Robust non-Gaussian matched subspace detectors

We address the problem of matched subspace detection in the presence of arbitrary noise and interferents, or interfering signals that may lie in a possibly unknown subspace, but that nevertheless corrupt the measurements. A hypothesis test that is robust to interferents yet sensitive to the signal of interest is formulated. The test is applicable to a large class of noise density functions. In addition, specific expressions for the generalized likelihood ratio (GLR) detectors are derived for the class of Generalized Gaussian noise. The detectors are generalizations of the χ
²
, t, and F statistics used with Gaussian noise. For matched filter detection, these expressions are simpler and computationally efficient. ROC performance results based on simulation demonstrate the superior performance obtained with detectors based on the correct noise model. The results also demonstrate the improved performance robust detectors offer when interferents are present.