Radar target characterization using model-based bicoherence

Fourier techniques are often used in radar imaging and feature extraction of ISAR data. One drawback is the appearance of artifacts due to scatterer interactions within the target. This paper defines a modified bicoherence, based on a specific target scattering model, to distinguish target scattering centers from target interactions and develop target-specific features for identification. Third-order statistics such as the bicoherence measure the asymmetric properties of distributions and are therefore zero for Gaussian processes such as additive white Gaussian noise. In addition to scatterer range, the model-based bicoherence described in this paper can also estimate the scatterer separation distance. This target information can be used for defining a feature set for target identification that is independent of target aspect angle, assuming the target is modeled as a collection of point scatterers. This information can also be used to determine subspace separation in eigenspace techniques such as the MUSIC algorithm, thereby increasing ISAR image accuracy.