spkmodem-decode: learn tone per frame, not sample

the fir filter produces stable frequencies per frame,
but learning per sample (within a frame) means we
record the same value roughly 240 times.

here, we are syncing up at just the right moment
instead, and only at that moment, this increasing
both performance and reliability.

Signed-off-by: Leah Rowe <leah@libreboot.org>

1 files changed, 72 insertions, 86 deletions

diff --git a/util/spkmodem_decode/spkmodem-decode.c b/util/spkmodem_decode/spkmodem-decode.c
index 8c57ff9a..0a22c6d5 100644
--- a/util/spkmodem_decode/spkmodem-decode.c
+++ b/util/spkmodem_decode/spkmodem-decode.c
@@ -157,15 +157,10 @@
 /*
  * 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
@@ -209,20 +204,20 @@ struct decoder_state {
 	int freq_min;
 	int freq_max;
 	int freq_threshold;
-	int learn_samples;
+	int learn_frames;
 };
 
 static const char *argv0;
 
 static int host_is_big_endian(void);
 static void handle_audio(struct decoder_state *st);
+static void auto_detect_tone(struct decoder_state *st);
+static int silent_signal(struct decoder_state *st);
+static void select_low_tone(struct decoder_state *st);
 static void collect_separator_tone(struct decoder_state *st);
 static int valid_signal(struct decoder_state *st);
 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 select_low_tone(struct decoder_state *st);
 static int set_ascii_bit(struct decoder_state *st);
 static void print_char(struct decoder_state *st);
 static void print_stats(struct decoder_state *st);
@@ -317,6 +312,72 @@ handle_audio(struct decoder_state *st)
 	st->sample_count = 0;
 	for (sample = 0; sample < SAMPLES_PER_FRAME; sample++)
 		decode_pulse(st);
+
+	/* Detect tone per each frame */
+	auto_detect_tone(st);
+}
+
+/*
+ * Automatically detect spkmodem tone
+ */
+static void
+auto_detect_tone(struct decoder_state *st)
+{
+	if (st->learn_frames >= LEARN_FRAMES)
+		return;
+
+	st->learn_frames++;
+
+	if (silent_signal(st))
+		return;
+
+	select_low_tone(st);
+
+	if (st->learn_frames == LEARN_FRAMES) {
+		st->freq_threshold =
+		    (st->freq_min + st->freq_max) / 2;
+
+		if (st->debug)
+			printf("auto threshold: %dHz\n",
+			    st->freq_threshold * FRAME_RATE);
+	}
+}
+
+/*
+ * Ignore silence / near silence.
+ * Both FIR windows will be near zero when no signal exists.
+ */
+static int
+silent_signal(struct decoder_state *st)
+{
+	return (st->freq_data <= 2 &&
+	    st->freq_separator <= 2);
+}
+
+/*
+ * Choose the lowest active tone.
+ * Separator frames carry tone in the separator window,
+ * data frames carry tone in the data window.
+ */
+static void
+select_low_tone(struct decoder_state *st)
+{
+	int f;
+
+	f = st->freq_data;
+
+	if (f <= 0 || (st->freq_separator > 0 &&
+	    st->freq_separator < f))
+		f = st->freq_separator;
+
+	if (f <= 0)
+		return;
+
+	if (f < st->freq_min)
+		st->freq_min = f;
+
+	if (f > st->freq_max)
+		st->freq_max = f;
 }
 
 /*
@@ -439,84 +500,9 @@ decode_pulse(struct decoder_state *st)
 	st->ringpos = ringpos;
 	st->sep_pos = sep_pos;
 
-	/*
-	 * Attempt to auto-detect spkmodem tone
-	 */
-	auto_detect_tone(st);
-
 	st->sample_count++;
 }
 
-/*
- * 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.
- */
-static void
-auto_detect_tone(struct decoder_state *st)
-{
-	if (st->learn_samples >= LEARN_SAMPLES)
-		return;
-
-	st->learn_samples++;
-
-	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->debug)
-			printf("auto threshold: %dHz\n",
-			    st->freq_threshold * FRAME_RATE);
-	}
-}
-
-/*
- * Ignore silence / near silence.
- * Both FIR windows will be near zero when no signal exists.
- */
-static int
-silent_signal(struct decoder_state *st)
-{
-	return (st->freq_data <= 2 &&
-	    st->freq_separator <= 2);
-}
-
-/*
- * Choose the lowest active tone.
- * Separator frames carry tone in the separator window,
- * data frames carry tone in the data window.
- */
-static void
-select_low_tone(struct decoder_state *st)
-{
-	int f;
-
-	f = st->freq_data;
-
-	if (f <= 0 || (st->freq_separator > 0 &&
-	    st->freq_separator < f))
-		f = st->freq_separator;
-
-	if (f <= 0)
-		return;
-
-	if (f < st->freq_min)
-		st->freq_min = f;
-
-	if (f > st->freq_max)
-		st->freq_max = f;
-}
-
 static signed short
 read_sample(struct decoder_state *st)
 {