Distributed Cfar Detection Theory Research

The distributed CFAR detection is studied in this paper.First, the present research situation and problem in distributed CFAR detection are discussed after briefly reviewing the previous research. Then the research content of this dissertation is introduced.The nonparametric distributed detection is studied in chapter 2. Firstly, when the weights of LDs(local detector) are decided, the threshold in fusion centre is solved by an on-line method. Based on this method, a novel self-learning method is proposed to get the optimum local detector weights in NP sense. The two methods can be united easily and they can be used in other situations. After analysed the performance of nonparametric detector particularly, three new fusion rules are proposed based on local statistics, their asymptotic relative efficiency is studied.The distributed CFAR detection that fusion centre receives local decisions is studied in chapter 3. In the hypothesis that the clutter samplings are independent identically distributed random variables and that follow an exponential distribution, a general numerical method is proposed for parallel and serial network, and it can be used for any fusion rule and other clutter distribution. The system performance is analysed in homogeneous and nonhomogeneous background.The distributed CFAR detection that fusion centre receives local statistics is studied in chapter 4. In the case of same clutter power in LDs, based on the fusion rule proposed in [171], a novel method is proposed to solve the distributed system included any kind of LDs by the property of Laplace transform. This method overcomes the shortcoming of supposing the same local SNR in other method. Under the conditions that the clutter power of LDs is not same, a lot of emulating results are given, and part of available conclusions are find based on these results.In chapter 5 the distributed CFAR detection is studied when LDs are correlated. Due to the correlation of local decisions is not known, empirical estimation is adopt to resolve this problem, the relationship of the training sample size and the estimated confidence is analyzed. When the fusion centre receives local statistics, the probability of detection is studied on the situation that only the signal is correlated and the clutter is independent.The problems and content needed further research are pointed out in chapter 6.