diff options
Diffstat (limited to 'util/spkmodem_decode')
| -rw-r--r-- | util/spkmodem_decode/spkmodem-decode.c | 339 |
1 files changed, 203 insertions, 136 deletions
diff --git a/util/spkmodem_decode/spkmodem-decode.c b/util/spkmodem_decode/spkmodem-decode.c index 8c57ff9a..3b3b33f8 100644 --- a/util/spkmodem_decode/spkmodem-decode.c +++ b/util/spkmodem_decode/spkmodem-decode.c @@ -63,6 +63,7 @@ #endif #include <errno.h> +#include <limits.h> #include <stdio.h> #include <stdarg.h> #include <stdlib.h> @@ -143,7 +144,7 @@ * Convert tone frequency ranges into pulse counts within the * sliding analysis window. * - * pulse_count ≈ tone_frequency / FRAME_RATE + * pulse_count = tone_frequency / FRAME_RATE * where FRAME_RATE = SAMPLE_RATE / SAMPLES_PER_FRAME. */ #define FREQ_SEP_MIN ((SEP_TONE_MIN_HZ) / (FRAME_RATE)) @@ -157,15 +158,10 @@ /* * These determine how long the program will wait during * tone auto-detection, before shifting to defaults. - * - * For tone auto-detection (time waiting for detection) - * NOTE: you could multiply SAMPLE_PER_FRAME instead - * of SAMPLE_RATE in LEARN_SAMPLES, for more granularity. - * Here, 1 * SAMPLE_RATE represents 1 second, which seems - * like a reasonable, conservative default wait time. + * It is done every LEARN_FRAMES number of frames. */ #define LEARN_SECONDS 1 -#define LEARN_SAMPLES ((LEARN_SECONDS) * (SAMPLE_RATE)) +#define LEARN_FRAMES ((LEARN_SECONDS) * (FRAME_RATE)) /* * Sample amplitude threshold used to convert the waveform @@ -209,35 +205,70 @@ struct decoder_state { int freq_min; int freq_max; int freq_threshold; - int learn_samples; + int learn_frames; + + /* previous sample used for edge detection */ + signed short prev_sample; }; static const char *argv0; +/* + * 16-bit little endian words are read + * continuously. we will swap them, if + * the host cpu is big endian. + */ static int host_is_big_endian(void); + +/* main loop */ static void handle_audio(struct decoder_state *st); -static void collect_separator_tone(struct decoder_state *st); -static int valid_signal(struct decoder_state *st); + +/* separate tone tolerances */ +static void select_separator_tone(struct decoder_state *st); +static int is_valid_signal(struct decoder_state *st); + +/* output to terminal */ +static int set_ascii_bit(struct decoder_state *st); +static void print_char(struct decoder_state *st); +static void reset_char(struct decoder_state *st); + +/* process samples/frames */ static void decode_pulse(struct decoder_state *st); -static void auto_detect_tone(struct decoder_state *st); -static int silent_signal(struct decoder_state *st); static signed short read_sample(struct decoder_state *st); +static void read_words(struct decoder_state *st); + +/* continually adjust tone */ +static void detect_tone(struct decoder_state *st); +static int silent_signal(struct decoder_state *st); static void select_low_tone(struct decoder_state *st); -static int set_ascii_bit(struct decoder_state *st); -static void print_char(struct decoder_state *st); + +/* debug */ static void print_stats(struct decoder_state *st); -static void reset_char(struct decoder_state *st); +/* error handling / usage */ static void err(int errval, const char *msg, ...); static void usage(void); static const char *progname(void); +/* portability (old systems) */ int getopt(int, char * const *, const char *); extern char *optarg; extern int optind; extern int opterr; extern int optopt; +#ifndef CHAR_BIT +#define CHAR_BIT 8 +#endif + +typedef char static_assert_char_is_8_bits[(CHAR_BIT == 8) ? 1 : -1]; +typedef char static_assert_char_is_1[(sizeof(char) == 1) ? 1 : -1]; +typedef char static_assert_short[(sizeof(short) == 2) ? 1 : -1]; +typedef char static_assert_int_is_4[(sizeof(int) >= 4) ? 1 : -1]; +typedef char static_assert_twos_complement[ + ((-1 & 3) == 3) ? 1 : -1 +]; + int main(int argc, char **argv) { @@ -302,21 +333,26 @@ handle_audio(struct decoder_state *st) int sample; /* - * If the modem signal disappears for several frames, - * discard the partially assembled character. + * If the modem signal disappears for several (read: 3) + * frames, discard the partially assembled character. */ - if (st->sample_count >= (3 * SAMPLES_PER_FRAME)) + if (st->sample_count >= (3 * SAMPLES_PER_FRAME) || + st->freq_separator <= 0) reset_char(st); - collect_separator_tone(st); - decode_pulse(st); + st->sample_count = 0; + + /* process exactly one frame */ + for (sample = 0; sample < SAMPLES_PER_FRAME; sample++) + decode_pulse(st); + + select_separator_tone(st); if (set_ascii_bit(st) < 0) print_char(st); - st->sample_count = 0; - for (sample = 0; sample < SAMPLES_PER_FRAME; sample++) - decode_pulse(st); + /* Detect tone per each frame */ + detect_tone(st); } /* @@ -324,14 +360,11 @@ handle_audio(struct decoder_state *st) * (and auto-adjust tolerances) */ static void -collect_separator_tone(struct decoder_state *st) +select_separator_tone(struct decoder_state *st) { int avg; - if (valid_signal(st)) - return; - - if (st->sep_samples >= 50 && st->freq_separator <= 0) + if (!is_valid_signal(st)) return; st->sep_sum += st->freq_separator; @@ -345,6 +378,10 @@ collect_separator_tone(struct decoder_state *st) st->sep_min = avg - SEP_TOLERANCE_PULSES; st->sep_max = avg + SEP_TOLERANCE_PULSES; + /* reset calibration accumulators */ + st->sep_sum = 0; + st->sep_samples = 0; + if (st->debug) printf("separator calibrated: %dHz\n", avg * FRAME_RATE); @@ -356,10 +393,55 @@ collect_separator_tone(struct decoder_state *st) * from being misinterpreted as data. */ static int -valid_signal(struct decoder_state *st) +is_valid_signal(struct decoder_state *st) +{ + if (st->freq_data <= 0) + return 0; + + if (st->freq_separator < st->sep_min || + st->freq_separator > st->sep_max) + return 0; + + return 1; +} + +/* + * Each validated frame contributes one bit of modem data. + * Bits are accumulated MSB-first into the ASCII byte. + */ +static int +set_ascii_bit(struct decoder_state *st) +{ + if (st->debug) + print_stats(st); + + if (!is_valid_signal(st)) + return st->ascii_bit; + + if (st->freq_data < st->freq_threshold) + st->ascii |= (1 << st->ascii_bit); + + st->ascii_bit--; + + return st->ascii_bit; +} + +static void +print_char(struct decoder_state *st) { - return (st->freq_separator > 0 && - st->freq_data > 0); + if (st->debug) + printf("<%c,%x>", st->ascii, st->ascii); + else + putchar(st->ascii); + + reset_char(st); +} + +static void +reset_char(struct decoder_state *st) +{ + st->ascii = 0; + st->ascii_bit = 7; } /* @@ -370,9 +452,11 @@ decode_pulse(struct decoder_state *st) { unsigned char old_ring, old_sep; unsigned char new_pulse; + signed short sample; int ringpos; int sep_pos; - signed short sample; + int diff_edge; + int diff_amp; ringpos = st->ringpos; sep_pos = st->sep_pos; @@ -408,18 +492,33 @@ decode_pulse(struct decoder_state *st) sample = read_sample(st); /* - * Convert the waveform sample into a pulse (0 or 1). + * Avoid startup edge. Since + * it's zero at startup, this + * may wrongly produce a pulse + */ + if (st->sample_count == 0) + st->prev_sample = sample; + + /* + * Detect edges instead of amplitude. + * This is more tolerant of weak microphones + * and speaker distortion.. * - * The unsigned comparison creates a small dead zone near zero, - * suppressing small amplitude noise from microphones or - * cheap ADCs. Real PC speaker tones are far outside this - * range, so they still produce clean pulses. + * However, we check both slope edges and + * amplitude, to mitagate noise. */ - if ((unsigned)(sample + THRESHOLD) - > (unsigned)(2 * THRESHOLD)) + diff_amp = sample; + diff_edge = sample - st->prev_sample; + if (diff_edge < 0) + diff_edge = -diff_edge; + if (diff_amp < 0) + diff_amp = -diff_amp; + if (diff_edge > THRESHOLD && + diff_amp > THRESHOLD) new_pulse = 1; else new_pulse = 0; + st->prev_sample = sample; st->pulse[ringpos] = new_pulse; st->freq_separator += new_pulse; @@ -429,55 +528,92 @@ decode_pulse(struct decoder_state *st) * The separator window always stays one frame ahead * of the data window. */ - ringpos++; - if (ringpos >= MAX_SAMPLES) + if (++ringpos >= MAX_SAMPLES) ringpos = 0; - sep_pos++; - if (sep_pos >= MAX_SAMPLES) + if (++sep_pos >= MAX_SAMPLES) sep_pos = 0; st->ringpos = ringpos; st->sep_pos = sep_pos; - /* - * Attempt to auto-detect spkmodem tone - */ - auto_detect_tone(st); - st->sample_count++; } +static signed short +read_sample(struct decoder_state *st) +{ + signed short sample; + unsigned short u; + + while (st->inpos >= st->inlen) + read_words(st); + + sample = st->inbuf[st->inpos++]; + + if (st->swap_bytes) { + u = (unsigned short)sample; + u = (u >> 8) | (u << 8); + + sample = (signed short)u; + } + + return sample; +} + +static void +read_words(struct decoder_state *st) +{ + size_t n; + + n = fread(st->inbuf, sizeof(st->inbuf[0]), + READ_BUF, stdin); + + if (n != 0) { + st->inpos = 0; + st->inlen = n; + + return; + } + + if (ferror(stdin)) + err(errno, "stdin read"); + if (feof(stdin)) + exit(EXIT_SUCCESS); +} + /* - * Observe signal for LEARN_SAMPLES samples (e.g. 1 second). - * The exact amount of time is determined by LEARN_SAMPLES - * divided by SAMPLE_RATE, logically. For example, if - * LEARN_SAMPLES were half of the SAMPLE_RATE, this - * corresponds to roughly 500ms before timeout. - * - * to guess the correct timing. If it fails, - * fall back to known good values. + * Automatically detect spkmodem tone */ static void -auto_detect_tone(struct decoder_state *st) +detect_tone(struct decoder_state *st) { - if (st->learn_samples >= LEARN_SAMPLES) + if (st->learn_frames >= LEARN_FRAMES) return; - st->learn_samples++; + st->learn_frames++; if (silent_signal(st)) return; select_low_tone(st); - if (st->learn_samples == LEARN_SAMPLES) { - st->freq_threshold = - (st->freq_min + st->freq_max) / 2; + if (st->learn_frames != LEARN_FRAMES) + return; - if (st->debug) - printf("auto threshold: %dHz\n", - st->freq_threshold * FRAME_RATE); - } + /* + * If the observed frequencies are too close, + * learning likely failed (only one tone seen). + * Keep the default threshold. + */ + if (st->freq_max - st->freq_min < 2) + return; + + st->freq_threshold = + (st->freq_min + st->freq_max) / 2; + + if (st->debug) + printf("auto threshold: %dHz\n", + st->freq_threshold * FRAME_RATE); } /* @@ -517,68 +653,6 @@ select_low_tone(struct decoder_state *st) st->freq_max = f; } -static signed short -read_sample(struct decoder_state *st) -{ - size_t n; - signed short sample; - unsigned short u; - - while (st->inpos >= st->inlen) { - - n = fread(st->inbuf, sizeof(st->inbuf[0]), - READ_BUF, stdin); - - if (n == 0) { - if (ferror(stdin)) - err(errno, "stdin read"); - if (feof(stdin)) - exit(EXIT_SUCCESS); - } - - st->inpos = 0; - st->inlen = n; - } - - sample = st->inbuf[st->inpos++]; - - if (st->swap_bytes) { - u = (unsigned short)sample; - u = (u >> 8) | (u << 8); - - sample = (signed short)u; - } - - return sample; -} - -/* - * Each validated frame contributes one bit of modem data. - * Bits are accumulated MSB-first into the ASCII byte. - */ -static int -set_ascii_bit(struct decoder_state *st) -{ - if (st->debug) - print_stats(st); - if (st->freq_data < st->freq_threshold) - st->ascii |= (1 << st->ascii_bit); - - st->ascii_bit--; - return st->ascii_bit; -} - -static void -print_char(struct decoder_state *st) -{ - if (st->debug) - printf("<%c,%x>", st->ascii, st->ascii); - else - putchar(st->ascii); - - reset_char(st); -} - static void print_stats(struct decoder_state *st) { @@ -613,13 +687,6 @@ print_stats(struct decoder_state *st) } static void -reset_char(struct decoder_state *st) -{ - st->ascii = 0; - st->ascii_bit = 7; -} - -static void err(int errval, const char *msg, ...) { va_list ap; |