Page 528 - Fundamentals of Radar Signal Processing

noncoherent integration and its behavior in Weibull clutter, are available in

Shor and Levanon (1991).

6.5.7 Additional CFAR Topics
“Adaptive CFAR” is the name given to a growing class of CFAR algorithms
17
designed to improve performance in nonhomogeneous clutter. Generally, they
dispense with the fixed CFAR window structure (usually with half of the cells

in each of the lead and lag windows). Instead they typically construct a
statistical test to determine if the reference cells span one clutter field or two,
i.e., whether or not the reference cell data are homogeneous. If not
homogeneous, the algorithms estimate not only the clutter statistics in each field,
but at which cell the transition from one to the other occurs (and therefore which
type of clutter is competing with the target in the CUT).
The basic approach to adaptive CFAR for nonhomogeneous clutter was
described in Finn (1986). The algorithm assumes a CFAR window of total

length N cells, including the CUT, that spans two clutter fields, i.e., two regions
with different statistical parameters. The clutter edge is presumed to occur
between samples M and M + 1; however, M is not known. Initially, it is also
assumed that the clutter follows the usual square-law detected, exponential
distribution. Let (M) denote the estimate of obtained by averaging the M

samples 1 through M forming the first clutter region, and (M) be the average
obtained from the N – M samples M + 1 through N forming the second region.
The algorithm starts by setting M = 1 and computing (1), which is the
“average” of only cell #1, and (1), the average of the remaining N - 1 cells.
This process is repeated for M = 2, …, N = 1. Thus, a pair of sample means
(M) and (M) are computed for each of the N – 1 possible transition points
between the two-clutter regions.

The next step is to choose the most likely transition point M . The maximum
t
likelihood estimate of this transition point is the value M of M that maximizes
t
the log-likelihood function (Finn, 1986):

(6.170)

Once M is identified it is also known if the CUT is in the first or second clutter
t
region. Standard CA CFAR, using the appropriate mean estimate and the number

of cells with which it is estimated, can then be applied. For example, if the CUT
is in the first region the threshold would be set according to

(6.171)