mirror of https://github.com/wb2osz/direwolf.git
662 lines
16 KiB
C
662 lines
16 KiB
C
//
|
|
// This file is part of Dire Wolf, an amateur radio packet TNC.
|
|
//
|
|
// Copyright (C) 2011, 2012, 2013 John Langner, WB2OSZ
|
|
//
|
|
// This program is free software: you can redistribute it and/or modify
|
|
// it under the terms of the GNU General Public License as published by
|
|
// the Free Software Foundation, either version 2 of the License, or
|
|
// (at your option) any later version.
|
|
//
|
|
// This program is distributed in the hope that it will be useful,
|
|
// but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
// GNU General Public License for more details.
|
|
//
|
|
// You should have received a copy of the GNU General Public License
|
|
// along with this program. If not, see <http://www.gnu.org/licenses/>.
|
|
//
|
|
|
|
|
|
/********************************************************************************
|
|
*
|
|
* File: hdlc_rec2.c
|
|
*
|
|
* Purpose: Extract HDLC frame from a block of bits after someone
|
|
* else has done the work of pulling it out from between
|
|
* the special "flag" sequences.
|
|
*
|
|
*******************************************************************************/
|
|
|
|
#include <stdio.h>
|
|
#include <assert.h>
|
|
#include <ctype.h>
|
|
|
|
#include "direwolf.h"
|
|
#include "hdlc_rec2.h"
|
|
#include "fcs_calc.h"
|
|
#include "textcolor.h"
|
|
#include "ax25_pad.h"
|
|
#include "rrbb.h"
|
|
#include "rdq.h"
|
|
#include "multi_modem.h"
|
|
|
|
|
|
/*
|
|
* Minimum & maximum sizes of an AX.25 frame including the 2 octet FCS.
|
|
*/
|
|
|
|
#define MIN_FRAME_LEN ((AX25_MIN_PACKET_LEN) + 2)
|
|
|
|
#define MAX_FRAME_LEN ((AX25_MAX_PACKET_LEN) + 2)
|
|
|
|
/*
|
|
* This is the current state of the HDLC decoder.
|
|
*
|
|
* It is possible to run multiple decoders concurrently by
|
|
* having a separate set of state variables for each.
|
|
*
|
|
* Should have a reset function instead of initializations here.
|
|
*/
|
|
|
|
struct hdlc_state_s {
|
|
|
|
int prev_raw; /* Keep track of previous bit so */
|
|
/* we can look for transitions. */
|
|
/* Should be only 0 or 1. */
|
|
|
|
unsigned char pat_det; /* 8 bit pattern detector shift register. */
|
|
/* See below for more details. */
|
|
|
|
unsigned char oacc; /* Accumulator for building up an octet. */
|
|
|
|
int olen; /* Number of bits in oacc. */
|
|
/* When this reaches 8, oacc is copied */
|
|
/* to the frame buffer and olen is zeroed. */
|
|
|
|
unsigned char frame_buf[MAX_FRAME_LEN];
|
|
/* One frame is kept here. */
|
|
|
|
int frame_len; /* Number of octets in frame_buf. */
|
|
/* Should be in range of 0 .. MAX_FRAME_LEN. */
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
static int try_decode (rrbb_t block, int chan, int subchan, int alevel, retry_t bits_flipped, int flip_a, int flip_b, int flip_c);
|
|
static int try_to_fix_quick_now (rrbb_t block, int chan, int subchan, int alevel, retry_t fix_bits);
|
|
static int sanity_check (unsigned char *buf, int blen, retry_t bits_flipped);
|
|
#if DEBUG
|
|
static double dtime_now (void);
|
|
#endif
|
|
|
|
/***********************************************************************************
|
|
*
|
|
* Name: hdlc_rec2_block
|
|
*
|
|
* Purpose: Extract HDLC frame from a stream of bits.
|
|
*
|
|
* Inputs: block - Handle for bit array.
|
|
* fix_bits - Level of effort to recover frames with bad FCS.
|
|
*
|
|
* Description: The other (original) hdlc decoder took one bit at a time
|
|
* right out of the demodulator.
|
|
*
|
|
* This is different in that it processes a block of bits
|
|
* previously extracted from between two "flag" patterns.
|
|
*
|
|
* This allows us to try decoding the same received data more
|
|
* than once.
|
|
*
|
|
* Bugs: This does not work for 9600 baud, more accurately when
|
|
* the transmitted bits are scrambled.
|
|
*
|
|
* Currently we unscramble the bits as they come from the
|
|
* receiver. Instead we need to save the original received
|
|
* bits and apply the descrambling after flipping the bits.
|
|
*
|
|
***********************************************************************************/
|
|
|
|
|
|
void hdlc_rec2_block (rrbb_t block, retry_t fix_bits)
|
|
{
|
|
int chan = rrbb_get_chan(block);
|
|
int subchan = rrbb_get_subchan(block);
|
|
int alevel = rrbb_get_audio_level(block);
|
|
int ok;
|
|
int n;
|
|
|
|
#if DEBUGx
|
|
text_color_set(DW_COLOR_DEBUG);
|
|
dw_printf ("\n--- try to decode ---\n");
|
|
#endif
|
|
|
|
#if SLICENDICE
|
|
/*
|
|
* Unfinished experiment. Get back to this again someday.
|
|
* The demodulator output is (should be) roughly in the range of -1 to 1.
|
|
* Formerly we sliced it at 0 and saved only a single bit for the sample.
|
|
* Now we save the sample so we can try adjusting the slicing point.
|
|
*
|
|
* First time thru, set the slicing point to 0.
|
|
*/
|
|
|
|
for (n = 0; n < 1 ; n++) {
|
|
|
|
rrbb_set_slice_val (block, n);
|
|
|
|
ok = try_decode (block, chan, subchan, alevel, RETRY_NONE, -1, -1, -1);
|
|
|
|
if (ok) {
|
|
//#if DEBUG
|
|
text_color_set(DW_COLOR_INFO);
|
|
dw_printf ("Got it with no errors. Slice val = %d \n", n);
|
|
//#endif
|
|
rrbb_delete (block);
|
|
return;
|
|
}
|
|
}
|
|
rrbb_set_slice_val (block, 0);
|
|
|
|
#else /* not SLICENDICE */
|
|
|
|
ok = try_decode (block, chan, subchan, alevel, RETRY_NONE, -1, -1, -1);
|
|
|
|
if (ok) {
|
|
#if DEBUG
|
|
text_color_set(DW_COLOR_INFO);
|
|
dw_printf ("Got it the first time.\n");
|
|
#endif
|
|
rrbb_delete (block);
|
|
return;
|
|
}
|
|
#endif
|
|
|
|
if (try_to_fix_quick_now (block, chan, subchan, alevel, fix_bits)) {
|
|
rrbb_delete (block);
|
|
return;
|
|
}
|
|
|
|
/*
|
|
* Put in queue for retrying later at lower priority.
|
|
*/
|
|
|
|
if (fix_bits < RETRY_TWO_SEP) {
|
|
rrbb_delete (block);
|
|
return;
|
|
}
|
|
|
|
rdq_append (block);
|
|
|
|
}
|
|
|
|
|
|
static int try_to_fix_quick_now (rrbb_t block, int chan, int subchan, int alevel, retry_t fix_bits)
|
|
{
|
|
int ok;
|
|
int len, i;
|
|
|
|
|
|
len = rrbb_get_len(block);
|
|
|
|
/*
|
|
* Try fixing one bit.
|
|
*/
|
|
if (fix_bits < RETRY_SINGLE) {
|
|
return 0;
|
|
}
|
|
|
|
for (i=0; i<len; i++) {
|
|
ok = try_decode (block, chan, subchan, alevel, RETRY_SINGLE, i, -1, -1);
|
|
if (ok) {
|
|
#if DEBUG
|
|
text_color_set(DW_COLOR_ERROR);
|
|
dw_printf ("*** Success by flipping SINGLE bit %d of %d ***\n", i, len);
|
|
#endif
|
|
return 1;
|
|
}
|
|
}
|
|
|
|
/*
|
|
* Try fixing two adjacent bits.
|
|
*/
|
|
if (fix_bits < RETRY_DOUBLE) {
|
|
return 0;
|
|
}
|
|
|
|
for (i=0; i<len-1; i++) {
|
|
ok = try_decode (block, chan, subchan, alevel, RETRY_DOUBLE, i, i+1, -1);
|
|
if (ok) {
|
|
#if DEBUG
|
|
text_color_set(DW_COLOR_ERROR);
|
|
dw_printf ("*** Success by flipping DOUBLE bit %d of %d ***\n", i, len);
|
|
#endif
|
|
return 1;
|
|
}
|
|
}
|
|
|
|
/*
|
|
* Try fixing adjacent three bits.
|
|
*/
|
|
if (fix_bits < RETRY_TRIPLE) {
|
|
return 0;
|
|
}
|
|
|
|
len = rrbb_get_len(block);
|
|
for (i=0; i<len-2; i++) {
|
|
ok = try_decode (block, chan, subchan, alevel, RETRY_TRIPLE, i, i+1, i+2);
|
|
if (ok) {
|
|
#if DEBUG
|
|
text_color_set(DW_COLOR_ERROR);
|
|
dw_printf ("*** Success by flipping TRIPLE bit %d of %d ***\n", i, len);
|
|
#endif
|
|
return 1;
|
|
}
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
|
|
void hdlc_rec2_try_to_fix_later (rrbb_t block, int chan, int subchan, int alevel)
|
|
{
|
|
int ok;
|
|
int len, i;
|
|
#if DEBUG
|
|
double tstart, tend;
|
|
#endif
|
|
|
|
len = rrbb_get_len(block);
|
|
|
|
/*
|
|
* Two non-adjacent ("separated") single bits.
|
|
* It chews up a lot of CPU time. Test takes 4 times longer to run.
|
|
*
|
|
* Ran up to 4.82 seconds for 1040 bits before giving up.
|
|
* Processing time is order N squared so time goes up rapidly with larger frames.
|
|
*/
|
|
|
|
#if DEBUG
|
|
tstart = dtime_now();
|
|
#endif
|
|
len = rrbb_get_len(block);
|
|
for (i=0; i<len-2; i++) {
|
|
int j;
|
|
|
|
ok = 0;
|
|
for (j=i+2; j<len; j++) {
|
|
ok = try_decode (block, chan, subchan, alevel, RETRY_TWO_SEP, i, j, -1);
|
|
if (ok)
|
|
break;
|
|
}
|
|
if (ok) {
|
|
#if DEBUG
|
|
tend = dtime_now();
|
|
text_color_set(DW_COLOR_ERROR);
|
|
dw_printf ("*** Success by flipping TWO SEPARATED bits %d and %d of %d *** %.3f sec.\n", i, j, len, tend-tstart);
|
|
#endif
|
|
return;
|
|
}
|
|
}
|
|
#if DEBUGx
|
|
tend = dtime_now();
|
|
text_color_set(DW_COLOR_ERROR);
|
|
dw_printf ("*** No luck flipping TWO SEPARATED bits of %d *** %.3f sec.\n", len, tend-tstart);
|
|
#endif
|
|
|
|
return;
|
|
}
|
|
|
|
|
|
|
|
|
|
static int try_decode (rrbb_t block, int chan, int subchan, int alevel, retry_t bits_flipped, int flip_a, int flip_b, int flip_c)
|
|
{
|
|
struct hdlc_state_s H;
|
|
int blen; /* Block length in bits. */
|
|
int i;
|
|
int raw; /* From demodulator. */
|
|
int dbit; /* Data bit after undoing NRZI. */
|
|
|
|
|
|
H.prev_raw = rrbb_get_bit (block, 0); /* Actually last bit of the */
|
|
/* opening flag so we can derive the */
|
|
/* first data bit. */
|
|
|
|
/* Does this make sense? */
|
|
/* This is the last bit of the "flag" pattern. */
|
|
/* If it was corrupted we wouldn't have detected */
|
|
/* the start of frame. */
|
|
|
|
if (0 == flip_a || 0 == flip_b || 0 == flip_c){
|
|
H.prev_raw = ! H.prev_raw;
|
|
}
|
|
|
|
H.pat_det = 0;
|
|
H.oacc = 0;
|
|
H.olen = 0;
|
|
H.frame_len = 0;
|
|
|
|
blen = rrbb_get_len (block);
|
|
|
|
#if DEBUGx
|
|
text_color_set(DW_COLOR_DEBUG);
|
|
dw_printf ("try_decode: blen=%d\n", blen);
|
|
#endif
|
|
|
|
for (i=1; i<blen; i++) {
|
|
|
|
raw = rrbb_get_bit (block, i);
|
|
|
|
if (i == flip_a || i == flip_b || i == flip_c){
|
|
raw = ! raw;
|
|
}
|
|
|
|
/*
|
|
* Using NRZI encoding,
|
|
* A '0' bit is represented by an inversion since previous bit.
|
|
* A '1' bit is represented by no change.
|
|
*/
|
|
|
|
dbit = (raw == H.prev_raw);
|
|
H.prev_raw = raw;
|
|
|
|
/*
|
|
* Octets are sent LSB first.
|
|
* Shift the most recent 8 bits thru the pattern detector.
|
|
*/
|
|
H.pat_det >>= 1;
|
|
if (dbit) {
|
|
H.pat_det |= 0x80;
|
|
}
|
|
|
|
if (H.pat_det == 0x7e) {
|
|
/* The special pattern 01111110 indicates beginning and ending of a frame. */
|
|
#if DEBUGx
|
|
text_color_set(DW_COLOR_DEBUG);
|
|
dw_printf ("try_decode: found flag, i=%d\n", i);
|
|
#endif
|
|
return 0;
|
|
}
|
|
else if (H.pat_det == 0xfe) {
|
|
/* Valid data will never have 7 one bits in a row. */
|
|
#if DEBUGx
|
|
text_color_set(DW_COLOR_DEBUG);
|
|
dw_printf ("try_decode: found abort, i=%d\n", i);
|
|
#endif
|
|
return 0;
|
|
}
|
|
else if ( (H.pat_det & 0xfc) == 0x7c ) {
|
|
/*
|
|
* If we have five '1' bits in a row, followed by a '0' bit,
|
|
*
|
|
* 0111110xx
|
|
*
|
|
* the current '0' bit should be discarded because it was added for
|
|
* "bit stuffing."
|
|
*/
|
|
;
|
|
} else {
|
|
|
|
/*
|
|
* In all other cases, accumulate bits into octets, and complete octets
|
|
* into the frame buffer.
|
|
*/
|
|
|
|
H.oacc >>= 1;
|
|
if (dbit) {
|
|
H.oacc |= 0x80;
|
|
}
|
|
H.olen++;
|
|
|
|
if (H.olen == 8) {
|
|
H.olen = 0;
|
|
|
|
if (H.frame_len < MAX_FRAME_LEN) {
|
|
H.frame_buf[H.frame_len] = H.oacc;
|
|
H.frame_len++;
|
|
}
|
|
}
|
|
}
|
|
|
|
} /* end of loop on all bits in block */
|
|
|
|
/*
|
|
* Do we have a minimum number of complete bytes?
|
|
*/
|
|
|
|
#if DEBUGx
|
|
text_color_set(DW_COLOR_DEBUG);
|
|
dw_printf ("try_decode: olen=%d, frame_len=%d\n", H.olen, H.frame_len);
|
|
#endif
|
|
|
|
if (H.olen == 0 && H.frame_len >= MIN_FRAME_LEN) {
|
|
|
|
unsigned short actual_fcs, expected_fcs;
|
|
|
|
#if DEBUGx
|
|
int j;
|
|
text_color_set(DW_COLOR_DEBUG);
|
|
dw_printf ("NEW WAY: frame len = %d\n", H.frame_len);
|
|
for (j=0; j<H.frame_len; j++) {
|
|
dw_printf (" %02x", H.frame_buf[j]);
|
|
}
|
|
dw_printf ("\n");
|
|
#endif
|
|
/* Check FCS, low byte first, and process... */
|
|
|
|
/* Alternatively, it is possible to include the two FCS bytes */
|
|
/* in the CRC calculation and look for a magic constant. */
|
|
/* That would be easier in the case where the CRC is being */
|
|
/* accumulated along the way as the octets are received. */
|
|
/* I think making a second pass over it and comparing is */
|
|
/* easier to understand. */
|
|
|
|
actual_fcs = H.frame_buf[H.frame_len-2] | (H.frame_buf[H.frame_len-1] << 8);
|
|
|
|
expected_fcs = fcs_calc (H.frame_buf, H.frame_len - 2);
|
|
|
|
if (actual_fcs == expected_fcs && sanity_check (H.frame_buf, H.frame_len - 2, bits_flipped)) {
|
|
|
|
|
|
// TODO: Shouldn't be necessary to pass chan, subchan, alevel into
|
|
// try_decode because we can obtain them from block.
|
|
// Let's make sure that assumption is good...
|
|
|
|
assert (rrbb_get_chan(block) == chan);
|
|
assert (rrbb_get_subchan(block) == subchan);
|
|
assert (rrbb_get_audio_level(block) == alevel);
|
|
|
|
multi_modem_process_rec_frame (chan, subchan, H.frame_buf, H.frame_len - 2, alevel, bits_flipped); /* len-2 to remove FCS. */
|
|
return 1; /* success */
|
|
}
|
|
}
|
|
return 0; /* failure. */
|
|
|
|
} /* end try_decode */
|
|
|
|
|
|
/*
|
|
* Try to weed out bogus packets from initially failed FCS matches.
|
|
*/
|
|
|
|
static int sanity_check (unsigned char *buf, int blen, retry_t bits_flipped)
|
|
{
|
|
int alen; /* Length of address part. */
|
|
int j;
|
|
|
|
/*
|
|
* No sanity check if we didn't try fixing the data.
|
|
* Should we have different levels of checking depending on
|
|
* how much we try changing the raw data?
|
|
*/
|
|
if (bits_flipped == RETRY_NONE) {
|
|
return 1;
|
|
}
|
|
|
|
#if DEBUGx
|
|
text_color_set(DW_COLOR_XMIT);
|
|
dw_printf ("sanity_check: address part length = %d\n", alen);
|
|
#endif
|
|
|
|
/*
|
|
* Address part must be a multiple of 7.
|
|
*/
|
|
|
|
alen = 0;
|
|
for (j=0; j<blen && alen==0; j++) {
|
|
if (buf[j] & 0x01) {
|
|
alen = j + 1;
|
|
}
|
|
}
|
|
|
|
if (alen % 7 != 0) {
|
|
#if DEBUGx
|
|
text_color_set(DW_COLOR_ERROR);
|
|
dw_printf ("sanity_check: FAILED. Address part not multiple of 7.\n");
|
|
#endif
|
|
return 0;
|
|
}
|
|
|
|
/*
|
|
* Need at least 2 addresses and maximum of 8 digipeaters.
|
|
*/
|
|
|
|
if (alen/7 < 2 || alen/7 > 10) {
|
|
#if DEBUGx
|
|
text_color_set(DW_COLOR_ERROR);
|
|
dw_printf ("sanity_check: FAILED. Too few or many addresses.\n");
|
|
#endif
|
|
return 0;
|
|
}
|
|
|
|
/*
|
|
* Addresses can contain only upper case letters, digits, and space.
|
|
*/
|
|
|
|
for (j=0; j<alen; j+=7) {
|
|
|
|
char addr[7];
|
|
|
|
addr[0] = buf[j+0] >> 1;
|
|
addr[1] = buf[j+1] >> 1;
|
|
addr[2] = buf[j+2] >> 1;
|
|
addr[3] = buf[j+3] >> 1;
|
|
addr[4] = buf[j+4] >> 1;
|
|
addr[5] = buf[j+5] >> 1;
|
|
addr[6] = '\0';
|
|
|
|
|
|
if ( (! isupper(addr[0]) && ! isdigit(addr[0])) ||
|
|
(! isupper(addr[1]) && ! isdigit(addr[1]) && addr[1] != ' ') ||
|
|
(! isupper(addr[2]) && ! isdigit(addr[2]) && addr[2] != ' ') ||
|
|
(! isupper(addr[3]) && ! isdigit(addr[3]) && addr[3] != ' ') ||
|
|
(! isupper(addr[4]) && ! isdigit(addr[4]) && addr[4] != ' ') ||
|
|
(! isupper(addr[5]) && ! isdigit(addr[5]) && addr[5] != ' ')) {
|
|
#if DEBUGx
|
|
text_color_set(DW_COLOR_ERROR);
|
|
dw_printf ("sanity_check: FAILED. Invalid characters in addresses \"%s\"\n", addr);
|
|
#endif
|
|
return 0;
|
|
}
|
|
}
|
|
|
|
/*
|
|
* The next two bytes should be 0x03 and 0xf0 for APRS.
|
|
* Checking that would mean precluding use for other types of packet operation.
|
|
*
|
|
* The next section is also assumes APRS and might discard good data
|
|
* for other protocols.
|
|
*/
|
|
|
|
|
|
/*
|
|
* Finally, look for bogus characters in the information part.
|
|
* In theory, the bytes could have any values.
|
|
* In practice, we find only printable ASCII characters and:
|
|
*
|
|
* 0x0a line feed
|
|
* 0x0d carriage return
|
|
* 0x1c MIC-E
|
|
* 0x1d MIC-E
|
|
* 0x1e MIC-E
|
|
* 0x1f MIC-E
|
|
* 0x7f MIC-E
|
|
* 0x80 "{UIV32N}<0x0d><0x9f><0x80>"
|
|
* 0x9f "{UIV32N}<0x0d><0x9f><0x80>"
|
|
* 0xb0 degree symbol, ISO LATIN1
|
|
* (Note: UTF-8 uses two byte sequence 0xc2 0xb0.)
|
|
* 0xbe invalid MIC-E encoding.
|
|
* 0xf8 degree symbol, Microsoft code page 437
|
|
*
|
|
* So, if we have something other than these (in English speaking countries!),
|
|
* chances are that we have bogus data from twiddling the wrong bits.
|
|
*
|
|
* Notice that we shouldn't get here for good packets. This extra level
|
|
* of checking happens only if we twiddled a couple of bits, possibly
|
|
* creating bad data. We want to be very fussy.
|
|
*/
|
|
|
|
for (j=alen+2; j<blen; j++) {
|
|
int ch = buf[j];
|
|
|
|
if ( ! (( ch >= 0x1c && ch <= 0x7f)
|
|
|| ch == 0x0a
|
|
|| ch == 0x0d
|
|
|| ch == 0x80
|
|
|| ch == 0x9f
|
|
|| ch == 0xb0
|
|
|| ch == 0xf8) ) {
|
|
#if DEBUGx
|
|
text_color_set(DW_COLOR_ERROR);
|
|
dw_printf ("sanity_check: FAILED. Probably bogus info char 0x%02x\n", ch);
|
|
#endif
|
|
return 0;
|
|
}
|
|
}
|
|
|
|
return 1;
|
|
}
|
|
|
|
|
|
/* end hdlc_rec2.c */
|
|
|
|
|
|
|
|
|
|
// TODO: Also in xmit.c. Move to some common location.
|
|
|
|
|
|
/* Current time in seconds but more resolution than time(). */
|
|
|
|
/* We don't care what date a 0 value represents because we */
|
|
/* only use this to calculate elapsed time. */
|
|
|
|
|
|
#if DEBUG
|
|
|
|
static double dtime_now (void)
|
|
{
|
|
#if __WIN32__
|
|
/* 64 bit integer is number of 100 nanosecond intervals from Jan 1, 1601. */
|
|
|
|
FILETIME ft;
|
|
|
|
GetSystemTimeAsFileTime (&ft);
|
|
|
|
return ((( (double)ft.dwHighDateTime * (256. * 256. * 256. * 256.) +
|
|
(double)ft.dwLowDateTime ) / 10000000.) - 11644473600.);
|
|
#else
|
|
/* tv_sec is seconds from Jan 1, 1970. */
|
|
|
|
struct timespec ts;
|
|
|
|
clock_gettime (CLOCK_REALTIME, &ts);
|
|
|
|
return (ts.tv_sec + ts.tv_nsec / 1000000000.);
|
|
#endif
|
|
}
|
|
|
|
#endif
|