Page 50 - Phase Space Optics Fundamentals and Applications
P. 50

Wigner Distribution in Optics   31


               From the latter equality we conclude that the value of the Wigner
               distribution at some phase-space point (r, q) is related to the values
                                                  1
               of all those pairs of points (r ± r , q ± q ) for which (r, q) is the

                                           1
                                           2      2
               midpoint. Using, as we generally do, the analytic signal f (r) instead
                              1
               of the real signal [ f (r) + f (r)] avoids the cross-terms that otherwise
                                      ∗
                              2
               would automatically appear around q = 0.
                 The requirement of removing cross-terms without seriously affect-
               ing the auto-terms has led to the Cohen class of bilinear signal rep-
               resentations. All members C f (r, q) of this class can be generated by
               a convolution (for both r and q) of the Wigner distribution with an
               appropriate kernel K(r, q):
                       C f (r, q) = K(r, q)∗ ∗W f (r, q)
                                      r q

                              =     K(r − r o , q − q ) W f (r o , q ) dr o dq  (1.76)
                                                 o        o      o
               Note that a convolution keeps the important property of shift covari-
               ance! After Fourier transforming the latter equation, we are led to an
               equation in the “ambiguity domain,” and the convolution becomes a
               product:
                                           ¯
                                ¯


                                C f (r , q ) = K(r , q )A f (r , q )  (1.77)




               with
                                ¯
                               C f (r , q ) = F[C f (r, q)](r , q )  (1.78a)








                                A f (r , q ) = F[W f (r, q)](r , q )  (1.78b)
                                 ¯




                                K(r , q ) = F[K(r, q)](r , q )     (1.78c)
               The product form (1.77) offers an easy way in the design of appropriate
               kernels.
                 Again, cf. Fig. 1.1, we position the different signal and kernel rep-
               resentations at the corners of a rectangle, see Fig. 1.4. For complete-
               ness we have also introduced the kernels R(r 1 , r 2 ) and ¯ R(q , q ) that
                                                                 1  2
                                                     ∗
               operate on the product   f (r 1 , r 2 ) = f (r 1 ) f (r 2 ) and ¯   f (q , q ) =
                                                                  1
                                                                     2
                ¯
                f (q ) f (q ), respectively, by means of a convolution for r or q. Again,
                     ¯ ∗
                   1    2
               we have Fourier transformations along the sides of the rectangle, and
               we readily see that the kernel K(r, q) is related to the kernels R(r 1 , r 2 )
               and ¯ R(q , q )as
                      1  2


                                                           t
                                        1
                                              1
                        K(r, q) =  R r + r , r − r exp(−i2 q r ) dr    (1.79a)
                                        2     2

                                        1       1
                                                           t


                        K(r, q) =  ¯ R q + q , q − q exp(i2 r q ) dq    (1.79b)
                                        2       2
   45   46   47   48   49   50   51   52   53   54   55