A robust signal subspace estimator

Abstract

International audienceAn original estimator of the orthogonal projector onto the signal subspace is proposed. This estimator is derived as the maximum likelihood estimator for a model of sources plus orthogonal outliers, both with varying power (modeled by Compound Gaussians process), embedded in a white Gaus-sian noise. Validity and interest-in terms of performance and robustness-of this estimator is illustrated through simulation results on a low rank STAP filtering application

Robust Covariance Matrix Estimation in Heterogeneous Low Rank Context

“…Nevertheless, we point out that, though it appears that the choice of the second step does not impact mainly the performance in terms of error on the estimation of the shape matrix, the exact resolution of this step may improve the performance in terms of CSP estimation as shown in [39]. Moreover this step can also provide an additional robustness to corruption of the samples by outliers, as shown in [58]. Also notice that when few samples are available, the approached solutions may provide lower NMSE (which is also observed in Fig.…”

Bayesian Robust Signal Subspace Estimation in Non-Gaussian Environment

“…Notice that, as done in [9], [11], the non-null eigenvalues of Σ P are assumed to be equal to 1. The hypothesis of eigenvalues equality is a relaxation that is made for tractability purposes but still offers interesting performance in practice [11]. Additionally, any scaling of the signal CM is absorbed in the textures τ k of the CG distribution.…”

“…This update has the same complexity as the deterministic case [11]. It does not require a Gibbs sampling step [4].…”

“…In order to enjoy best of both the CG modeling and the Bayesian subspace estimation approaches, we derive a new MMSD estimator for the data modeled as a CG distributed signal of interest embedded in heterogeneous noise (CG outliers plus white Gaussian noise) [11]. For this model, the complex invariant Bingham (CIB) [12] is considered as a prior of the subspace of interest.…”

Robust rank constrained kronecker covariance matrix estimation

“…In this paper, we follow the second approach and specifically focus on robustly estimating CM that can be expressed as the Kronecker product of (structured) low rank matrices. Indeed, this structure often arise in the context of array processing, such as in MIMO-Radar, COLD array and STAP: the Kronecker product structure generally comes from a redundancy induced by the multiplication of sensors and/or signal emissions [13,14], while the low rank structure is induced by signals (or interference) being contained in a low dimensional subspace [15]. This work was supported by the Hong Kong RGC 16207814 research grant.…”