Author: Michael Richard Riedl

Publisher:

Published: 2016

Total Pages: 156

ISBN-13:

Increasing the resolution of radar imaging and ground moving target indicator (GMTI) systems puts a stress on both hardware and processing limitations. Hardware must be able to handle the transfer of the large amounts of data generated. Additionally, the processing must be robust to any heterogeneity of the data that is introduced by collecting returns from large swaths. This dissertation presents system architectures and knowledge-aided processing techniques to combat the large data rates and data heterogeneity. Ground moving target indicator radar techniques for airborne platforms require spatial and Doppler signal diversity for separating the returns of moving targets from the returns of ground clutter. The traditional use of multiple receive antennas for jointly imaging a scene and detecting moving objects is prohibited by the system bottleneck at the data down-link. We present a frequency division multiple access, multiple-transmit single-receive radar architecture, with associated waveform design and data processing procedure. The proposed approach is demonstrated to jointly provide imaging and GMTI modalities while maintaining the data rate to that of a single antenna imaging system. Heterogeneity of the radar backscatter data degrades detection performance by biasing statistical parameters estimated from the data. A GMTI processing technique, known as space-time adaptive processing (STAP), requires estimation of the space-time covariance of the clutter for use in a generalized likelihood ratio test. Consequently, the performance of STAP is related to the quality of the estimated clutter covariance matrix; however, in practice it is common for the data to be limited, contaminated, and heterogeneous. In this dissertation, we introduce and evaluate two estimators for the clutter covariance and a purely Bayesian detection scheme. A Bayesian model is postulated for the angle/Doppler scene to incorporate approximate prior knowledge of the terrain height and the platform kinematics. Posterior probabilities computed using the model are then used to either estimate a covariance matrix or directly report posterior probabilities of the presence of a target. The approach is a novel means for incorporating operational knowledge into GMTI processing and admits low-complexity algorithmic implementation via recent advances in Bayesian message passing algorithms. In the second covariance estimator, a regularized shrinkage approach is proposed, whereby prior knowledge is expressed through an elastic net regularization penalty on a minimum expected squared error estimation cost. The regularized shrinkage estimator is shown to coincide with a minimax robust covariance estimator and offers simplicity in modeling and computation that may facilitate use by practitioners. In the third approach, the Bayesian model is augmented to jointly estimate calibration parameters for unknown antenna phases and detect moving targets. The performances of the proposed estimators and detectors are evaluated using the KASSPER I dataset. We conclude that the proposed approaches extend the state-of-the-art to provide reliable detection performance when the training data is limited to a number of range bins less than the rank of the true covariance matrix. Further, when presented no training data, the Bayesian approach is shown to maintain performance using only the data under test. Finally, the purely Bayesian detection approach, when combined with antenna calibration, is observed to provide enhanced resolution, allowing reliable detection of multiple targets within a single range bin not achievable with traditional STAP.