Research On Statistical Signal Detection In Hybrid Method

Signal Detection, which possesses a crucial application value to state economy and defense. is a very important topic in the world. On the basis of the study of the classic signal detection theory and the characteristics of sonic advanced subjects. the research on statistical signal detection in hybrid method has been presented in this dissertation. The main contribution of the thesis are shown as follows: 1 Discussing and reviewing the classic signal detection theory The principle of hypotheses testing and sonic decision criterions are reviewed and discussed. Much attention has been paid on the study of target fluctuating models(Swerling ones) and the performance of detection system in the different environment are formulized. 2 Researching the constant false-alarm ratio (CFAR) detection Based on the order statistics theory and CFAR method, a new kind of detection scheme and performance analysis method are put forward to detect the dependent target in nonhomogeneous environment. The detector based on the new scheme are build up in t~ o forms: OS ?CFAR with one reference window and (MX)OS ?CFAR ~vith two reference windows. Performance analyses and simulations have been made upon the two forms of detectors. The results of simulations show that this new detection scheme is valid on the signal detection in nonhomogeneous environment. 3 Studying the distributed fusion algorithm and analyzing its performance The different fusion systems are characterized and the concept of distributed detection is put forward. The analyses of optimal distributed decision fusion under the Bayes and Neyman?Pearson rules show the relation between the two optimal detection and likelihood ratio test. According to the above discussion, the fusion algorithm of multi-sensors system has been studied. Based on the known local detection rules :K/N~ AND~ OR a kind of information fusion optimization is proposed from algorithm and performance. The evaluation of detection performance under the different fusion rules and the observation ~vith Rayleigh distribution is given by theory method. Computer simulations are presented to confirm the validity of the algorithms. 4 Researching the applications of neural networks and signal processing in signal detection A new signal detection scheme is presented to solve the calculation difflculb of likelihood ratio in the actual environment with fluctuate signal and inference. This new method is based on neural network which can change the weight value of the network to fit the coniplex nonlinear function at suitable precision. This new detector is composed of three components, that are classification network. average procession and decision net~vork. This new method is compared with the traditional ones in detection performance to show that the...