PageRenderTime 41ms CodeModel.GetById 19ms app.highlight 17ms RepoModel.GetById 0ms app.codeStats 1ms

/drivers/staging/echo/echo.c

https://bitbucket.org/wisechild/galaxy-nexus
C | 666 lines | 272 code | 95 blank | 299 comment | 43 complexity | 7b5faf5cf303616e9659ad8edb25d13c MD5 | raw file
Possible License(s): GPL-2.0, LGPL-2.0, AGPL-1.0
  1/*
  2 * SpanDSP - a series of DSP components for telephony
  3 *
  4 * echo.c - A line echo canceller.  This code is being developed
  5 *          against and partially complies with G168.
  6 *
  7 * Written by Steve Underwood <steveu@coppice.org>
  8 *         and David Rowe <david_at_rowetel_dot_com>
  9 *
 10 * Copyright (C) 2001, 2003 Steve Underwood, 2007 David Rowe
 11 *
 12 * Based on a bit from here, a bit from there, eye of toad, ear of
 13 * bat, 15 years of failed attempts by David and a few fried brain
 14 * cells.
 15 *
 16 * All rights reserved.
 17 *
 18 * This program is free software; you can redistribute it and/or modify
 19 * it under the terms of the GNU General Public License version 2, as
 20 * published by the Free Software Foundation.
 21 *
 22 * This program is distributed in the hope that it will be useful,
 23 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 24 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 25 * GNU General Public License for more details.
 26 *
 27 * You should have received a copy of the GNU General Public License
 28 * along with this program; if not, write to the Free Software
 29 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 30 */
 31
 32/*! \file */
 33
 34/* Implementation Notes
 35   David Rowe
 36   April 2007
 37
 38   This code started life as Steve's NLMS algorithm with a tap
 39   rotation algorithm to handle divergence during double talk.  I
 40   added a Geigel Double Talk Detector (DTD) [2] and performed some
 41   G168 tests.  However I had trouble meeting the G168 requirements,
 42   especially for double talk - there were always cases where my DTD
 43   failed, for example where near end speech was under the 6dB
 44   threshold required for declaring double talk.
 45
 46   So I tried a two path algorithm [1], which has so far given better
 47   results.  The original tap rotation/Geigel algorithm is available
 48   in SVN http://svn.rowetel.com/software/oslec/tags/before_16bit.
 49   It's probably possible to make it work if some one wants to put some
 50   serious work into it.
 51
 52   At present no special treatment is provided for tones, which
 53   generally cause NLMS algorithms to diverge.  Initial runs of a
 54   subset of the G168 tests for tones (e.g ./echo_test 6) show the
 55   current algorithm is passing OK, which is kind of surprising.  The
 56   full set of tests needs to be performed to confirm this result.
 57
 58   One other interesting change is that I have managed to get the NLMS
 59   code to work with 16 bit coefficients, rather than the original 32
 60   bit coefficents.  This reduces the MIPs and storage required.
 61   I evaulated the 16 bit port using g168_tests.sh and listening tests
 62   on 4 real-world samples.
 63
 64   I also attempted the implementation of a block based NLMS update
 65   [2] but although this passes g168_tests.sh it didn't converge well
 66   on the real-world samples.  I have no idea why, perhaps a scaling
 67   problem.  The block based code is also available in SVN
 68   http://svn.rowetel.com/software/oslec/tags/before_16bit.  If this
 69   code can be debugged, it will lead to further reduction in MIPS, as
 70   the block update code maps nicely onto DSP instruction sets (it's a
 71   dot product) compared to the current sample-by-sample update.
 72
 73   Steve also has some nice notes on echo cancellers in echo.h
 74
 75   References:
 76
 77   [1] Ochiai, Areseki, and Ogihara, "Echo Canceller with Two Echo
 78       Path Models", IEEE Transactions on communications, COM-25,
 79       No. 6, June
 80       1977.
 81       http://www.rowetel.com/images/echo/dual_path_paper.pdf
 82
 83   [2] The classic, very useful paper that tells you how to
 84       actually build a real world echo canceller:
 85	 Messerschmitt, Hedberg, Cole, Haoui, Winship, "Digital Voice
 86	 Echo Canceller with a TMS320020,
 87	 http://www.rowetel.com/images/echo/spra129.pdf
 88
 89   [3] I have written a series of blog posts on this work, here is
 90       Part 1: http://www.rowetel.com/blog/?p=18
 91
 92   [4] The source code http://svn.rowetel.com/software/oslec/
 93
 94   [5] A nice reference on LMS filters:
 95	 http://en.wikipedia.org/wiki/Least_mean_squares_filter
 96
 97   Credits:
 98
 99   Thanks to Steve Underwood, Jean-Marc Valin, and Ramakrishnan
100   Muthukrishnan for their suggestions and email discussions.  Thanks
101   also to those people who collected echo samples for me such as
102   Mark, Pawel, and Pavel.
103*/
104
105#include <linux/kernel.h>
106#include <linux/module.h>
107#include <linux/slab.h>
108
109#include "echo.h"
110
111#define MIN_TX_POWER_FOR_ADAPTION	64
112#define MIN_RX_POWER_FOR_ADAPTION	64
113#define DTD_HANGOVER			600	/* 600 samples, or 75ms     */
114#define DC_LOG2BETA			3	/* log2() of DC filter Beta */
115
116/* adapting coeffs using the traditional stochastic descent (N)LMS algorithm */
117
118#ifdef __bfin__
119static inline void lms_adapt_bg(struct oslec_state *ec, int clean, int shift)
120{
121	int i, j;
122	int offset1;
123	int offset2;
124	int factor;
125	int exp;
126	int16_t *phist;
127	int n;
128
129	if (shift > 0)
130		factor = clean << shift;
131	else
132		factor = clean >> -shift;
133
134	/* Update the FIR taps */
135
136	offset2 = ec->curr_pos;
137	offset1 = ec->taps - offset2;
138	phist = &ec->fir_state_bg.history[offset2];
139
140	/* st: and en: help us locate the assembler in echo.s */
141
142	/* asm("st:"); */
143	n = ec->taps;
144	for (i = 0, j = offset2; i < n; i++, j++) {
145		exp = *phist++ * factor;
146		ec->fir_taps16[1][i] += (int16_t) ((exp + (1 << 14)) >> 15);
147	}
148	/* asm("en:"); */
149
150	/* Note the asm for the inner loop above generated by Blackfin gcc
151	   4.1.1 is pretty good (note even parallel instructions used):
152
153	   R0 = W [P0++] (X);
154	   R0 *= R2;
155	   R0 = R0 + R3 (NS) ||
156	   R1 = W [P1] (X) ||
157	   nop;
158	   R0 >>>= 15;
159	   R0 = R0 + R1;
160	   W [P1++] = R0;
161
162	   A block based update algorithm would be much faster but the
163	   above can't be improved on much.  Every instruction saved in
164	   the loop above is 2 MIPs/ch!  The for loop above is where the
165	   Blackfin spends most of it's time - about 17 MIPs/ch measured
166	   with speedtest.c with 256 taps (32ms).  Write-back and
167	   Write-through cache gave about the same performance.
168	 */
169}
170
171/*
172   IDEAS for further optimisation of lms_adapt_bg():
173
174   1/ The rounding is quite costly.  Could we keep as 32 bit coeffs
175   then make filter pluck the MS 16-bits of the coeffs when filtering?
176   However this would lower potential optimisation of filter, as I
177   think the dual-MAC architecture requires packed 16 bit coeffs.
178
179   2/ Block based update would be more efficient, as per comments above,
180   could use dual MAC architecture.
181
182   3/ Look for same sample Blackfin LMS code, see if we can get dual-MAC
183   packing.
184
185   4/ Execute the whole e/c in a block of say 20ms rather than sample
186   by sample.  Processing a few samples every ms is inefficient.
187*/
188
189#else
190static inline void lms_adapt_bg(struct oslec_state *ec, int clean, int shift)
191{
192	int i;
193
194	int offset1;
195	int offset2;
196	int factor;
197	int exp;
198
199	if (shift > 0)
200		factor = clean << shift;
201	else
202		factor = clean >> -shift;
203
204	/* Update the FIR taps */
205
206	offset2 = ec->curr_pos;
207	offset1 = ec->taps - offset2;
208
209	for (i = ec->taps - 1; i >= offset1; i--) {
210		exp = (ec->fir_state_bg.history[i - offset1] * factor);
211		ec->fir_taps16[1][i] += (int16_t) ((exp + (1 << 14)) >> 15);
212	}
213	for (; i >= 0; i--) {
214		exp = (ec->fir_state_bg.history[i + offset2] * factor);
215		ec->fir_taps16[1][i] += (int16_t) ((exp + (1 << 14)) >> 15);
216	}
217}
218#endif
219
220static inline int top_bit(unsigned int bits)
221{
222	if (bits == 0)
223		return -1;
224	else
225		return (int)fls((int32_t) bits) - 1;
226}
227
228struct oslec_state *oslec_create(int len, int adaption_mode)
229{
230	struct oslec_state *ec;
231	int i;
232
233	ec = kzalloc(sizeof(*ec), GFP_KERNEL);
234	if (!ec)
235		return NULL;
236
237	ec->taps = len;
238	ec->log2taps = top_bit(len);
239	ec->curr_pos = ec->taps - 1;
240
241	for (i = 0; i < 2; i++) {
242		ec->fir_taps16[i] =
243		    kcalloc(ec->taps, sizeof(int16_t), GFP_KERNEL);
244		if (!ec->fir_taps16[i])
245			goto error_oom;
246	}
247
248	fir16_create(&ec->fir_state, ec->fir_taps16[0], ec->taps);
249	fir16_create(&ec->fir_state_bg, ec->fir_taps16[1], ec->taps);
250
251	for (i = 0; i < 5; i++)
252		ec->xvtx[i] = ec->yvtx[i] = ec->xvrx[i] = ec->yvrx[i] = 0;
253
254	ec->cng_level = 1000;
255	oslec_adaption_mode(ec, adaption_mode);
256
257	ec->snapshot = kcalloc(ec->taps, sizeof(int16_t), GFP_KERNEL);
258	if (!ec->snapshot)
259		goto error_oom;
260
261	ec->cond_met = 0;
262	ec->Pstates = 0;
263	ec->Ltxacc = ec->Lrxacc = ec->Lcleanacc = ec->Lclean_bgacc = 0;
264	ec->Ltx = ec->Lrx = ec->Lclean = ec->Lclean_bg = 0;
265	ec->tx_1 = ec->tx_2 = ec->rx_1 = ec->rx_2 = 0;
266	ec->Lbgn = ec->Lbgn_acc = 0;
267	ec->Lbgn_upper = 200;
268	ec->Lbgn_upper_acc = ec->Lbgn_upper << 13;
269
270	return ec;
271
272error_oom:
273	for (i = 0; i < 2; i++)
274		kfree(ec->fir_taps16[i]);
275
276	kfree(ec);
277	return NULL;
278}
279
280EXPORT_SYMBOL_GPL(oslec_create);
281
282void oslec_free(struct oslec_state *ec)
283{
284	int i;
285
286	fir16_free(&ec->fir_state);
287	fir16_free(&ec->fir_state_bg);
288	for (i = 0; i < 2; i++)
289		kfree(ec->fir_taps16[i]);
290	kfree(ec->snapshot);
291	kfree(ec);
292}
293
294EXPORT_SYMBOL_GPL(oslec_free);
295
296void oslec_adaption_mode(struct oslec_state *ec, int adaption_mode)
297{
298	ec->adaption_mode = adaption_mode;
299}
300
301EXPORT_SYMBOL_GPL(oslec_adaption_mode);
302
303void oslec_flush(struct oslec_state *ec)
304{
305	int i;
306
307	ec->Ltxacc = ec->Lrxacc = ec->Lcleanacc = ec->Lclean_bgacc = 0;
308	ec->Ltx = ec->Lrx = ec->Lclean = ec->Lclean_bg = 0;
309	ec->tx_1 = ec->tx_2 = ec->rx_1 = ec->rx_2 = 0;
310
311	ec->Lbgn = ec->Lbgn_acc = 0;
312	ec->Lbgn_upper = 200;
313	ec->Lbgn_upper_acc = ec->Lbgn_upper << 13;
314
315	ec->nonupdate_dwell = 0;
316
317	fir16_flush(&ec->fir_state);
318	fir16_flush(&ec->fir_state_bg);
319	ec->fir_state.curr_pos = ec->taps - 1;
320	ec->fir_state_bg.curr_pos = ec->taps - 1;
321	for (i = 0; i < 2; i++)
322		memset(ec->fir_taps16[i], 0, ec->taps * sizeof(int16_t));
323
324	ec->curr_pos = ec->taps - 1;
325	ec->Pstates = 0;
326}
327
328EXPORT_SYMBOL_GPL(oslec_flush);
329
330void oslec_snapshot(struct oslec_state *ec)
331{
332	memcpy(ec->snapshot, ec->fir_taps16[0], ec->taps * sizeof(int16_t));
333}
334
335EXPORT_SYMBOL_GPL(oslec_snapshot);
336
337/* Dual Path Echo Canceller */
338
339int16_t oslec_update(struct oslec_state *ec, int16_t tx, int16_t rx)
340{
341	int32_t echo_value;
342	int clean_bg;
343	int tmp, tmp1;
344
345	/*
346	 * Input scaling was found be required to prevent problems when tx
347	 * starts clipping.  Another possible way to handle this would be the
348	 * filter coefficent scaling.
349	 */
350
351	ec->tx = tx;
352	ec->rx = rx;
353	tx >>= 1;
354	rx >>= 1;
355
356	/*
357	 * Filter DC, 3dB point is 160Hz (I think), note 32 bit precision
358	 * required otherwise values do not track down to 0. Zero at DC, Pole
359	 * at (1-Beta) on real axis.  Some chip sets (like Si labs) don't
360	 * need this, but something like a $10 X100P card does.  Any DC really
361	 * slows down convergence.
362	 *
363	 * Note: removes some low frequency from the signal, this reduces the
364	 * speech quality when listening to samples through headphones but may
365	 * not be obvious through a telephone handset.
366	 *
367	 * Note that the 3dB frequency in radians is approx Beta, e.g. for Beta
368	 * = 2^(-3) = 0.125, 3dB freq is 0.125 rads = 159Hz.
369	 */
370
371	if (ec->adaption_mode & ECHO_CAN_USE_RX_HPF) {
372		tmp = rx << 15;
373
374		/*
375		 * Make sure the gain of the HPF is 1.0. This can still
376		 * saturate a little under impulse conditions, and it might
377		 * roll to 32768 and need clipping on sustained peak level
378		 * signals. However, the scale of such clipping is small, and
379		 * the error due to any saturation should not markedly affect
380		 * the downstream processing.
381		 */
382		tmp -= (tmp >> 4);
383
384		ec->rx_1 += -(ec->rx_1 >> DC_LOG2BETA) + tmp - ec->rx_2;
385
386		/*
387		 * hard limit filter to prevent clipping.  Note that at this
388		 * stage rx should be limited to +/- 16383 due to right shift
389		 * above
390		 */
391		tmp1 = ec->rx_1 >> 15;
392		if (tmp1 > 16383)
393			tmp1 = 16383;
394		if (tmp1 < -16383)
395			tmp1 = -16383;
396		rx = tmp1;
397		ec->rx_2 = tmp;
398	}
399
400	/* Block average of power in the filter states.  Used for
401	   adaption power calculation. */
402
403	{
404		int new, old;
405
406		/* efficient "out with the old and in with the new" algorithm so
407		   we don't have to recalculate over the whole block of
408		   samples. */
409		new = (int)tx *(int)tx;
410		old = (int)ec->fir_state.history[ec->fir_state.curr_pos] *
411		    (int)ec->fir_state.history[ec->fir_state.curr_pos];
412		ec->Pstates +=
413		    ((new - old) + (1 << (ec->log2taps - 1))) >> ec->log2taps;
414		if (ec->Pstates < 0)
415			ec->Pstates = 0;
416	}
417
418	/* Calculate short term average levels using simple single pole IIRs */
419
420	ec->Ltxacc += abs(tx) - ec->Ltx;
421	ec->Ltx = (ec->Ltxacc + (1 << 4)) >> 5;
422	ec->Lrxacc += abs(rx) - ec->Lrx;
423	ec->Lrx = (ec->Lrxacc + (1 << 4)) >> 5;
424
425	/* Foreground filter */
426
427	ec->fir_state.coeffs = ec->fir_taps16[0];
428	echo_value = fir16(&ec->fir_state, tx);
429	ec->clean = rx - echo_value;
430	ec->Lcleanacc += abs(ec->clean) - ec->Lclean;
431	ec->Lclean = (ec->Lcleanacc + (1 << 4)) >> 5;
432
433	/* Background filter */
434
435	echo_value = fir16(&ec->fir_state_bg, tx);
436	clean_bg = rx - echo_value;
437	ec->Lclean_bgacc += abs(clean_bg) - ec->Lclean_bg;
438	ec->Lclean_bg = (ec->Lclean_bgacc + (1 << 4)) >> 5;
439
440	/* Background Filter adaption */
441
442	/* Almost always adap bg filter, just simple DT and energy
443	   detection to minimise adaption in cases of strong double talk.
444	   However this is not critical for the dual path algorithm.
445	 */
446	ec->factor = 0;
447	ec->shift = 0;
448	if ((ec->nonupdate_dwell == 0)) {
449		int P, logP, shift;
450
451		/* Determine:
452
453		   f = Beta * clean_bg_rx/P ------ (1)
454
455		   where P is the total power in the filter states.
456
457		   The Boffins have shown that if we obey (1) we converge
458		   quickly and avoid instability.
459
460		   The correct factor f must be in Q30, as this is the fixed
461		   point format required by the lms_adapt_bg() function,
462		   therefore the scaled version of (1) is:
463
464		   (2^30) * f  = (2^30) * Beta * clean_bg_rx/P
465		   factor      = (2^30) * Beta * clean_bg_rx/P     ----- (2)
466
467		   We have chosen Beta = 0.25 by experiment, so:
468
469		   factor      = (2^30) * (2^-2) * clean_bg_rx/P
470
471		   (30 - 2 - log2(P))
472		   factor      = clean_bg_rx 2                     ----- (3)
473
474		   To avoid a divide we approximate log2(P) as top_bit(P),
475		   which returns the position of the highest non-zero bit in
476		   P.  This approximation introduces an error as large as a
477		   factor of 2, but the algorithm seems to handle it OK.
478
479		   Come to think of it a divide may not be a big deal on a
480		   modern DSP, so its probably worth checking out the cycles
481		   for a divide versus a top_bit() implementation.
482		 */
483
484		P = MIN_TX_POWER_FOR_ADAPTION + ec->Pstates;
485		logP = top_bit(P) + ec->log2taps;
486		shift = 30 - 2 - logP;
487		ec->shift = shift;
488
489		lms_adapt_bg(ec, clean_bg, shift);
490	}
491
492	/* very simple DTD to make sure we dont try and adapt with strong
493	   near end speech */
494
495	ec->adapt = 0;
496	if ((ec->Lrx > MIN_RX_POWER_FOR_ADAPTION) && (ec->Lrx > ec->Ltx))
497		ec->nonupdate_dwell = DTD_HANGOVER;
498	if (ec->nonupdate_dwell)
499		ec->nonupdate_dwell--;
500
501	/* Transfer logic */
502
503	/* These conditions are from the dual path paper [1], I messed with
504	   them a bit to improve performance. */
505
506	if ((ec->adaption_mode & ECHO_CAN_USE_ADAPTION) &&
507	    (ec->nonupdate_dwell == 0) &&
508	    /* (ec->Lclean_bg < 0.875*ec->Lclean) */
509	    (8 * ec->Lclean_bg < 7 * ec->Lclean) &&
510	    /* (ec->Lclean_bg < 0.125*ec->Ltx) */
511	    (8 * ec->Lclean_bg < ec->Ltx)) {
512		if (ec->cond_met == 6) {
513			/*
514			 * BG filter has had better results for 6 consecutive
515			 * samples
516			 */
517			ec->adapt = 1;
518			memcpy(ec->fir_taps16[0], ec->fir_taps16[1],
519			       ec->taps * sizeof(int16_t));
520		} else
521			ec->cond_met++;
522	} else
523		ec->cond_met = 0;
524
525	/* Non-Linear Processing */
526
527	ec->clean_nlp = ec->clean;
528	if (ec->adaption_mode & ECHO_CAN_USE_NLP) {
529		/*
530		 * Non-linear processor - a fancy way to say "zap small
531		 * signals, to avoid residual echo due to (uLaw/ALaw)
532		 * non-linearity in the channel.".
533		 */
534
535		if ((16 * ec->Lclean < ec->Ltx)) {
536			/*
537			 * Our e/c has improved echo by at least 24 dB (each
538			 * factor of 2 is 6dB, so 2*2*2*2=16 is the same as
539			 * 6+6+6+6=24dB)
540			 */
541			if (ec->adaption_mode & ECHO_CAN_USE_CNG) {
542				ec->cng_level = ec->Lbgn;
543
544				/*
545				 * Very elementary comfort noise generation.
546				 * Just random numbers rolled off very vaguely
547				 * Hoth-like.  DR: This noise doesn't sound
548				 * quite right to me - I suspect there are some
549				 * overflow issues in the filtering as it's too
550				 * "crackly".
551				 * TODO: debug this, maybe just play noise at
552				 * high level or look at spectrum.
553				 */
554
555				ec->cng_rndnum =
556				    1664525U * ec->cng_rndnum + 1013904223U;
557				ec->cng_filter =
558				    ((ec->cng_rndnum & 0xFFFF) - 32768 +
559				     5 * ec->cng_filter) >> 3;
560				ec->clean_nlp =
561				    (ec->cng_filter * ec->cng_level * 8) >> 14;
562
563			} else if (ec->adaption_mode & ECHO_CAN_USE_CLIP) {
564				/* This sounds much better than CNG */
565				if (ec->clean_nlp > ec->Lbgn)
566					ec->clean_nlp = ec->Lbgn;
567				if (ec->clean_nlp < -ec->Lbgn)
568					ec->clean_nlp = -ec->Lbgn;
569			} else {
570				/*
571				 * just mute the residual, doesn't sound very
572				 * good, used mainly in G168 tests
573				 */
574				ec->clean_nlp = 0;
575			}
576		} else {
577			/*
578			 * Background noise estimator.  I tried a few
579			 * algorithms here without much luck.  This very simple
580			 * one seems to work best, we just average the level
581			 * using a slow (1 sec time const) filter if the
582			 * current level is less than a (experimentally
583			 * derived) constant.  This means we dont include high
584			 * level signals like near end speech.  When combined
585			 * with CNG or especially CLIP seems to work OK.
586			 */
587			if (ec->Lclean < 40) {
588				ec->Lbgn_acc += abs(ec->clean) - ec->Lbgn;
589				ec->Lbgn = (ec->Lbgn_acc + (1 << 11)) >> 12;
590			}
591		}
592	}
593
594	/* Roll around the taps buffer */
595	if (ec->curr_pos <= 0)
596		ec->curr_pos = ec->taps;
597	ec->curr_pos--;
598
599	if (ec->adaption_mode & ECHO_CAN_DISABLE)
600		ec->clean_nlp = rx;
601
602	/* Output scaled back up again to match input scaling */
603
604	return (int16_t) ec->clean_nlp << 1;
605}
606
607EXPORT_SYMBOL_GPL(oslec_update);
608
609/* This function is separated from the echo canceller is it is usually called
610   as part of the tx process.  See rx HP (DC blocking) filter above, it's
611   the same design.
612
613   Some soft phones send speech signals with a lot of low frequency
614   energy, e.g. down to 20Hz.  This can make the hybrid non-linear
615   which causes the echo canceller to fall over.  This filter can help
616   by removing any low frequency before it gets to the tx port of the
617   hybrid.
618
619   It can also help by removing and DC in the tx signal.  DC is bad
620   for LMS algorithms.
621
622   This is one of the classic DC removal filters, adjusted to provide
623   sufficient bass rolloff to meet the above requirement to protect hybrids
624   from things that upset them. The difference between successive samples
625   produces a lousy HPF, and then a suitably placed pole flattens things out.
626   The final result is a nicely rolled off bass end. The filtering is
627   implemented with extended fractional precision, which noise shapes things,
628   giving very clean DC removal.
629*/
630
631int16_t oslec_hpf_tx(struct oslec_state * ec, int16_t tx)
632{
633	int tmp, tmp1;
634
635	if (ec->adaption_mode & ECHO_CAN_USE_TX_HPF) {
636		tmp = tx << 15;
637
638		/*
639		 * Make sure the gain of the HPF is 1.0. The first can still
640		 * saturate a little under impulse conditions, and it might
641		 * roll to 32768 and need clipping on sustained peak level
642		 * signals. However, the scale of such clipping is small, and
643		 * the error due to any saturation should not markedly affect
644		 * the downstream processing.
645		 */
646		tmp -= (tmp >> 4);
647
648		ec->tx_1 += -(ec->tx_1 >> DC_LOG2BETA) + tmp - ec->tx_2;
649		tmp1 = ec->tx_1 >> 15;
650		if (tmp1 > 32767)
651			tmp1 = 32767;
652		if (tmp1 < -32767)
653			tmp1 = -32767;
654		tx = tmp1;
655		ec->tx_2 = tmp;
656	}
657
658	return tx;
659}
660
661EXPORT_SYMBOL_GPL(oslec_hpf_tx);
662
663MODULE_LICENSE("GPL");
664MODULE_AUTHOR("David Rowe");
665MODULE_DESCRIPTION("Open Source Line Echo Canceller");
666MODULE_VERSION("0.3.0");