An adaptive thresholding algorithm is presented that can be used in conjunction with the multi-pass generalized inner product (GIP)-based editing method to eliminate non-homogeneities from the training data used for STAP applications, such as adaptive radars. The algorithm exploits a property of the generic structure of the ordered GIP statistic, along with a single user-specified parameter related to the Type I error of incorrectly excising target-free training data, to adaptively determine the thresholds for excising target-contaminated training data. The performance of the method is demonstrated using both high-fidelity site-specific simulated data with both ideal and realistic waveforms as well as measured data from the Multi-Channel Airborne Radar Measurement (MCARM) experiment.

Secondary data selection for estimation of the clutter covariance matrix, needed in space-time adaptive processing (STAP), is normally obtained from range rings nearby the cell under test. The assumption is that these range rings contain cells that are representative of the clutter statistics in the test cell. However, in a nonhomogeneous terrain environment, this may not be true. An innovative approach is presented, in the area of knowledge-aided STAP, which utilizes terrain data from the United States Geological Survey (USGS) to aid in the selection of secondary data cells. Results have been obtained and compared with the sliding (cell averaging symmetric) window method of secondary data selection. This comparison indicates that making use of the surveillance terrain knowledge improves STAP performance.

A particularly active area of research in space-time adaptive processing (STAP) involves scenarios in which the sample support available for training the adaptive processor is limited. Several of these scenarios are of significant current interest. One of those scenarios is an environment in which targets are potentially so dense (relative to the sample support requirements) that they bias the weight training, thereby causing significant performance degradation of the STAP processor. Such environments include those containing roads and highways, for example. Other related problems include scenarios in which the clutter itself is not homogeneous over significant ranges, e.g. conditions where the terrain type is highly variable, urban environments, etc. One technique that addresses the low-sample support conditions described above is the Parametric Adaptive Matched Filter (PAMF). Performance of this technique and several contending STAP approaches will be demonstrated using the KASSPER challenge dataset only.

This paper presents further developments and results of the FRACTA algorithm which has been shown to be robust to nonhomogeneous environments containing outliers. The main focus here is upon the detection of targets in the KASSPER I challenge data cube which possesses dense clusters of targets and the highly nonhomogeneous KASSPER II data in which severe clutter is present over all ranges and dopplers thereby hindering the identification of a dominant clutter ridge. The KASSPER II dataset is further exacerbated by dense clusters of targets as well as the presence of several deep shadow regions that not only prevent target detection but may also skew covariance matrix estimation. A doppler-dependent thresholding technique is developed which is then incorporated into the FRACTA.E framework and then applied to the KASSPER II dataset. Simulation results are compared with the standard sliding window scheme as well as when clairvoyant knowledge of the covariance matrices is employed. For the KASSPER I data set the standard approach detects only 11 of the 268 total targets while the FRACTA.E algorithm detects 192 which is the same amount achieved when clairvoyant clutter covariance knowledge is used. For a selected CPI of the KASSPER II dataset the standard approach is found to perform poorly with only 14 targets detected out of 127 while the FRACTA.E algorithm with the proposed modification detects 51 targets; 73 targets are found when clairvoyant clutter covariance knowledge is employed.

We simulate the performance of existing and planned tactical GMTI systems using data cubes derived from high-fidelity interferometric SAR, to assess the utility of these GMTI systems for an auxiliary terrain height estimation function. The two systems are current and next generation GMTI radars with linear and planar arrays that could be mounted on a manned aircraft or a large UAV. In order to achieve the vertical element separation required for interferometric terrain height estimation, the antenna array in the first case must be pitched up relative to the horizontal position that is ordinarily used for DPCA or STAP clutter suppression. The purpose of the study is to determine whether useable terrain elevation maps can be generated by interferometric techniques within the operational constraints of these systems. Such elevation map data, obtained using a GMTI radar, would be valuable to knowledge-aided algorithms which rely on precise three-dimensional registration of radar data with terrain or road databases.

New algorithms are being developed in the radar community that blend a priori knowledge source processing with traditional digital signal processing concepts. This operational blend necessitates a system-level architecture capable of delivering both high processing throughput and memory bandwidth. This paper derives these system parameters from the Knowledge Aided Pre-Whitening algorithm and evaluates the performance of two High Performance Embedded Computing architectures, the Imagine and Raw processors, on these kernels. The implementation results are compared with the measured performance of a conventional system based on the PowerPC with Altivec. The results show these processors exhibit significant improvements over conventional systems and that each architecture has its own strengths and weaknesses.

In this paper, we investigate a spectral-domain approach to estimating the interference covariance matrix used in space-time adaptive processing. Traditionally, an estimate of the interference covariance matrix is obtained by averaging the space-time covariance matrices of multiple range bins. Unfortunately, the spectral content of these data snapshots usually varies, which corrupts the covariance estimate for the desired range. We propose to use knowledge sources to identify angle-Doppler spectral regions having the same underlying scattering statistics. Then, we use real-time data to form a synthetic aperture radar image, which is inherently an estimate of non-moving ground clutter. We then average the SAR pixels within each homogeneous region. The resulting clutter power map is used, along with knowledge of the radar system and scenario geometry, to compute the interference covariance matrix. Using simulated data, we demonstrate the potential performance of such a technique, demonstrate its dependence on accurate space-time steering vectors, and provide an example of using data to compensate for imperfect knowledge.

The KASSPER project is a Defense Advanced Research Projects Agency (DARPA) program which has the goal of improving the performance of Ground Moving Target Indicator (GMTI) radar systems by incorporating external sources of knowledge into the signal processing chain. The KASSPER Real-Time Signal Processing Architecture is a radar system scheduling and signal processing framework that is being developed at Massachusetts Institute of Technology Lincoln Laboratory (MIT LL). This paper discusses the design of the architecture, knowledge handling issues, resource scheduling issues, the current state of the prototype implementation of the framework, and the current state of the project?s real-time processor testbed.

Internal Clutter Motion (ICM) places significant limits on the effectiveness of STAP clutter suppression techniques, for example in achieving the smallest minimum detectable velocity in GMTI radar surveillance applications. To simulate this effect with maximum fidelity, the required correlation must be impressed on the returns from individual scatterers during the construction of the data cube. Doing so can represent a substantial computational burden on the simulation process when a clutter scene is characterized by many millions of individual pixels. Such numbers are typical when using high fidelity SAR maps as the basis of the clutter model. The paper reports on a fast computational technique for wind-blown clutter simulation that works within a flexible data-generation system employing real high-fidelity IFSAR maps with co-registered elevation data as ground truth.