Studie av CFAR-algoritmer i Weibull-fördelat klotter

Fredrik Sundin

Master Thesis, Report UPTEC 97121E, 51 pages, October 1997.

Abstract:

The use of five different constant false-alarm rate (CFAR) algorithms for target detection in Weibull-clutter has been studied. The algorithms considered were Cell Averaging (CA), Censored CA (CCA), Maximum-Likelihood (ML), Censored ML (CML) and Order Statistic (OS). The investigation has been carried out both by simulations and by applying the algorithms on physical data.

Two different ML algorithms for estimation of the shape parameter of the Weibull distribution, have also been studied. The shape parameter estimation was used to improve the robustness of the algorithms.

The investigation shows that unbiased estimation of the shape parameter can be performed with sufficiently low variance, based on a small number of samples. Investigations also show that the CA and ML algorithms respond poorly to a multiple-target situation. The censored algorithms perform much better. In the tests with physical data, the false-alarm rate was much higher than expected. The reason for this is that the probability of clutter samples with high power is higher than permitted by a Weibull-distribution.

Company:

Communicator C3 Consult

Thesis Advisor:

Mikael Sternad