proxmark3/fpga/hi_read_rx_xcorr.v
Andreas Dröscher 83a4259ddc change: fixed xcorrelation for strong signal
The initial code assumed phase shift modulation only. Lately,
xcorrelation is also used for load modulation. But the initial
the assumption that 11 bits are enough isn't true for load
modulation.

This change extends the registers by 2 bits and compresses the
uper bits to preserve the sensitivity on the lower end.
2018-08-10 02:36:04 +02:00

197 lines
6.4 KiB
Verilog

//-----------------------------------------------------------------------------
//
// Jonathan Westhues, April 2006
//-----------------------------------------------------------------------------
module hi_read_rx_xcorr(
pck0, ck_1356meg, ck_1356megb,
pwr_lo, pwr_hi, pwr_oe1, pwr_oe2, pwr_oe3, pwr_oe4,
adc_d, adc_clk,
ssp_frame, ssp_din, ssp_dout, ssp_clk,
cross_hi, cross_lo,
dbg,
xcorr_is_848, snoop, xcorr_quarter_freq
);
input pck0, ck_1356meg, ck_1356megb;
output pwr_lo, pwr_hi, pwr_oe1, pwr_oe2, pwr_oe3, pwr_oe4;
input [7:0] adc_d;
output adc_clk;
input ssp_dout;
output ssp_frame, ssp_din, ssp_clk;
input cross_hi, cross_lo;
output dbg;
input xcorr_is_848, snoop, xcorr_quarter_freq;
// Carrier is steady on through this, unless we're snooping.
assign pwr_hi = ck_1356megb & (~snoop);
assign pwr_oe1 = 1'b0;
assign pwr_oe3 = 1'b0;
assign pwr_oe4 = 1'b0;
reg [2:0] fc_div;
always @(negedge ck_1356megb)
fc_div <= fc_div + 1;
(* clock_signal = "yes" *) reg adc_clk; // sample frequency, always 16 * fc
always @(ck_1356megb, xcorr_is_848, xcorr_quarter_freq, fc_div)
if (xcorr_is_848 & ~xcorr_quarter_freq) // fc = 847.5 kHz, standard ISO14443B
adc_clk <= ck_1356megb;
else if (~xcorr_is_848 & ~xcorr_quarter_freq) // fc = 423.75 kHz
adc_clk <= fc_div[0];
else if (xcorr_is_848 & xcorr_quarter_freq) // fc = 211.875 kHz
adc_clk <= fc_div[1];
else // fc = 105.9375 kHz
adc_clk <= fc_div[2];
// When we're a reader, we just need to do the BPSK demod; but when we're an
// eavesdropper, we also need to pick out the commands sent by the reader,
// using AM. Do this the same way that we do it for the simulated tag.
reg after_hysteresis, after_hysteresis_prev, after_hysteresis_prev_prev;
reg [11:0] has_been_low_for;
always @(negedge adc_clk)
begin
if(& adc_d[7:0]) after_hysteresis <= 1'b1;
else if(~(| adc_d[7:0])) after_hysteresis <= 1'b0;
if(after_hysteresis)
begin
has_been_low_for <= 7'b0;
end
else
begin
if(has_been_low_for == 12'd4095)
begin
has_been_low_for <= 12'd0;
after_hysteresis <= 1'b1;
end
else
has_been_low_for <= has_been_low_for + 1;
end
end
// Let us report a correlation every 4 subcarrier cycles, or 4*16=64 samples,
// so we need a 6-bit counter.
reg [5:0] corr_i_cnt;
// And a couple of registers in which to accumulate the correlations. Since
// load modulation saturates the ADC we have to use a large enough register
// 32 * 255 = 8160, which can be held in 13 bits. Add 1 bit for sign.
//
// The initial code assumed a phase shift of up to 25% and the accumulators were
// 11 bits (32 * 255 * 0,25 = 2040), we will pack all bits exceeding 11 bits into
// MSB. This prevents under/-overflows but preserves sensitivity on the lower end.
reg signed [13:0] corr_i_accum;
reg signed [13:0] corr_q_accum;
// we will report maximum 8 significant bits
reg signed [7:0] corr_i_out;
reg signed [7:0] corr_q_out;
// clock and frame signal for communication to ARM
reg ssp_clk;
reg ssp_frame;
always @(negedge adc_clk)
begin
corr_i_cnt <= corr_i_cnt + 1;
end
// ADC data appears on the rising edge, so sample it on the falling edge
always @(negedge adc_clk)
begin
// These are the correlators: we correlate against in-phase and quadrature
// versions of our reference signal, and keep the (signed) result to
// send out later over the SSP.
if(corr_i_cnt == 6'd0)
begin
// send 10 bits of tag signal, 4 MSBs are stuffed into 2 MSB
if(~corr_i_accum[13])
corr_i_out <= {corr_i_accum[13],
corr_i_accum[12] | corr_i_accum[11] | corr_i_accum[10],
corr_i_accum[12] | corr_i_accum[11] | corr_i_accum[9],
corr_i_accum[8:4]};
else
corr_i_out <= {corr_i_accum[13],
corr_i_accum[12] & corr_i_accum[11] & corr_i_accum[10],
corr_i_accum[12] & corr_i_accum[11] & corr_i_accum[9],
corr_i_accum[8:4]};
if(~corr_q_accum[13])
corr_q_out <= {corr_q_accum[13],
corr_q_accum[12] | corr_q_accum[11] | corr_q_accum[10],
corr_q_accum[12] | corr_q_accum[11] | corr_q_accum[9],
corr_q_accum[8:4]};
else
corr_q_out <= {corr_q_accum[13],
corr_q_accum[12] & corr_q_accum[11] & corr_q_accum[10],
corr_q_accum[12] & corr_q_accum[11] & corr_q_accum[9],
corr_q_accum[8:4]};
if(snoop)
begin
// replace LSB with 1 bit reader signal
corr_i_out[0] <= after_hysteresis_prev_prev;
corr_q_out[0] <= after_hysteresis_prev;
after_hysteresis_prev_prev <= after_hysteresis;
end
corr_i_accum <= adc_d;
corr_q_accum <= adc_d;
end
else
begin
if(corr_i_cnt[3])
corr_i_accum <= corr_i_accum - adc_d;
else
corr_i_accum <= corr_i_accum + adc_d;
if(corr_i_cnt[3] == corr_i_cnt[2]) // phase shifted by pi/2
corr_q_accum <= corr_q_accum + adc_d;
else
corr_q_accum <= corr_q_accum - adc_d;
end
// The logic in hi_simulate.v reports 4 samples per bit. We report two
// (I, Q) pairs per bit, so we should do 2 samples per pair.
if(corr_i_cnt == 6'd32)
after_hysteresis_prev <= after_hysteresis;
// Then the result from last time is serialized and send out to the ARM.
// We get one report each cycle, and each report is 16 bits, so the
// ssp_clk should be the adc_clk divided by 64/16 = 4.
if(corr_i_cnt[1:0] == 2'b10)
ssp_clk <= 1'b0;
if(corr_i_cnt[1:0] == 2'b00)
begin
ssp_clk <= 1'b1;
// Don't shift if we just loaded new data, obviously.
if(corr_i_cnt != 6'd0)
begin
corr_i_out[7:0] <= {corr_i_out[6:0], corr_q_out[7]};
corr_q_out[7:1] <= corr_q_out[6:0];
end
end
// set ssp_frame signal for corr_i_cnt = 0..3 and corr_i_cnt = 32..35
// (send two frames with 8 Bits each)
if(corr_i_cnt[5:2] == 4'b0000 || corr_i_cnt[5:2] == 4'b1000)
ssp_frame = 1'b1;
else
ssp_frame = 1'b0;
end
assign ssp_din = corr_i_out[7];
assign dbg = corr_i_cnt[3];
// Unused.
assign pwr_lo = 1'b0;
assign pwr_oe2 = 1'b0;
endmodule