Neighborhood Selection and Cholesky Matrix for CAPON-based SAR Tomography Inversion

Synthetic Aperture Radar (SAR) Tomography gain particular interest due to its 3-D reconstruction capability in built-up areas. Spectral estimation approach enables the scene reflectivity to be restored from multi-baseline SAR datasets. The lack of a priori information about the observed regions and the need for high-accuracy reconstructions encourage researchers to use non-parametric multi-looking spectral estimation methods. These estimators require the generation of the Multidimensional Covariance Matrix (MCM) by means of a local spatial averaging operation. In this context, the paper's aim is to contribute to the tomographic process development by: 1) considering an appropriate selection of surrounding neighbors for the data covariance matrix estimation, where the boxcar window is replaced by a range-oriented rectangular window; and 2) incorporating the root square matrix of the MCM (also known as the Cholesky matrix) into the CAPON-based spectral analysis algorithm. The experimental results on data acquired by TerraSAR-X satellite demonstrate the effectiveness of the proposed tomographic neighborhood selection, along with the potential of the improved CAPON-based estimator in regard to scatterers' height estimation.