[2] The classic, very useful paper that tells you how to
actually build a real world echo canceller:
- Messerschmitt, Hedberg, Cole, Haoui, Winship, "Digital Voice
- Echo Canceller with a TMS320020,
- http://www.rowetel.com/images/echo/spra129.pdf
+ Messerschmitt, Hedberg, Cole, Haoui, Winship, "Digital Voice
+ Echo Canceller with a TMS320020,
+ http://www.rowetel.com/images/echo/spra129.pdf
[3] I have written a series of blog posts on this work, here is
Part 1: http://www.rowetel.com/blog/?p=18
[4] The source code http://svn.rowetel.com/software/oslec/
[5] A nice reference on LMS filters:
- http://en.wikipedia.org/wiki/Least_mean_squares_filter
+ http://en.wikipedia.org/wiki/Least_mean_squares_filter
Credits:
Mark, Pawel, and Pavel.
*/
-#include <linux/kernel.h> /* We're doing kernel work */
+#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/slab.h>
#include "bit_operations.h"
#include "echo.h"
-#define MIN_TX_POWER_FOR_ADAPTION 64
-#define MIN_RX_POWER_FOR_ADAPTION 64
-#define DTD_HANGOVER 600 /* 600 samples, or 75ms */
-#define DC_LOG2BETA 3 /* log2() of DC filter Beta */
+#define MIN_TX_POWER_FOR_ADAPTION 64
+#define MIN_RX_POWER_FOR_ADAPTION 64
+#define DTD_HANGOVER 600 /* 600 samples, or 75ms */
+#define DC_LOG2BETA 3 /* log2() of DC filter Beta */
-/*-----------------------------------------------------------------------*\
- FUNCTIONS
-\*-----------------------------------------------------------------------*/
/* adapting coeffs using the traditional stochastic descent (N)LMS algorithm */
}
EXPORT_SYMBOL_GPL(oslec_snapshot);
-/* Dual Path Echo Canceller ------------------------------------------------*/
+/* Dual Path Echo Canceller */
int16_t oslec_update(struct oslec_state *ec, int16_t tx, int16_t rx)
{
int clean_bg;
int tmp, tmp1;
- /* Input scaling was found be required to prevent problems when tx
- starts clipping. Another possible way to handle this would be the
- filter coefficent scaling. */
+ /*
+ * Input scaling was found be required to prevent problems when tx
+ * starts clipping. Another possible way to handle this would be the
+ * filter coefficent scaling.
+ */
ec->tx = tx;
ec->rx = rx;
rx >>= 1;
/*
- Filter DC, 3dB point is 160Hz (I think), note 32 bit precision required
- otherwise values do not track down to 0. Zero at DC, Pole at (1-Beta)
- only real axis. Some chip sets (like Si labs) don't need
- this, but something like a $10 X100P card does. Any DC really slows
- down convergence.
-
- Note: removes some low frequency from the signal, this reduces
- the speech quality when listening to samples through headphones
- but may not be obvious through a telephone handset.
-
- Note that the 3dB frequency in radians is approx Beta, e.g. for
- Beta = 2^(-3) = 0.125, 3dB freq is 0.125 rads = 159Hz.
+ * Filter DC, 3dB point is 160Hz (I think), note 32 bit precision
+ * required otherwise values do not track down to 0. Zero at DC, Pole
+ * at (1-Beta) only real axis. Some chip sets (like Si labs) don't
+ * need this, but something like a $10 X100P card does. Any DC really
+ * slows down convergence.
+ *
+ * Note: removes some low frequency from the signal, this reduces the
+ * speech quality when listening to samples through headphones but may
+ * not be obvious through a telephone handset.
+ *
+ * Note that the 3dB frequency in radians is approx Beta, e.g. for Beta
+ * = 2^(-3) = 0.125, 3dB freq is 0.125 rads = 159Hz.
*/
if (ec->adaption_mode & ECHO_CAN_USE_RX_HPF) {
tmp = rx << 15;
#if 1
- /* Make sure the gain of the HPF is 1.0. This can still saturate a little under
- impulse conditions, and it might roll to 32768 and need clipping on sustained peak
- level signals. However, the scale of such clipping is small, and the error due to
- any saturation should not markedly affect the downstream processing. */
+ /*
+ * Make sure the gain of the HPF is 1.0. This can still
+ * saturate a little under impulse conditions, and it might
+ * roll to 32768 and need clipping on sustained peak level
+ * signals. However, the scale of such clipping is small, and
+ * the error due to any saturation should not markedly affect
+ * the downstream processing.
+ */
tmp -= (tmp >> 4);
#endif
ec->rx_1 += -(ec->rx_1 >> DC_LOG2BETA) + tmp - ec->rx_2;
- /* hard limit filter to prevent clipping. Note that at this stage
- rx should be limited to +/- 16383 due to right shift above */
+ /*
+ * hard limit filter to prevent clipping. Note that at this
+ * stage rx should be limited to +/- 16383 due to right shift
+ * above
+ */
tmp1 = ec->rx_1 >> 15;
if (tmp1 > 16383)
tmp1 = 16383;
ec->Lrxacc += abs(rx) - ec->Lrx;
ec->Lrx = (ec->Lrxacc + (1 << 4)) >> 5;
- /* Foreground filter --------------------------------------------------- */
+ /* Foreground filter */
ec->fir_state.coeffs = ec->fir_taps16[0];
echo_value = fir16(&ec->fir_state, tx);
ec->Lcleanacc += abs(ec->clean) - ec->Lclean;
ec->Lclean = (ec->Lcleanacc + (1 << 4)) >> 5;
- /* Background filter --------------------------------------------------- */
+ /* Background filter */
echo_value = fir16(&ec->fir_state_bg, tx);
clean_bg = rx - echo_value;
ec->Lclean_bgacc += abs(clean_bg) - ec->Lclean_bg;
ec->Lclean_bg = (ec->Lclean_bgacc + (1 << 4)) >> 5;
- /* Background Filter adaption ----------------------------------------- */
+ /* Background Filter adaption */
/* Almost always adap bg filter, just simple DT and energy
detection to minimise adaption in cases of strong double talk.
if (ec->nonupdate_dwell)
ec->nonupdate_dwell--;
- /* Transfer logic ------------------------------------------------------ */
+ /* Transfer logic */
/* These conditions are from the dual path paper [1], I messed with
them a bit to improve performance. */
/* (ec->Lclean_bg < 0.125*ec->Ltx) */
(8 * ec->Lclean_bg < ec->Ltx)) {
if (ec->cond_met == 6) {
- /* BG filter has had better results for 6 consecutive samples */
+ /*
+ * BG filter has had better results for 6 consecutive
+ * samples
+ */
ec->adapt = 1;
memcpy(ec->fir_taps16[0], ec->fir_taps16[1],
ec->taps * sizeof(int16_t));
} else
ec->cond_met = 0;
- /* Non-Linear Processing --------------------------------------------------- */
+ /* Non-Linear Processing */
ec->clean_nlp = ec->clean;
if (ec->adaption_mode & ECHO_CAN_USE_NLP) {
- /* Non-linear processor - a fancy way to say "zap small signals, to avoid
- residual echo due to (uLaw/ALaw) non-linearity in the channel.". */
+ /*
+ * Non-linear processor - a fancy way to say "zap small
+ * signals, to avoid residual echo due to (uLaw/ALaw)
+ * non-linearity in the channel.".
+ */
if ((16 * ec->Lclean < ec->Ltx)) {
- /* Our e/c has improved echo by at least 24 dB (each factor of 2 is 6dB,
- so 2*2*2*2=16 is the same as 6+6+6+6=24dB) */
+ /*
+ * Our e/c has improved echo by at least 24 dB (each
+ * factor of 2 is 6dB, so 2*2*2*2=16 is the same as
+ * 6+6+6+6=24dB)
+ */
if (ec->adaption_mode & ECHO_CAN_USE_CNG) {
ec->cng_level = ec->Lbgn;
- /* Very elementary comfort noise generation. Just random
- numbers rolled off very vaguely Hoth-like. DR: This
- noise doesn't sound quite right to me - I suspect there
- are some overlfow issues in the filtering as it's too
- "crackly". TODO: debug this, maybe just play noise at
- high level or look at spectrum.
+ /*
+ * Very elementary comfort noise generation.
+ * Just random numbers rolled off very vaguely
+ * Hoth-like. DR: This noise doesn't sound
+ * quite right to me - I suspect there are some
+ * overlfow issues in the filtering as it's too
+ * "crackly".
+ * TODO: debug this, maybe just play noise at
+ * high level or look at spectrum.
*/
ec->cng_rndnum =
if (ec->clean_nlp < -ec->Lbgn)
ec->clean_nlp = -ec->Lbgn;
} else {
- /* just mute the residual, doesn't sound very good, used mainly
- in G168 tests */
+ /*
+ * just mute the residual, doesn't sound very
+ * good, used mainly in G168 tests
+ */
ec->clean_nlp = 0;
}
} else {
- /* Background noise estimator. I tried a few algorithms
- here without much luck. This very simple one seems to
- work best, we just average the level using a slow (1 sec
- time const) filter if the current level is less than a
- (experimentally derived) constant. This means we dont
- include high level signals like near end speech. When
- combined with CNG or especially CLIP seems to work OK.
+ /*
+ * Background noise estimator. I tried a few
+ * algorithms here without much luck. This very simple
+ * one seems to work best, we just average the level
+ * using a slow (1 sec time const) filter if the
+ * current level is less than a (experimentally
+ * derived) constant. This means we dont include high
+ * level signals like near end speech. When combined
+ * with CNG or especially CLIP seems to work OK.
*/
if (ec->Lclean < 40) {
ec->Lbgn_acc += abs(ec->clean) - ec->Lbgn;
It can also help by removing and DC in the tx signal. DC is bad
for LMS algorithms.
- This is one of the classic DC removal filters, adjusted to provide sufficient
- bass rolloff to meet the above requirement to protect hybrids from things that
- upset them. The difference between successive samples produces a lousy HPF, and
- then a suitably placed pole flattens things out. The final result is a nicely
- rolled off bass end. The filtering is implemented with extended fractional
- precision, which noise shapes things, giving very clean DC removal.
+ This is one of the classic DC removal filters, adjusted to provide
+ sufficient bass rolloff to meet the above requirement to protect hybrids
+ from things that upset them. The difference between successive samples
+ produces a lousy HPF, and then a suitably placed pole flattens things out.
+ The final result is a nicely rolled off bass end. The filtering is
+ implemented with extended fractional precision, which noise shapes things,
+ giving very clean DC removal.
*/
int16_t oslec_hpf_tx(struct oslec_state *ec, int16_t tx)
if (ec->adaption_mode & ECHO_CAN_USE_TX_HPF) {
tmp = tx << 15;
#if 1
- /* Make sure the gain of the HPF is 1.0. The first can still saturate a little under
- impulse conditions, and it might roll to 32768 and need clipping on sustained peak
- level signals. However, the scale of such clipping is small, and the error due to
- any saturation should not markedly affect the downstream processing. */
+ /*
+ * Make sure the gain of the HPF is 1.0. The first can still
+ * saturate a little under impulse conditions, and it might
+ * roll to 32768 and need clipping on sustained peak level
+ * signals. However, the scale of such clipping is small, and
+ * the error due to any saturation should not markedly affect
+ * the downstream processing.
+ */
tmp -= (tmp >> 4);
#endif
ec->tx_1 += -(ec->tx_1 >> DC_LOG2BETA) + tmp - ec->tx_2;
projects / karo-tx-linux.git / commitdiff