The main objective of this dissertation work is to develop an unified CFAR structure and to derive uniformed formulas for detection and false alarm probabilities in closed form. By this approach, one can obtain performance of various CFAR detectors, particularly whose performance has not been analyzed yet.
There is a lot of CFAR detector for single pulse detection using the sliding window techniques. These CFAR detectors have their own fixed structure, respectively, i.e., fixed estimate of the background clutter-plus-noise level. A generalized CFAR model, termed the GOS CFAR detector which is a generalized structure of the OS CFAR processors, is proposed. Unified formulas for the false alarm and detection probabilities in homogeneous as well as nonhomogeneous situations are derived in closed form. By properly choosing the coefficients of the GOS filter, one can realize the OS CFAR processor, the TM CFAR processor, and CMLD as well as the CA CFAR processor as special cases. By this model, one can implement the $2^N-1$ kinds of CFAR processor by the combination of coefficients of the GOS filter.
In order to get rid of the capturing effect in multiple target situations, we also propose ACMLD, which is obtained by applying the GOS CFAR detector to this environments. ACMLD estimates adaptively the number of interferers in a reference window in short time interval compared to GCMLD when there exist a few interfering targets. This scheme can also be applied to the OS CFAR processor to determine the number of interferers and the detection threshold. Using the unified CFAR structure and the performance formulas, we compare the detection performance of ACMLD with those of other CFAR processors in multiple target situations. And then we compare the false alarm rates of various CFAR processors at clutter edge. The analytical results show that ACMLD reveals little CFAR loss compared with the CA CFAR processor in homogeneous situation. In multiple target situations, ACMLD can maintain good detection performance by estimating the number of interfering targets when there exist unknown number of interfering targets in the reference window.
Second, we analyze the performance of the GOS CFAR detector that employs M-pulse noncoherent integration in nonhomogeneous situation. We derive the unified formulas of the detection and false alarm probabilities for the GOS CFAR detector with noncoherent integration. Unified formulas of false alarm and the detection probabilities are derived in closed form. The analytical results show that the detection performance of the CA CFAR detector is better than any other CFAR detectors in homogeneous situation when we use the same number of integrated pulses. In nonhomogeneous clutter-plus-noise situation, CMLD has better detection performance than the other CFAR detectors, such as the OS, the TM, and the CA CFAR detectors, as long as one correctly censors the cells from clutter-plus-noise. On the other hand, analytical results of four types of Swerling target models show that for about $P_d>0.3$, the detection probability of the completely decorrelated target is better than that of the completely correlated target.
Finally, we give the analysis of the GOS CFAR detector for partially correlated target returns. We also derive the unified formulas of the detection and false alarm probabilities for the GOS CFAR detector for partially correlated target. By the proper choice of the GOS CFAR detector, we can obtain the detection performance of each CFAR detector selected, particularly whose performance has not been analyzed yet. The analytical results for a first order Markov exponentially correlation model of signal fluctuation are given for partially correlated target model as well as four Swerling target models. It can be shown that for about $P_d>0.3$, the more the correlation, the poorer the detectability. In order to raise the detectability of a target, one must adopt the pulse-to-pulse waveform diversity to decorrelate the target returns.