful signal as a vector belonging to a proper cone whose axis coincides with the whitened direction of the nominal array response. Then we develop two new robust adaptive detectors resorting to the two-step generalized-likelihood ratio test (GLRT) design procedure without assignment of a distinct set of secondary data. Finally, a performance assessment, conducted by Monte Carlo simulation, show that the proposed detectors achieve a visible performance improvement over their natural counterparts.

We address adaptive detection of Swerling 2 pulse trains by an array of antennas. The disturbance is modeled in terms of a state space model and the ideas of subspace identification are used to come up with a GLRT-based detector. Such detector is compared by Monte Carlo simulation with a Kelly's detector derived assuming that returns are temporally uncorrelated (but spatially correlated) and that a proper set of secondary data is available

We address adaptive discrimination between the signal of interest and a sidelobe interferer. To this end, we propose a detector derived resorting to a GLRT implementation of the generalized Neyman-Pearson rule and a two-stage detection scheme. The adaptive detectors rely on secondary data, free of signal components, but sharing the statistical characterization of the noise in the cell under test, in order to guarantee the CFAR property

This letter focuses on the design of selective receivers for homogeneous scenarios where a very small number of secondary data are available. To this end, at the design stage it is assumed that the cell under test (CUT) contains a fictitious signal orthogonal to the nominal steering vector under the null hypothesis; the clutter covariance matrix is modeled as a random matrix with an inverse complex Wishart distribution. Under the above assumptions, we devise two Bayesian detectors based on the GLRT criterion, both one-step and two-step. It is shown that the proposed detectors have the same detection structure as their non-Bayesian counterparts, substituting the colored diagonal sample covariance matrix (SCM) for the classic one. Finally, a performance assessment, conducted by Monte Carlo simulations, has shown that our detectors ensure better rejection capabilities of mismatched signals than the existing Bayesian detectors, at the price of a certain loss in terms of detection of matched signals.

We consider the problem of detecting a signal of interest in the presence of compound-Gaussian clutter, without resorting to secondary data in order to infer the clutter covariance matrix. Towards this end, we assume that both the texture τ and the speckle covariance matrix R are random variables with some a priori distributions. Marginalizing with respect to these variables, the probability density function of the observed primary data is derived, leading to a closed-form expression for the generalized likelihood ratio test (GLRT) of the problem at hand. Accordingly, the GLRT assuming that τ is deterministic is also derived. The two detectors are assessed through numerical simulations

We address adaptive detection of coherent signals backscattered by possible point-like targets or originated from electronic countermeasure (ECM) systems in presence of thermal noise, clutter, and possible noise-like interferers. In order to come up with a class of decision schemes capable of discriminating between targets and ECM signals, we resort to generalized likelihood ratio test (GLRT) implementations of a generalized Neyman-Pearson rule (i.e., for multiple hypotheses). The adaptive detectors rely on secondary data, free of signal components, but sharing the statistical characterization of the noise in the cell under test. The performance assessment focuses on an adaptive beamformer orthogonal rejection test (ABORT)-like detector; analytical expressions for the probability of false alarm, the probability of detection of the target, and the probability of blanking the ECM (coherent) signal are given. More remarkably, it guarantees the constant false alarm rate (CFAR) property. The performance assessment shows that it can outperform the adaptive sidelobe blanker (ASB) in presence of ECM systems.

We address the problem of estimating a covariance matrix R using K samples z k whose covariance matrices are τ kR, where τ k are random variables. This problem naturally arises in radar applications in the case of compound-Gaussian clutter. In contrast to the conventional approach which consists in considering R as a deterministic quantity, a knowledge-aided (KA) approach is advocated here, where R is assumed to be a random matrix with some prior distribution. The posterior distribution of R is derived. Since it does not lead to a closed-form expression for the minimum mean-square error (MMSE) estimate of R, both R and τ k are estimated using a Gibbs-sampling strategy. The maximum a posteriori (MAP) estimator of R is also derived. It is shown that it obeys an implicit equation which can be solved through an iterative procedure, similarly to the case of deterministic τ ks, except that KA is now introduced in the iterative scheme. The new estimators are shown to improve over conventional estimators, especially in small sample support.

We address the problem of adaptive detection of a signal of interest embedded in colored noise modeled in terms of a compound-Gaussian process. The covariance matrices of the primary and the secondary data share a common structure while having different power levels. A Bayesian approach is proposed here, where both the power levels and the structure are assumed to be random, with some appropriate distributions. Within this framework we propose MMSE and MAP estimators of the covariance structure and their application to adaptive detection using the NMF test statistic and an optimized GLRT herein derived. Some results, also in comparison with existing algorithms, are presented to illustrate the performances of the proposed detectors. The relevant result is that the solutions presented herein allows to improve the performance over conventional ones, especially in presence of a small number of training data.

In this correspondence, we focus on the design and the analysis of schemes aimed at estimating the position of multiple point-like targets that fall among three adjacent samples of the matched filter output (typically present in the processing chain of a radar). To this end, we exploit spillover of targets’ energy to adjacent range cells. The number of targets (and the corresponding Doppler frequency shifts) is assumed to be known. Moreover, we suppose that received useful signals can be modeled in terms of vectors known up to deterministic parameters and that they are embedded in correlated Gaussian noise with unknown covariance matrix. For estimation purposes we assume that a set of secondary data, free of signal components, but sharing the same covariance matrix of the noise in the cells containing signal returns, is available. The analysis, also in comparison to a possible competitor, proves the superiority of multitarget schemes with respect to single target ones, even under reasonably mismatched scenarios.