ua1zbe
/
dsd-fme-01.09.2024
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

M17 - Implement libM17 Viterbi Decoder for LSF;

This commit is contained in:
lwvmobile 2024-03-08 09:41:19 -05:00
parent 677b064e97
commit 5c09130782
2 changed files with 231 additions and 272 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
src/3.c
View File

@ -2,7 +2,7 @@
* 3.c
* Placeholder File 3 -- This file is for temporary code storage
*
* A 1/2 Rate Convolutional Decoder for NXDN/M17/YSF
* Multiple 1/2 Rate Convolutional Decoders for NXDN/M17/YSF
* Used as a secondary decoder in case convolutional decoder fails (second opinion)
*-----------------------------------------------------------------------------*/
@ -87,15 +87,6 @@ void trellis_decode(uint8_t result[], const uint8_t source[], int result_len)
*/
//Ripped from libM17
//prototypes
// uint32_t viterbi_decode(uint8_t* out, const uint16_t* in, const uint16_t len);
// uint32_t viterbi_decode_punctured(uint8_t* out, const uint16_t* in, const uint8_t* punct, const uint16_t in_len, const uint16_t p_len);
// void viterbi_decode_bit(uint16_t s0, uint16_t s1, const size_t pos);
// uint32_t viterbi_chainback(uint8_t* out, size_t pos, uint16_t len);
// void viterbi_reset(void);
// uint16_t q_abs_diff(const uint16_t v1, const uint16_t v2);
#define K 5 //constraint length
#define NUM_STATES (1 << (K - 1)) //number of states
@ -129,8 +120,14 @@ uint32_t viterbi_decode(uint8_t* out, const uint16_t* in, const uint16_t len)
viterbi_decode_bit(s0, s1, pos);
pos++;
}
uint32_t err = viterbi_chainback(out, pos, len/2);
return viterbi_chainback(out, pos, len/2);
//debug
// fprintf (stderr, "\n vcb: \n");
// for (size_t i = 0; i < len; i++)
// fprintf (stderr, "%02X", out[i]);
return err;
}
/**
@ -170,6 +167,9 @@ uint32_t viterbi_decode_punctured(uint8_t* out, const uint16_t* in, const uint8_
p%=p_len;
}
//debug
// fprintf (stderr, " p: %d, u: %d; p_len: %d; len: %d;", p, u, p_len, (u-in_len)*0x7FFF);
return viterbi_decode(out, umsg, u) - (u-in_len)*0x7FFF;
}
@ -272,6 +272,11 @@ uint32_t viterbi_chainback(uint8_t* out, size_t pos, uint16_t len)
if(m < cost) cost = m;
}
//debug
// fprintf (stderr, "\n cb: \n");
// for (size_t i = 0; i < len; i++)
// fprintf (stderr, "%02X", out[i]);
return cost;
}

250
src/m17.c
View File

@ -684,26 +684,36 @@ void processM17STR(dsd_opts * opts, dsd_state * state)
void processM17LSF(dsd_opts * opts, dsd_state * state)
{
//WIP: Switch to libM17 Viterbi Decoder for better performance
//Not yet working on actual signal, but debug version works
//most likely frame sync / demodulation related issue
//NOTE: Have not been able to get this to work successfully, but considering
//there is only one LSF frame at the beginning of TX, its probably just as easy to
//use and rely on the STR frame and embedded LSF chunk, which does work properly
int i, x;
uint8_t dbuf[184]; //384-bit frame - 16-bit (8 symbol) sync pattern (184 dibits)
uint8_t m17_rnd_bits[368]; //368 bits that are still scrambled (randomized)
//TODO: Work on implementing a symbol buffer of some sort
//when the preamble starts?
int i, j, k, x;
uint8_t dbuf[184]; //384-bit (192 symbol) frame - 16-bit (8 symbol) sync pattern (184 dibits)
uint8_t m17_int_bits[368]; //368 bits that are still interleaved
uint8_t m17_bits[368]; //368 bits that have been de-interleaved and de-scramble
uint8_t m17_depunc[488]; //488 bits after depuncturing
uint8_t m17_rnd_bits[368]; //368 bits that are still scrambled
uint16_t m17_bits[368]; //368 bits that have been de-interleaved and de-scrambled
uint16_t m17_depunc[488]; //488 weighted byte representation of bits after depuncturing
uint8_t lsf_bytes[31];
uint8_t lsf_packed[30];
uint32_t v_err = 0; //errors in viterbi decoder
UNUSED(v_err);
memset (dbuf, 0, sizeof(dbuf));
memset (m17_rnd_bits, 0, sizeof(m17_rnd_bits));
memset (state->m17_lsf, 0, sizeof(state->m17_lsf));
memset (m17_int_bits, 0, sizeof(m17_int_bits));
memset (m17_bits, 0, sizeof(m17_bits));
memset (m17_rnd_bits, 0, sizeof(m17_rnd_bits));
memset (m17_depunc, 0, sizeof(m17_depunc));
memset (lsf_packed, 0, sizeof(lsf_packed));
memset (lsf_bytes, 0, sizeof(lsf_bytes));
//load dibits into dibit buffer
for (i = 0; i < 184; i++)
dbuf[i] = (uint8_t) getDibit(opts, state);
dbuf[i] = getDibit(opts, state);
//convert dbuf into a bit array
for (i = 0; i < 184; i++)
@ -724,67 +734,54 @@ void processM17LSF(dsd_opts * opts, dsd_state * state)
m17_bits[i] = m17_int_bits[x];
}
// P1 Depuncture
//P1 Depuncture & Add Weights
x = 0;
for (i = 0; i < 488; i++)
{
if (p1[i%61] == 1)
m17_depunc[i] = m17_bits[x++];
else m17_depunc[i] = 0;
if (m17_depunc[i])
m17_depunc[i] = 0xFFFF;
else m17_depunc[i] = 0x7FFF;
}
//debug -- values seem okay at end of run
// fprintf (stderr, "K = %d; J = %d; X = %d", k, j, x);
//trailing zeroes
// for (i = 484; i < 488; i++)
// m17_depunc[i] = 0x7FFF;
//setup the convolutional decoder
uint8_t temp[488];
uint8_t s0;
uint8_t s1;
uint8_t m_data[30];
uint8_t trellis_buf[240]; //30*8 = 240
memset (trellis_buf, 0, sizeof(trellis_buf));
memset (temp, 0, sizeof (temp));
memset (m_data, 0, sizeof (m_data));
//debug -- fill all of the byte array
// memset (m17_depunc, 0xFFFF, sizeof(m17_depunc));
for (i = 0; i < 488; i++)
temp[i] = m17_depunc[i] << 1;
//debug
// fprintf (stderr, "\n depunc: \n");
// for (i = 0; i < 488; i++)
// fprintf (stderr, " %04X", m17_depunc[i]);
CNXDNConvolution_start();
for (i = 0; i < 244; i++)
//use the libM17 Viterbi Decoder
uint16_t len = 488;
v_err = viterbi_decode(lsf_bytes, m17_depunc, len);
// v_err -= 3932040; //cost negation (double check this as well as unit, meaning, etc)
//debug
// fprintf (stderr, "\n lsf_bytes: \n");
// for (i = 0; i < 31; i++)
// fprintf (stderr, " %02X", lsf_bytes[i]);
//copy + left shift one octet
memcpy (lsf_packed, lsf_bytes+1, 30);
//Unpack bytes into m17_lsf bits
k = 0;
for (j = 0; j < 30; j++)
{
s0 = temp[(2*i)];
s1 = temp[(2*i)+1];
CNXDNConvolution_decode(s0, s1);
for (i = 0; i < 8; i++)
state->m17_lsf[k++] = (lsf_packed[j] >> 7-i) & 1;
}
CNXDNConvolution_chainback(m_data, 240);
//244/8 = 30, last 4 (244-248) are trailing zeroes
for(i = 0; i < 30; i++)
{
trellis_buf[(i*8)+0] = (m_data[i] >> 7) & 1;
trellis_buf[(i*8)+1] = (m_data[i] >> 6) & 1;
trellis_buf[(i*8)+2] = (m_data[i] >> 5) & 1;
trellis_buf[(i*8)+3] = (m_data[i] >> 4) & 1;
trellis_buf[(i*8)+4] = (m_data[i] >> 3) & 1;
trellis_buf[(i*8)+5] = (m_data[i] >> 2) & 1;
trellis_buf[(i*8)+6] = (m_data[i] >> 1) & 1;
trellis_buf[(i*8)+7] = (m_data[i] >> 0) & 1;
}
memset (state->m17_lsf, 0, sizeof(state->m17_lsf));
memcpy (state->m17_lsf, trellis_buf, 240);
uint8_t lsf_packed[30];
memset (lsf_packed, 0, sizeof(lsf_packed));
//need to pack bytes for the sw5wwp variant of the crc (might as well, may be useful in the future)
for (i = 0; i < 30; i++)
lsf_packed[i] = (uint8_t)ConvertBitIntoBytes(&state->m17_lsf[i*8], 8);
uint16_t crc_cmp = crc16m17(lsf_packed, 28);
uint16_t crc_ext = (uint16_t)ConvertBitIntoBytes(&state->m17_lsf[224], 16);
uint16_t crc_ext = (lsf_packed[28] << 8) + (lsf_packed[29] << 0);
int crc_err = 0;
if (crc_cmp != crc_ext) crc_err = 1;
@ -800,6 +797,7 @@ void processM17LSF(dsd_opts * opts, dsd_state * state)
if (i == 15) fprintf (stderr, "\n ");
fprintf (stderr, "[%02X]", lsf_packed[i]);
}
// fprintf (stderr, " V Err: %d", v_err);
}
//debug
@ -807,6 +805,9 @@ void processM17LSF(dsd_opts * opts, dsd_state * state)
if (crc_err == 1) fprintf (stderr, " CRC ERR");
//zero out after decoding
memset (state->m17_lsf, 0, sizeof(state->m17_lsf));
//ending linebreak
fprintf (stderr, "\n");
@ -816,43 +817,15 @@ void processM17LSF(dsd_opts * opts, dsd_state * state)
void processM17LSF_debug(dsd_opts * opts, dsd_state * state, uint8_t * m17_depunc)
{
//NOTE: Having run a full debug on all steps, I am convinced that the number of punctures
//renders the current convolutional decoder useless to fix the holes in an LSF frame
//Skipping to the deconvolution step renders the correct frame info and crc value
//We can still send the full LSF as intended over IP frame or RF without this step
//TODO: Switch to M17 Viterbi Decoder for better performance
//NOTE: This version has been stripped down, but will not be used,
//but will keep in case of future debug for ease of use
int i, x; UNUSED(x);
uint8_t m17_int_bits[368]; //368 bits that are still interleaved
uint8_t m17_bits[368]; //368 bits that have been de-interleaved and de-scramble
// uint8_t m17_depunc[488]; //488 bits after depuncturing
memset (m17_int_bits, 0, sizeof(m17_int_bits));
memset (m17_bits, 0, sizeof(m17_bits));
// memset (m17_depunc, 0, sizeof(m17_depunc));
//descramble the frame
// for (i = 0; i < 368; i++)
// m17_int_bits[i] = (m17_rnd_bits[i] ^ m17_scramble[i]) & 1;
//deinterleave the bit array using Quadratic Permutation Polynomial
//function π(x) = (45x + 92x^2 ) mod 368
// for (i = 0; i < 368; i++)
// {
// x = ((45*i)+(92*i*i)) % 368;
// m17_bits[i] = m17_int_bits[x];
// }
//P1 Depuncture
// x = 0;
// for (i = 0; i < 488; i++)
// {
// if (p1[i%61] == 1)
// m17_depunc[i] = m17_bits[x++];
// else m17_depunc[i] = 0;
// }
//setup the convolutional decoder
uint8_t temp[488];
@ -925,31 +898,26 @@ void processM17LSF_debug(dsd_opts * opts, dsd_state * state, uint8_t * m17_depun
if (crc_err == 1) fprintf (stderr, " CRC ERR");
//ending linebreak
// fprintf (stderr, "\n");
} //end processM17LSF_debug
//This will be the stripped out version to switch over to the libM17 viterbi decoder
void processM17LSF_debug2(dsd_opts * opts, dsd_state * state, uint8_t * m17_rnd_bits)
{
//NOTE: Having run a full debug on all steps, I am convinced that the number of punctures
//renders the current convolutional decoder useless to fix the holes in an LSF frame
//Skipping to the deconvolution step renders the correct frame info and crc value
//We can still send the full LSF as intended over IP frame or RF without this step
//TODO: Switch to M17 Viterbi Decoder for better performance
int i, x;
//Working: Switched to libM17 Viterbi Decoder
int i, j, k, x;
uint8_t m17_int_bits[368]; //368 bits that are still interleaved
uint8_t m17_bits[368]; //368 bits that have been de-interleaved and de-scramble
uint8_t m17_depunc[488]; //488 bits after depuncturing
uint16_t m17_bits[368]; //368 bits that have been de-interleaved and de-scrambled
uint16_t m17_depunc[488]; //488 weighted byte representation of bits after depuncturing
uint8_t lsf_bytes[31];
uint8_t lsf_packed[30];
uint32_t v_err = 0; //errors in viterbi decoder
UNUSED(v_err);
memset (state->m17_lsf, 0, sizeof(state->m17_lsf));
memset (m17_int_bits, 0, sizeof(m17_int_bits));
memset (m17_bits, 0, sizeof(m17_bits));
memset (m17_depunc, 0, sizeof(m17_depunc));
memset (lsf_packed, 0, sizeof(lsf_packed));
memset (lsf_bytes, 0, sizeof(lsf_bytes));
//descramble the frame
for (i = 0; i < 368; i++)
@ -963,64 +931,50 @@ void processM17LSF_debug2(dsd_opts * opts, dsd_state * state, uint8_t * m17_rnd_
m17_bits[i] = m17_int_bits[x];
}
//P1 Depuncture
//P1 Depuncture & Add Weights
x = 0;
for (i = 0; i < 488; i++)
{
if (p1[i%61] == 1)
m17_depunc[i] = m17_bits[x++];
else m17_depunc[i] = 0;
if (m17_depunc[i])
m17_depunc[i] = 0xFFFF;
else m17_depunc[i] = 0x7FFF;
}
//setup the convolutional decoder
uint8_t temp[488];
uint8_t s0;
uint8_t s1;
uint8_t m_data[30];
uint8_t trellis_buf[240]; //30*8 = 240
memset (trellis_buf, 0, sizeof(trellis_buf));
memset (temp, 0, sizeof (temp));
memset (m_data, 0, sizeof (m_data));
//debug -- fill all of the byte array
// memset (m17_depunc, 0xFFFF, sizeof(m17_depunc));
for (i = 0; i < 488; i++)
temp[i] = m17_depunc[i] << 1;
//debug
// fprintf (stderr, "\n depunc: \n");
// for (i = 0; i < 488; i++)
// fprintf (stderr, " %04X", m17_depunc[i]);
CNXDNConvolution_start();
for (i = 0; i < 244; i++)
//use the libM17 Viterbi Decoder
uint16_t len = 488;
v_err = viterbi_decode(lsf_bytes, m17_depunc, len);
// v_err -= 3932040; //cost negation (double check this as well as unit, meaning, etc)
//debug
// fprintf (stderr, "\n lsf_bytes: \n");
// for (i = 0; i < 31; i++)
// fprintf (stderr, " %02X", lsf_bytes[i]);
//copy + left shift one octet
memcpy (lsf_packed, lsf_bytes+1, 30);
//Unpack bytes into m17_lsf bits
k = 0;
for (j = 0; j < 30; j++)
{
s0 = temp[(2*i)];
s1 = temp[(2*i)+1];
CNXDNConvolution_decode(s0, s1);
for (i = 0; i < 8; i++)
state->m17_lsf[k++] = (lsf_packed[j] >> 7-i) & 1;
}
CNXDNConvolution_chainback(m_data, 240);
//244/8 = 30, last 4 (244-248) are trailing zeroes
for(i = 0; i < 30; i++)
{
trellis_buf[(i*8)+0] = (m_data[i] >> 7) & 1;
trellis_buf[(i*8)+1] = (m_data[i] >> 6) & 1;
trellis_buf[(i*8)+2] = (m_data[i] >> 5) & 1;
trellis_buf[(i*8)+3] = (m_data[i] >> 4) & 1;
trellis_buf[(i*8)+4] = (m_data[i] >> 3) & 1;
trellis_buf[(i*8)+5] = (m_data[i] >> 2) & 1;
trellis_buf[(i*8)+6] = (m_data[i] >> 1) & 1;
trellis_buf[(i*8)+7] = (m_data[i] >> 0) & 1;
}
memset (state->m17_lsf, 0, sizeof(state->m17_lsf));
memcpy (state->m17_lsf, trellis_buf, 240);
uint8_t lsf_packed[30];
memset (lsf_packed, 0, sizeof(lsf_packed));
//need to pack bytes for the sw5wwp variant of the crc (might as well, may be useful in the future)
for (i = 0; i < 30; i++)
lsf_packed[i] = (uint8_t)ConvertBitIntoBytes(&state->m17_lsf[i*8], 8);
uint16_t crc_cmp = crc16m17(lsf_packed, 28);
uint16_t crc_ext = (uint16_t)ConvertBitIntoBytes(&state->m17_lsf[224], 16);
uint16_t crc_ext = (lsf_packed[28] << 8) + (lsf_packed[29] << 0);
int crc_err = 0;
if (crc_cmp != crc_ext) crc_err = 1;
@ -1036,6 +990,7 @@ void processM17LSF_debug2(dsd_opts * opts, dsd_state * state, uint8_t * m17_rnd_
if (i == 15) fprintf (stderr, "\n ");
fprintf (stderr, "[%02X]", lsf_packed[i]);
}
// fprintf (stderr, " V Err: %d", v_err);
}
//debug
@ -1824,8 +1779,7 @@ void encodeM17STR(dsd_opts * opts, dsd_state * state)
fprintf (stderr, "\n M17 LSF (ENCODER): ");
if (opts->monitor_input_audio == 0)
processM17LSF_debug(opts, state, m17_lsfc);
// processM17LSF_debug(opts, state, m17_lsfs);
processM17LSF_debug2(opts, state, m17_lsfs);
else fprintf (stderr, " To Audio Out Device Type: %d; ", opts->audio_out_type);
//encodeM17RF