Toward Data-Driven STAP Radar

Abstract

Catalyzed by the recent emergence of site-specific, high-fidelity radio frequency (RF) modeling and simulation tools purposed for radar, data-driven formulations of classical methods in radar have rapidly grown in popularity over the past decade. Despite this surge, limited focus has been directed toward the theoretical foundations of these classical methods. In this regard, as part of our ongoing datadriven approach to radar space-time adaptive processing (STAP), we analyze the asymptotic performance guarantees of select subspace separation methods in the context of radar target localization, and augment this analysis through a proposed deep learning framework for target location estimation. In our approach, we generate comprehensive datasets by randomly placing targets of variable strengths in predetermined constrained areas using RFView ® , a site-specific RF modeling and simulation tool developed by ISL Inc.For each radar return signal from these constrained areas, we generate heatmap tensors in range, azimuth, and elevation of the normalized adaptive matched filter (NAMF) test statistic, and of the output power of a generalized sidelobe canceller (GSC). Using our deep learning framework, we estimate target locations from these heatmap tensors to demonstrate the feasibility of and significant improvements provided by our data-driven approach in matched and mismatched settings.

Subspace Perturbation Analysis for Data-Driven Radar Target Localization

“…Our objective is to predict targets locations from these 3D heatmap tensors using a robust deep learning framework for target localization. In our prior work (see [1]), we demonstrated that this framework achieves considerable gains over traditional approaches that invoke linear estimation principles. However, much of this analysis was performed in a 'matched' setting, where our convolutional neural network (CNN) architecture was trained and validated on heatmap tensors from the same constrained area (the 'original platform location' instance in [2]).…”

“…In our prior work (see [1]), we demonstrated that this framework achieves considerable gains over traditional approaches that invoke linear estimation principles. However, much of this analysis was performed in a 'matched' setting, where our convolutional neural network (CNN) architecture was trained and validated on heatmap tensors from the same constrained area (the 'original platform location' instance in [2]). To benchmark the localization performance of our CNN framework across 'mismatched' scenarios -where network training and testing is performed with tensors from disparate constrained areas -we separately used the mean normalized output SCNR metric [8].…”

“…The limiting performance of space-time adaptive processing (STAP) techniques for radar target localization has been extensively detailed in previous data-driven explorations of classical problems in radar STAP (see [1], [2]). Fundamentally, much of these limitations are attributable to underlying assumptions made within classical radar STAP literature that degrade target localization post-STAP detection (see [3]- [5]).…”

“…Developed as a means to address the limiting performance of traditional radar STAP methods, data-driven approaches to these classical problems (see [1], [2]) have shown considerable promise, paralleling the emergence of big data in deep learning applications. The rise of these methods has been made possible by the development of simulators that generate highly accurate radar data.…”

“…In Section II, we introduce RFView ® and further describe the robust mismatched case RFView ® example scenario our CNN framework is tested on. This robust scenario is an extension of the mismatched case scenario introduced in [2]. In Section III, we review the NAMF test statistic, which is utilized to generate the heatmap tensors for our training and test datasets.…”