Page 536 - DSP Integrated Circuits
P. 536

11.19 FPT Processor, Cont.                                           521


        tively. In k\ the bits denoted BF and N s are set to 0. Hence, the addresses are com-
        puted by setting these bits to 1.


        11.19.4 Base Index Generator
        The base index generator must have three basic functions: a binary counter, a
        Gray encoder, and a unit to compute P(ra). The counter can be implemented as a
        bit-serial counter. This implementation requires a set of D flip-flops (a shift regis-
        ter) and a bit-serial half-adder.
           The translation of a binary number, x = x nx n_i--x\XQ, to its Gray coded equiv-
        alent, y = y ny n-iyiyo, can be described by
                                    © x • , -, if 0 < i < n



            This operation can also be performed bit-serially. The computation of P(m) can
        be described as

                     P(m) = (((...((m 0 )0m 1 )©m 2 )©m 3 )© ...)©m ?
           This computation can also be done bit-serially. The implementation of a device
        that increments m, computes the Gray code representation of m, computes P(m),
        and extends this by two bits to k\ is shown in Figure 11.56.
           The genera-
        tor consists of                ,  [v^                m - Gary encoded
        four parts: the
        counter, the Gray
        encoder, the unit
        that   computes        D
        P(ra), and the
        unit that extends
                           — »  ^ nut
        m by two bits.
                              HA
        The counter is
        formed   around
        the m register
        and   the  half-
        adder. To incre-             Figure 11.56 The base index generator
        ment m by one's
        we first use "Set carry" to set the carry D flip-flop of the bit-serial half-adder. The
        half-adder adds 1 to the LSB of m, and thus the LSB of m+1 is present at the out-
        put. Next we let 10 (the m register length) clock periods pass. Simultaneously with
        the incrementation of m we compute the Gray code representation of m using an
        XOR gate. The i and P(ra) outputs are used during the input and output phases.
           To extend m by two bits, the switch is initially set to position 1. The register is
        clocked to the position where the extra bits should be inserted. The switch is then
        set to position 2 and two zeros are entered during two clock cycles. After this, the
        switch is set to 3 and the remaining part of the base index is clocked out.
           After 10 clock periods in the FFT computation phase, mg is checked. If mg = 1,
                                   8
        it means that m has reached 2  = 256, and we have computed all butterflies in a
        stage. When this happens, m is reset, and a done signal is sent to the Stage PE.
   531   532   533   534   535   536   537   538   539   540   541