Abstract：Two modified censored mean (MCM) and modified unbiased censored mean (MUCM) CFAR algorithms are proposed. Both split the reference window into two sub-windows which apply CM or UCM method to create two local noise power estimations, the mean value of them are taken to set an adaptive threshold. Both use the automatic censoring technique proposed by He You (1994). Under Swerling II assumption, considering Rayleigh distributed noise and single-pulse square-law detection, the analytic expressions of Pfa, Pdand ADT of both are derived. By comparison with other schemes, the results show that their performance are evidently superior to that of GOSCA and OS in homogeneous background and in multiple target situations, in the case of IL=4, IR=2, the CFAR loss of MCM is improved by 2 dB relative to that of OS, that of MUCM is improved by 1.5 dB over OS. The performance of MCM is slightly better than that of CM, the performance of MUCM is the nearly same as that of UCM, but their sample sorting time is less than half of that of CM, UCM and OS.
1997, Vol. 19 
