Page 282 - DSP Integrated Circuits
P. 282

6.10 Interpolator, Cont.                                             267















        Figure 6.58 Pole-zero configurations for the original and the pipelined filter obtained by
                   scattered look-ahead pipelining



















                  Figure 6.59 First-order scattered look-ahead pipelined structure




        data and coefficient word lengths a few bits longer than the original filter to main-
        tain the same round-off noise and appropriate pole-zero cancellation. Look-ahead
        pipelining can also be applied to a restricted set of nonlinear recursions [17,20-22].



        6.9.4 Synthesis of Fast Filter Structures
        An alternative, and possibly more efficient, approach is to use a computer to
        directly synthesize a transfer function of the desired form. The poles are placed
        symmetrically around the origin in the z-plane such that the recursive loops have
        the desired number of delay elements. In the method just discussed, the effect on
        the frequency response from some of the poles was canceled by placing zeros at the
        same locations. Now the zeros of the transfer function can be used more efficiently
        if they are placed on the unit circle instead. Their positions can be found by using
        a computer-based optimization procedure. Digital niters with this type of transfer
        function will have both high speed and low arithmetic work load.


        6.10 INTERPOLATOR, CONT.

        In order to schedule the operations and determine the necessary amount of compu-
        tational resources for the interpolator, we first need to determine the precedence
   277   278   279   280   281   282   283   284   285   286   287