On the Use of
Unstructured Array
Models for Radar
Space-Time Adaptive



A. Lee Swindlehurst
Brigham Young University
Department of Electrical and Computer Engineering
Provo, UT 84602
tel: (801) 378-4343

Petre Stoica
Uppsala University
Systems and Control Group
Uppsala, Sweden
tel: +46 18 183074

Abstract Most algorithms for space-time adaptive processing (STAP) in radar applications rely on the availability of array calibration data in order to detect the presence of point targets and estimate their azimuthal position. If calibration errors or multipath is present, standard STAP approaches may suffer from signal cancellation and even strong targets may be missed. Recently, in signal processing research conducted for wireless communications applications, so-called "blind" algorithms have been developed for such situations that employ an unstructured spatial signature model for the array response rather than one parameterized in terms of directions of arrival (DOAs). In this talk, we demonstrate how similar ideas may be applied to the STAP problem to render target detection and parameter estimation algorithms robust to array imperfections and multipath. Despite the fact that such models require more parameters than those using DOAs, the parameter dependence is linear, and simpler estimation procedures may be used. As a result, it becomes possible to consider optimal maximum likelihood (ML) approaches to the problem, which are not typically used for DOA-parameterized models due to the need for a complicated non-linear search. Depending on what assumptions are made about the additive clutter and jammer signals, different ML solutions result, and these are described in the talk. The extended invariance principle of ML estimation is used to show that, in situations where the DOA model holds exactly, the unstructured approach can still achieve asymptotically the same estimation accuracy as the standard approach using a weighted least-squares fit. We also derive a generalized likelihood ratio detection test (GLRT) for the unstructured case, and compare it with the one that results from a DOA-based model. When no array errors or multipath is present, the unstructured GLRT has a slightly higher detection threshold (1­2 dB) than the corresponding DOA-based detector. With calibration errors, however, the unstructured GLRT shows a significant performance advantage.



Direct comments and questions to:

MIT Lincoln Laboratory. All rights reserved.