Add some testing code that looks for ambiguous syndromes that

could confuse the partial correction used in DF11.

That code shows that yes, there are ambiguous syndromes in the
2-bit correction case only, so disable corrections of more than
1 bit in DF11.
This commit is contained in:
Oliver Jowett 2015-01-23 21:59:31 +00:00
parent 5ac90e30a8
commit 26feb1d078
2 changed files with 127 additions and 6 deletions

124
crc.c
View file

@ -421,7 +421,9 @@ void modesChecksumFix(uint8_t *msg, struct errorinfo *info)
#ifdef CRCDEBUG
int main(int argc, char **argv)
{
int len;
int shortlen, longlen;
int i;
struct errorinfo *shorttable, *longtable;
if (argc < 3) {
fprintf(stderr, "syntax: crctests <ncorrect> <ndetect>\n");
@ -429,8 +431,124 @@ int main(int argc, char **argv)
}
initLookupTables();
prepareErrorTable(MODES_SHORT_MSG_BITS, atoi(argv[1]), atoi(argv[2]), &len);
prepareErrorTable(MODES_LONG_MSG_BITS, atoi(argv[1]), atoi(argv[2]), &len);
shorttable = prepareErrorTable(MODES_SHORT_MSG_BITS, atoi(argv[1]), atoi(argv[2]), &shortlen);
longtable = prepareErrorTable(MODES_LONG_MSG_BITS, atoi(argv[1]), atoi(argv[2]), &longlen);
// check for DF11 correction syndromes where there is a syndrome with lower 7 bits all zero
// (which would be used for DF11 error correction), but there's also a syndrome which has
// the same upper 17 bits but nonzero lower 7 bits.
// empirically, with ncorrect=1 ndetect=2 we get no ambiguous syndromes;
// for ncorrect=2 ndetect=4 we get 11 ambiguous syndromes:
/*
syndrome 1 = 000C00 bits=[ 44 45 ]
syndrome 2 = 000C1B bits=[ 30 43 ]
syndrome 1 = 001400 bits=[ 43 45 ]
syndrome 2 = 00141B bits=[ 30 44 ]
syndrome 1 = 001800 bits=[ 43 44 ]
syndrome 2 = 00181B bits=[ 30 45 ]
syndrome 1 = 001800 bits=[ 43 44 ]
syndrome 2 = 001836 bits=[ 29 42 ]
syndrome 1 = 002400 bits=[ 42 45 ]
syndrome 2 = 00242D bits=[ 29 30 ]
syndrome 1 = 002800 bits=[ 42 44 ]
syndrome 2 = 002836 bits=[ 29 43 ]
syndrome 1 = 003000 bits=[ 42 43 ]
syndrome 2 = 003036 bits=[ 29 44 ]
syndrome 1 = 003000 bits=[ 42 43 ]
syndrome 2 = 00306C bits=[ 28 41 ]
syndrome 1 = 004800 bits=[ 41 44 ]
syndrome 2 = 00485A bits=[ 28 29 ]
syndrome 1 = 005000 bits=[ 41 43 ]
syndrome 2 = 00506C bits=[ 28 42 ]
syndrome 1 = 006000 bits=[ 41 42 ]
syndrome 2 = 00606C bits=[ 28 43 ]
*/
// So in the DF11 correction logic, we just discard messages that require more than a 1 bit fix.
fprintf(stderr, "checking %d syndromes for DF11 collisions..\n", shortlen);
for (i = 0; i < shortlen; ++i) {
if ((shorttable[i].syndrome & 0xFF) == 0) {
int j;
// all syndromes with the same first 17 bits should sort immediately after entry i,
// so this is fairly easy
for (j = i+1; j < shortlen; ++j) {
if ((shorttable[i].syndrome & 0xFFFF80) == (shorttable[j].syndrome & 0xFFFF80)) {
int k;
int mismatch = 0;
// we don't care if the only differences are in bits that lie in the checksum
for (k = 0; k < shorttable[i].errors; ++k) {
int l, matched = 0;
if (shorttable[i].bit[k] >= 49)
continue; // bit is in the final 7 bits, we don't care
for (l = 0; l < shorttable[j].errors; ++l) {
if (shorttable[i].bit[k] == shorttable[j].bit[l]) {
matched = 1;
break;
}
}
if (!matched)
mismatch = 1;
}
for (k = 0; k < shorttable[j].errors; ++k) {
int l, matched = 0;
if (shorttable[j].bit[k] >= 49)
continue; // bit is in the final 7 bits, we don't care
for (l = 0; l < shorttable[i].errors; ++l) {
if (shorttable[j].bit[k] == shorttable[i].bit[l]) {
matched = 1;
break;
}
}
if (!matched)
mismatch = 1;
}
if (mismatch) {
fprintf(stderr,
"DF11 correction collision: \n"
" syndrome 1 = %06X bits=[",
shorttable[i].syndrome);
for (k = 0; k < shorttable[i].errors; ++k)
fprintf(stderr, " %d", shorttable[i].bit[k]);
fprintf(stderr, " ]\n");
fprintf(stderr,
" syndrome 2 = %06X bits=[",
shorttable[j].syndrome);
for (k = 0; k < shorttable[j].errors; ++k)
fprintf(stderr, " %d", shorttable[j].bit[k]);
fprintf(stderr, " ]\n");
}
} else {
break;
}
}
}
}
free(shorttable);
free(longtable);
return 0;
}