| 1 |
36 |
dgisselq |
////////////////////////////////////////////////////////////////////////////////
|
| 2 |
|
|
//
|
| 3 |
|
|
// Filename: laststage.v
|
| 4 |
|
|
//
|
| 5 |
|
|
// Project: A General Purpose Pipelined FFT Implementation
|
| 6 |
|
|
//
|
| 7 |
|
|
// Purpose: This is part of an FPGA implementation that will process
|
| 8 |
|
|
// the final stage of a decimate-in-frequency FFT, running
|
| 9 |
|
|
// through the data at two samples per clock. If you notice from the
|
| 10 |
|
|
// derivation of an FFT, the only time both even and odd samples are
|
| 11 |
|
|
// used at the same time is in this stage. Therefore, other than this
|
| 12 |
|
|
// stage and these twiddles, all of the other stages can run two stages
|
| 13 |
|
|
// at a time at one sample per clock.
|
| 14 |
|
|
//
|
| 15 |
|
|
// Operation:
|
| 16 |
|
|
// Given a stream of values, operate upon them as though they were
|
| 17 |
|
|
// value pairs, x[2n] and x[2n+1]. The stream begins when n=0, and ends
|
| 18 |
|
|
// when n=1. When the first x[0] value enters, the synchronization
|
| 19 |
|
|
// input, i_sync, must be true as well.
|
| 20 |
|
|
//
|
| 21 |
|
|
// For this stream, produce outputs
|
| 22 |
|
|
// y[2n ] = x[2n] + x[2n+1], and
|
| 23 |
|
|
// y[2n+1] = x[2n] - x[2n+1]
|
| 24 |
|
|
//
|
| 25 |
|
|
// When y[0] is output, a synchronization bit o_sync will be true as
|
| 26 |
|
|
// well, otherwise it will be zero.
|
| 27 |
|
|
//
|
| 28 |
|
|
//
|
| 29 |
|
|
// In this implementation, the output is valid one clock after the input
|
| 30 |
|
|
// is valid. The output also accumulates one bit above and beyond the
|
| 31 |
|
|
// number of bits in the input.
|
| 32 |
|
|
//
|
| 33 |
|
|
// i_clk A system clock
|
| 34 |
|
|
// i_reset A synchronous reset
|
| 35 |
|
|
// i_ce Circuit enable--nothing happens unless this line is high
|
| 36 |
|
|
// i_sync A synchronization signal, high once per FFT at the start
|
| 37 |
|
|
// i_left The first (even) complex sample input. The higher order
|
| 38 |
|
|
// bits contain the real portion, low order bits the
|
| 39 |
|
|
// imaginary portion, all in two's complement.
|
| 40 |
|
|
// i_right The next (odd) complex sample input, same format as
|
| 41 |
|
|
// i_left.
|
| 42 |
|
|
// o_left The first (even) complex output.
|
| 43 |
|
|
// o_right The next (odd) complex output.
|
| 44 |
|
|
// o_sync Output synchronization signal.
|
| 45 |
|
|
//
|
| 46 |
|
|
// Creator: Dan Gisselquist, Ph.D.
|
| 47 |
|
|
// Gisselquist Technology, LLC
|
| 48 |
|
|
//
|
| 49 |
|
|
////////////////////////////////////////////////////////////////////////////////
|
| 50 |
|
|
//
|
| 51 |
|
|
// Copyright (C) 2015-2018, Gisselquist Technology, LLC
|
| 52 |
|
|
//
|
| 53 |
|
|
// This program is free software (firmware): you can redistribute it and/or
|
| 54 |
|
|
// modify it under the terms of the GNU General Public License as published
|
| 55 |
|
|
// by the Free Software Foundation, either version 3 of the License, or (at
|
| 56 |
|
|
// your option) any later version.
|
| 57 |
|
|
//
|
| 58 |
|
|
// This program is distributed in the hope that it will be useful, but WITHOUT
|
| 59 |
|
|
// ANY WARRANTY; without even the implied warranty of MERCHANTIBILITY or
|
| 60 |
|
|
// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
| 61 |
|
|
// for more details.
|
| 62 |
|
|
//
|
| 63 |
|
|
// You should have received a copy of the GNU General Public License along
|
| 64 |
|
|
// with this program. (It's in the $(ROOT)/doc directory, run make with no
|
| 65 |
|
|
// target there if the PDF file isn't present.) If not, see
|
| 66 |
|
|
// <http://www.gnu.org/licenses/> for a copy.
|
| 67 |
|
|
//
|
| 68 |
|
|
// License: GPL, v3, as defined and found on www.gnu.org,
|
| 69 |
|
|
// http://www.gnu.org/licenses/gpl.html
|
| 70 |
|
|
//
|
| 71 |
|
|
//
|
| 72 |
|
|
////////////////////////////////////////////////////////////////////////////////
|
| 73 |
|
|
//
|
| 74 |
|
|
//
|
| 75 |
|
|
`default_nettype none
|
| 76 |
|
|
//
|
| 77 |
|
|
module laststage(i_clk, i_reset, i_ce, i_sync, i_left, i_right, o_left, o_right, o_sync);
|
| 78 |
|
|
parameter IWIDTH=16,OWIDTH=IWIDTH+1, SHIFT=0;
|
| 79 |
|
|
input i_clk, i_reset, i_ce, i_sync;
|
| 80 |
|
|
input [(2*IWIDTH-1):0] i_left, i_right;
|
| 81 |
|
|
output reg [(2*OWIDTH-1):0] o_left, o_right;
|
| 82 |
|
|
output reg o_sync;
|
| 83 |
|
|
|
| 84 |
|
|
wire signed [(IWIDTH-1):0] i_in_0r, i_in_0i, i_in_1r, i_in_1i;
|
| 85 |
|
|
assign i_in_0r = i_left[(2*IWIDTH-1):(IWIDTH)];
|
| 86 |
|
|
assign i_in_0i = i_left[(IWIDTH-1):0];
|
| 87 |
|
|
assign i_in_1r = i_right[(2*IWIDTH-1):(IWIDTH)];
|
| 88 |
|
|
assign i_in_1i = i_right[(IWIDTH-1):0];
|
| 89 |
|
|
wire [(OWIDTH-1):0] o_out_0r, o_out_0i,
|
| 90 |
|
|
o_out_1r, o_out_1i;
|
| 91 |
|
|
|
| 92 |
|
|
|
| 93 |
|
|
// Handle a potential rounding situation, when IWIDTH>=OWIDTH.
|
| 94 |
|
|
|
| 95 |
|
|
|
| 96 |
|
|
|
| 97 |
|
|
// As with any register connected to the sync pulse, these must
|
| 98 |
|
|
// have initial values and be reset on the i_reset signal.
|
| 99 |
|
|
// Other data values need only restrict their updates to i_ce
|
| 100 |
|
|
// enabled clocks, but sync's must obey resets and initial
|
| 101 |
|
|
// conditions as well.
|
| 102 |
|
|
reg rnd_sync, r_sync;
|
| 103 |
|
|
|
| 104 |
|
|
initial rnd_sync = 1'b0; // Sync into rounding
|
| 105 |
|
|
initial r_sync = 1'b0; // Sync coming out
|
| 106 |
|
|
always @(posedge i_clk)
|
| 107 |
|
|
if (i_reset)
|
| 108 |
|
|
begin
|
| 109 |
|
|
rnd_sync <= 1'b0;
|
| 110 |
|
|
r_sync <= 1'b0;
|
| 111 |
|
|
end else if (i_ce)
|
| 112 |
|
|
begin
|
| 113 |
|
|
rnd_sync <= i_sync;
|
| 114 |
|
|
r_sync <= rnd_sync;
|
| 115 |
|
|
end
|
| 116 |
|
|
|
| 117 |
|
|
// As with other variables, these are really only updated when in
|
| 118 |
|
|
// the processing pipeline, after the first i_sync. However, to
|
| 119 |
|
|
// eliminate as much unnecessary logic as possible, we toggle
|
| 120 |
|
|
// these any time the i_ce line is enabled, and don't reset.
|
| 121 |
|
|
// them on i_reset.
|
| 122 |
|
|
// Don't forget that we accumulate a bit by adding two values
|
| 123 |
|
|
// together. Therefore our intermediate value must have one more
|
| 124 |
|
|
// bit than the two originals.
|
| 125 |
|
|
reg signed [(IWIDTH):0] rnd_in_0r, rnd_in_0i;
|
| 126 |
|
|
reg signed [(IWIDTH):0] rnd_in_1r, rnd_in_1i;
|
| 127 |
|
|
|
| 128 |
|
|
always @(posedge i_clk)
|
| 129 |
|
|
if (i_ce)
|
| 130 |
|
|
begin
|
| 131 |
|
|
//
|
| 132 |
|
|
rnd_in_0r <= i_in_0r + i_in_1r;
|
| 133 |
|
|
rnd_in_0i <= i_in_0i + i_in_1i;
|
| 134 |
|
|
//
|
| 135 |
|
|
rnd_in_1r <= i_in_0r - i_in_1r;
|
| 136 |
|
|
rnd_in_1i <= i_in_0i - i_in_1i;
|
| 137 |
|
|
//
|
| 138 |
|
|
end
|
| 139 |
|
|
|
| 140 |
|
|
convround #(IWIDTH+1,OWIDTH,SHIFT) do_rnd_0r(i_clk, i_ce,
|
| 141 |
|
|
rnd_in_0r, o_out_0r);
|
| 142 |
|
|
|
| 143 |
|
|
convround #(IWIDTH+1,OWIDTH,SHIFT) do_rnd_0i(i_clk, i_ce,
|
| 144 |
|
|
rnd_in_0i, o_out_0i);
|
| 145 |
|
|
|
| 146 |
|
|
convround #(IWIDTH+1,OWIDTH,SHIFT) do_rnd_1r(i_clk, i_ce,
|
| 147 |
|
|
rnd_in_1r, o_out_1r);
|
| 148 |
|
|
|
| 149 |
|
|
convround #(IWIDTH+1,OWIDTH,SHIFT) do_rnd_1i(i_clk, i_ce,
|
| 150 |
|
|
rnd_in_1i, o_out_1i);
|
| 151 |
|
|
|
| 152 |
|
|
|
| 153 |
|
|
// Prior versions of this routine did not include the extra
|
| 154 |
|
|
// clock and register/flip-flops that this routine requires.
|
| 155 |
|
|
// These are placed in here to correct a bug in Verilator, that
|
| 156 |
|
|
// otherwise struggles. (Hopefully this will fix the problem ...)
|
| 157 |
|
|
always @(posedge i_clk)
|
| 158 |
|
|
if (i_ce)
|
| 159 |
|
|
begin
|
| 160 |
|
|
o_left <= { o_out_0r, o_out_0i };
|
| 161 |
|
|
o_right <= { o_out_1r, o_out_1i };
|
| 162 |
|
|
end
|
| 163 |
|
|
|
| 164 |
|
|
initial o_sync = 1'b0; // Final sync coming out of module
|
| 165 |
|
|
always @(posedge i_clk)
|
| 166 |
|
|
if (i_reset)
|
| 167 |
|
|
o_sync <= 1'b0;
|
| 168 |
|
|
else if (i_ce)
|
| 169 |
|
|
o_sync <= r_sync;
|
| 170 |
|
|
|
| 171 |
|
|
endmodule
|