Performance Analysis Of Adaptive Detection Algorithms And Their Applications

In modern wars, radars are confronted with four main threats: penetration in lowand very low altitude, stealth aircrafts, electronic jamming and high-speed antiradiationmissiles. When detecting targets in low and very low altitude, radars receive clutter fromground/sea. This clutter is usually non-Gaussian when observed by high resolution radarsoperating at low grazing angles. Due to being painted wave-absorbing material and theirspecial shapes, stealth aircrafts appear with considerably small radar cross sections, andthus are difficult to detect. Electronic jamming can make hostile radars blind, and createfalse targets in order to cause hostile radars to fail in properly alarming. How to im-prove radars’ capacity of detecting (weak) targets in non-Gaussian clutter or electronicjamming is an urgent problem. This dissertation examines adaptive detection algorithmsin the non-Gaussian background, and proposes methods for improving the detection per-formance of traditional detection algorithms by using polarization diversity and multiple-input multiple-output (MIMO) techniques, and meanwhile discusses the problem of de-tecting distributed targets in the presence of jamming. The main contributions in this workfall into four aspects:Adaptive subspace detection algorithms in partially homogeneous environmentsThe adaptive subspace detection algorithms for unfluctuating and fluctuating targetmodels are studied in partially homogeneous environments, where the noise covariancematrix is assumed unknown, and is estimated with a set of secondary data. It is proved thatfor the nonfluctuating target model, the adaptive detector designed within one step is thesame as that developed with the two-step procedure. This conclusion does not hold true inhomogeneous environments. In particular, the probability of false alarm of the adaptivesubspace detector is derived, which is expressed as a sum of elementary functions. Itcan be used to simplify the selection of detection threshold. In addition, the approximateand exact detection probabilities of the adaptive subspace detector are also derived. Ifhigh accuracy in the detection probability is not required when evaluating the detectionperformance, the approximate expression for the detection probability which is simpleand easy to calculate is suggested. Otherwise, the exact expression for the detectionprobability is recommended.Performance enhancement of adaptive detection algorithmsIt is proved that all the detection probabilities of four adaptive detectors are strictlyincreasing functions with respect to some parameter which is related to the characteristicsof the transmitted pulses. By optimally selecting the transmitted pulses to maximize thisparameter, the maximum detection performance can be obtained. In polarized radar sys- tems, the model of the received data is first established. Then, the proposed optimizationalgorithm is used to optimally choose the transmitted pulses, which results in a significantdetection performance gain.Adaptive detection algorithms with MIMO radarWhen the noise covariance matrix is known, the detection performance of MIMO-matched filter detector is analyzed, and the probabilities of false alarm and detection ofthis detector are derived. When the noise covariance matrix is unknown, the generalizedlikelihood ratio test is employed to obtain the MIMO-generalized likelihood ratio testdetector, and the probability of false alarm of this detector is derived. The derived prob-abilities of false alarm can be used to obtain the detection threshold of the two detectorsfor a preassigned probability of false alarm.Distributed target detection algorithms in the presence of interferenceThe performance of distributed target detection algorithms are analyzed in the pres-ence of subspace interference plus white Gaussian noise. When the noise power is knownand unknown, the first kind of distributed matched subspace detector and the second kindof distributed matched subspace detector are adopted to detect targets, respectively. Par-ticularly, the probabilities of false alarm and detection of the two detectors for both unfluc-tuating and fluctuating target models are derived. Both of the detectors have the constantfalse alarm property with respect to the interference, and the second detector also pos-sesses the constant false alarm property against noise power. In addition, the interferencecan be completely eliminated by using the two detectors, and the detection performanceof the two detectors is independent of the interference power.