Page 532 - Fundamentals of Radar Signal Processing
P. 532

on  clutter  samples  taken  from  the  CUT  but  on  previous  time  intervals.  Thus,

               spatial  homogeneity  is  not  required,  but  homogeneity  in  time,  i.e.,  statistical
               stationarity,  is.  Both  types  of  CFAR  require  that  the  clutter  samples  be
               uncorrelated (in space or Doppler for CA CFAR, in time for clutter mapping)
               for  the  analyses  given  here  to  be  valid.  Finally, Eq.  (6.176)  shows  that  the
               threshold is based on an infinite, weighted sum of previous samples rather than
               the finite sums of CA CFAR and its variants.

                     It is also possible to estimate the clutter mapping threshold using a simple
               average  of  a  finite  number  of  previous  measurements.  One  version  of  the
               moving  target  detector  (MTD)  is  said  to  have  used  an  eight-scan  average,
               covering about 32 seconds of data (Skolnik, 2001). In this case, the CA CFAR
               analyses would apply to the clutter map as well. However, for computational
               simplicity  the  recursive  filter  of Eq. (6.173) is most often used. The average
               probabilities of false alarm and detection when the threshold is computed with

               the recursive scheme are (Nitzberg, 1986)






                                                                                                     (6.177)







                                                                                                     (6.178)

               These  formulas  are  slow  to  converge  in  practice;  a  more  rapidly  converging
               variation is given in Levanon (1988). Again, α can be varied to generate curves

               of    versus    and  CFAR  loss  can  then  be  determined  by  comparing  these
               curves to the case where the interference is known exactly. The case γ = 0 in
               fact corresponds to the ideal case. For every increase of 0.2 in γ, the CFAR loss
               increases approximately 3 dB. A crude approximate formula for the CFAR loss
               is (Taylor, 1990)








                                                                                                     (6.179)

                     This  formula  is  of  limited  accuracy,  especially  for  large γ,  but  may  be
               useful  for  rough  calculations.  Another  implementation  of  the  MTD  used  a
               recursive filter as described previously with a feedback coefficient of γ = 7/8
               (Nathanson, 1991). A rational value of γ with a denominator that is a power of
               two was particularly amenable to the fixed point implementations used in early
               versions of the MTD, since the division by the denominator value (eight in this
   527   528   529   530   531   532   533   534   535   536   537