Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
tjh.dev
kernel
os
linux
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 Steve Underwood and 2007 David Rowe
11 *
12 * All rights reserved.
13 *
14 * This program is free software; you can redistribute it and/or modify
15 * it under the terms of the GNU General Public License version 2, as
16 * published by the Free Software Foundation.
17 *
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 * GNU General Public License for more details.
22 *
23 * You should have received a copy of the GNU General Public License
24 * along with this program; if not, write to the Free Software
25 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
26 */
27
28#ifndef __ECHO_H
29#define __ECHO_H
30
31/*! \page echo_can_page Line echo cancellation for voice
32
33\section echo_can_page_sec_1 What does it do?
34This module aims to provide G.168-2002 compliant echo cancellation, to remove
35electrical echoes (e.g. from 2-4 wire hybrids) from voice calls.
36
37\section echo_can_page_sec_2 How does it work?
38The heart of the echo cancellor is FIR filter. This is adapted to match the
39echo impulse response of the telephone line. It must be long enough to
40adequately cover the duration of that impulse response. The signal transmitted
41to the telephone line is passed through the FIR filter. Once the FIR is
42properly adapted, the resulting output is an estimate of the echo signal
43received from the line. This is subtracted from the received signal. The result
44is an estimate of the signal which originated at the far end of the line, free
45from echos of our own transmitted signal.
46
47The least mean squares (LMS) algorithm is attributed to Widrow and Hoff, and
48was introduced in 1960. It is the commonest form of filter adaption used in
49things like modem line equalisers and line echo cancellers. There it works very
50well. However, it only works well for signals of constant amplitude. It works
51very poorly for things like speech echo cancellation, where the signal level
52varies widely. This is quite easy to fix. If the signal level is normalised -
53similar to applying AGC - LMS can work as well for a signal of varying
54amplitude as it does for a modem signal. This normalised least mean squares
55(NLMS) algorithm is the commonest one used for speech echo cancellation. Many
56other algorithms exist - e.g. RLS (essentially the same as Kalman filtering),
57FAP, etc. Some perform significantly better than NLMS. However, factors such
58as computational complexity and patents favour the use of NLMS.
59
60A simple refinement to NLMS can improve its performance with speech. NLMS tends
61to adapt best to the strongest parts of a signal. If the signal is white noise,
62the NLMS algorithm works very well. However, speech has more low frequency than
63high frequency content. Pre-whitening (i.e. filtering the signal to flatten its
64spectrum) the echo signal improves the adapt rate for speech, and ensures the
65final residual signal is not heavily biased towards high frequencies. A very
66low complexity filter is adequate for this, so pre-whitening adds little to the
67compute requirements of the echo canceller.
68
69An FIR filter adapted using pre-whitened NLMS performs well, provided certain
70conditions are met:
71
72 - The transmitted signal has poor self-correlation.
73 - There is no signal being generated within the environment being
74 cancelled.
75
76The difficulty is that neither of these can be guaranteed.
77
78If the adaption is performed while transmitting noise (or something fairly
79noise like, such as voice) the adaption works very well. If the adaption is
80performed while transmitting something highly correlative (typically narrow
81band energy such as signalling tones or DTMF), the adaption can go seriously
82wrong. The reason is there is only one solution for the adaption on a near
83random signal - the impulse response of the line. For a repetitive signal,
84there are any number of solutions which converge the adaption, and nothing
85guides the adaption to choose the generalised one. Allowing an untrained
86canceller to converge on this kind of narrowband energy probably a good thing,
87since at least it cancels the tones. Allowing a well converged canceller to
88continue converging on such energy is just a way to ruin its generalised
89adaption. A narrowband detector is needed, so adapation can be suspended at
90appropriate times.
91
92The adaption process is based on trying to eliminate the received signal. When
93there is any signal from within the environment being cancelled it may upset
94the adaption process. Similarly, if the signal we are transmitting is small,
95noise may dominate and disturb the adaption process. If we can ensure that the
96adaption is only performed when we are transmitting a significant signal level,
97and the environment is not, things will be OK. Clearly, it is easy to tell when
98we are sending a significant signal. Telling, if the environment is generating
99a significant signal, and doing it with sufficient speed that the adaption will
100not have diverged too much more we stop it, is a little harder.
101
102The key problem in detecting when the environment is sourcing significant
103energy is that we must do this very quickly. Given a reasonably long sample of
104the received signal, there are a number of strategies which may be used to
105assess whether that signal contains a strong far end component. However, by the
106time that assessment is complete the far end signal will have already caused
107major mis-convergence in the adaption process. An assessment algorithm is
108needed which produces a fairly accurate result from a very short burst of far
109end energy.
110
111\section echo_can_page_sec_3 How do I use it?
112The echo cancellor processes both the transmit and receive streams sample by
113sample. The processing function is not declared inline. Unfortunately,
114cancellation requires many operations per sample, so the call overhead is only
115a minor burden.
116*/
117
118#include "fir.h"
119#include "oslec.h"
120
121/*!
122 G.168 echo canceller descriptor. This defines the working state for a line
123 echo canceller.
124*/
125struct oslec_state {
126 int16_t tx, rx;
127 int16_t clean;
128 int16_t clean_nlp;
129
130 int nonupdate_dwell;
131 int curr_pos;
132 int taps;
133 int log2taps;
134 int adaption_mode;
135
136 int cond_met;
137 int32_t Pstates;
138 int16_t adapt;
139 int32_t factor;
140 int16_t shift;
141
142 /* Average levels and averaging filter states */
143 int Ltxacc, Lrxacc, Lcleanacc, Lclean_bgacc;
144 int Ltx, Lrx;
145 int Lclean;
146 int Lclean_bg;
147 int Lbgn, Lbgn_acc, Lbgn_upper, Lbgn_upper_acc;
148
149 /* foreground and background filter states */
150 struct fir16_state_t fir_state;
151 struct fir16_state_t fir_state_bg;
152 int16_t *fir_taps16[2];
153
154 /* DC blocking filter states */
155 int tx_1, tx_2, rx_1, rx_2;
156
157 /* optional High Pass Filter states */
158 int32_t xvtx[5], yvtx[5];
159 int32_t xvrx[5], yvrx[5];
160
161 /* Parameters for the optional Hoth noise generator */
162 int cng_level;
163 int cng_rndnum;
164 int cng_filter;
165
166 /* snapshot sample of coeffs used for development */
167 int16_t *snapshot;
168};
169
170#endif /* __ECHO_H */