Page 530 - Fundamentals of Radar Signal Processing

are normally distributed. The CFAR structure is a conventional cell-averaging

approach on the log data. The threshold is computed as follows:

(6.172)

This CFAR threshold calculation could clearly be combined with many of the
embellishments discussed earlier for CA CFAR, such as SOCA or GOCA rules
or censoring.
Because of the need to estimate two parameters, the CFAR loss is greater
with two parameter distributions than with the exponential distribution and in

fact can be very large, especially for small numbers of reference cells. For
–6
example, with = 0.9, = 10 , and N = 32 reference cells, the CFAR loss
using Eq. (6.172) in log-normal interference is approximately 13 dB (Schleher,
1977). From Fig. 6.21, a conventional CA CFAR in exponential interference

with the same detection statistics and window size has a CFAR loss of just
under 1 dB.
The same calculations are used to set the CFAR threshold in Weibull
clutter, though the specific values of α needed vary from the log-normal case.
Two proposed Weibull detectors, the so-called log-t detector and another based
on maximum likelihood estimates of the Weibull PDF parameters, have been
shown to be equivalent to Eq. (6.172) (Gandhi et al., 1995).

Order statistic CFARs have also been proposed for two-parameter clutter.
One example combines OS CFAR in each of the lead and lag windows with a
greatest-of logic to estimate the interference mean, and then uses the single
parameter Eq. (6.154) to set the threshold. Since the second (skewness)
parameter of the PDF is not estimated implicitly or explicitly, the multiplier α
must be made a function of the skewness, implying in turn that the skewness

parameter must be known to correctly set the threshold. Performance results
again suggest that choosing the order statistic k to be about 0.75N provides the
best performance against interferers and uncertainty in the skewness parameter
(Rifkin, 1994).
In Chap. 5 the technique of clutter mapping for detection of stationary or
slowly moving targets by ground-based, fixed-site radars when the competing
zero-Doppler clutter was not too strong was discussed. The threshold for each

range-angle cell was computed as a multiple of the measured clutter in the same
cell. The clutter measurement was obtained as a simple first-order recursive
filter of the form: