Statistical Regularization as an Alternative to Model Order Selection

Abstract

The correct functioning of parametric focusing techniques [e.g., MUltiple SIgnal Classification (MUSIC)] require a proper selection of the model order. For such aim, a methodology based on the Kullback-Leibler information criterion is commonly employed. These methods perform well due to its propensity to choose relatively large model orders, which tend to retrieve good-fitted responses when the data generating mechanism is more complex than the models used to fit. However, some solutions can be misleading, since only the most proper model order (i.e., the actual number of targets) guaranties best performance. As an alternative, this work suggests employing statistical regularization instead of model order selection (MOS) approaches. First, a model with large order is chosen to perform focusing via parametric methods; subsequently, statistical regularization is applied, seeking to attain good-fitted solutions. To demonstrate the capabilities of the addressed novel strategy, Synthetic Aperture Radar (SAR) Tomography (TomoSAR) is considered as application.