Principles of Modern Radar - Volume 2 1891121537

10.1 Introduction 455It is known that maximizing SINR at the output of the space-time processor equivalentlymaximizes probability of detection for a fixed false alarm rate [3]. Most STAPperformance metrics thus focus on characterizing algorithm impacts on SINR. SINR lossis generally spatially and temporally varying and characterizes SINR degradation due tothe presence of ground clutter.Reduced-dimension STAP (RD-STAP) techniques are common and use frequencydomaintransforms and selection operators to reduce the dimensionality of the adaptiveprocessor. Usually a small loss in performance relative to the joint space-time filter resultswhen using RD-STAP. However, its benefits include significantly reduced computationalloading and more efficient use of training data, often leading to more effective adaptivefilter implementation.Covariance estimation is a key consideration in the practical implementation of STAP.Heterogeneous or nonstationary clutter violates the requirement that training data usedto estimate the unknown covariance matrix be homogeneous with respect to the nullhypothesiscondition of the cell under test. Degraded output SINR and an increase in thedetector’s false alarm rate are the typical consequences. Robust STAP methods must makeprovision for heterogeneous or nonstationary clutter.10.1.2 OrganizationThe subsequent sections introduce the reader to common mathematical operations andnotation used throughout the chapter. STAP discussion relies extensively on the use oflinear algebra to describe signal and filter characteristics.After this preliminary discussion, this chapter develops spatial, temporal (Doppler),and space-time signal representation. Additionally, the early sections of this chapter introducethe reader to spatial beamforming, Doppler processing, and two-dimensional matchedfiltering, which are critical components of STAP. The space-time signal formulation is thenused to develop a ground clutter signal model and corresponding clutter covariance matrix.The middle sections of this chapter provide a general framework for space-time processingand introduce key STAP performance metrics, including probability of detection,probability of false alarm, SINR loss factors, and improvement factor. SINR loss is perhapsthe most commonly used STAP metric, and two variants are typical: clairvoyant SINRloss, which compares the SINR at the output of a space-time filter with known weights tothe achievable output SNR; and adaptive SINR loss, which compares the adaptive filteroutput SINR to the optimal (known weights) filter SINR. After a discussion of metrics,several methods used to calculate the optimal weight vector are given, including the maximumSINR filter, the minimum variance (MV) beamformer, and the generalized sidelobecanceller (GSC).Most STAP discussion centers on ways to approximate the unknown, space-timeweight vector and implement the filter. Two fundamental approaches exist and are discussedin the last third of the chapter. The first, and most popular, approach is known asreduced-dimension STAP, which applies deterministic filters and selection operations toreduce the dimensionality of the STAP filter and therefore the computational burden andrequirements on training sample support. The second approach is reduced-rank STAP, aweight calculation method based on eigendecomposition of the interference-plus-noisecovariance matrix. In the most general sense it results in a projection of the data to suppressinterference, followed by a matched filtering operation. Since reduced-rank STAP