Page 214 - DSP Integrated Circuits
P. 214

5.5 Scaling of Signal Levels                                         199

            Figure 5.13 shows a multiplier
        with a noninteger coefficient that
        is imbedded in a network, NZ- The
        input signal level to this multiplier
        must be properly scaled. This; is
        done by multiplying all signals
        entering the network NZ by the
        scaling coefficient c, and multiply-
                                            Figure 5.12 Signal-flow graph with only one
        ing all signals leaving the network           critical overflow node
        by 1/c.
            Scaling multipliers must not
        affect the transfer function such
        that the poles and zeros are
        changed. Only the gain from the
        input of the filter to the critical
        node may be changed. If the scaling
        multipliers are part of a recursive
        loop, it is necessary that the effect
        on the transfer function is elimi-
        nated by choosing c so that c(l/c) = 1,
        using binary values for both coeffi-  Figure 5.13 Scaling of the signal level incident
        cients. It can be shown that the only       to a noninteger multiplier
                                ±n
        possible values are c = 2 . The
        scaling multipliers with associated quantizations may introduce additional round-off
        noise sources, but proper scaling will nevertheless improve the SNR. Additional scal-
        ing nodes may in some cases be introduced by the noninteger scaling coefficients.


        5.5.1 Safe Scaling
        One strategy used to choose the scaling coefficient can be derived in the following
        way: The signal in the scaling node is given by


        where f(ri) is the impulse response from the input of the filter to the critical over-
        flow node. The magnitude of the output signal is bounded by




        where


            In this scaling approach, we insert a scaling multiplier(s), c, between the
        input and the critical overflow node, as shown in Figure 5.13. The resulting
        impulse response becomes


            Now, we choose c so that
   209   210   211   212   213   214   215   216   217   218   219