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 /* 32 Line echo cancellation for voice 33 34 What does it do? 35 36 This module aims to provide G.168-2002 compliant echo cancellation, to remove 37 electrical echoes (e.g. from 2-4 wire hybrids) from voice calls. 38 39 How does it work? 40 41 The heart of the echo cancellor is FIR filter. This is adapted to match the 42 echo impulse response of the telephone line. It must be long enough to 43 adequately cover the duration of that impulse response. The signal transmitted 44 to the telephone line is passed through the FIR filter. Once the FIR is 45 properly adapted, the resulting output is an estimate of the echo signal 46 received from the line. This is subtracted from the received signal. The result 47 is an estimate of the signal which originated at the far end of the line, free 48 from echos of our own transmitted signal. 49 50 The least mean squares (LMS) algorithm is attributed to Widrow and Hoff, and 51 was introduced in 1960. It is the commonest form of filter adaption used in 52 things like modem line equalisers and line echo cancellers. There it works very 53 well. However, it only works well for signals of constant amplitude. It works 54 very poorly for things like speech echo cancellation, where the signal level 55 varies widely. This is quite easy to fix. If the signal level is normalised - 56 similar to applying AGC - LMS can work as well for a signal of varying 57 amplitude as it does for a modem signal. This normalised least mean squares 58 (NLMS) algorithm is the commonest one used for speech echo cancellation. Many 59 other algorithms exist - e.g. RLS (essentially the same as Kalman filtering), 60 FAP, etc. Some perform significantly better than NLMS. However, factors such 61 as computational complexity and patents favour the use of NLMS. 62 63 A simple refinement to NLMS can improve its performance with speech. NLMS tends 64 to adapt best to the strongest parts of a signal. If the signal is white noise, 65 the NLMS algorithm works very well. However, speech has more low frequency than 66 high frequency content. Pre-whitening (i.e. filtering the signal to flatten its 67 spectrum) the echo signal improves the adapt rate for speech, and ensures the 68 final residual signal is not heavily biased towards high frequencies. A very 69 low complexity filter is adequate for this, so pre-whitening adds little to the 70 compute requirements of the echo canceller. 71 72 An FIR filter adapted using pre-whitened NLMS performs well, provided certain 73 conditions are met: 74 75 - The transmitted signal has poor self-correlation. 76 - There is no signal being generated within the environment being 77 cancelled. 78 79 The difficulty is that neither of these can be guaranteed. 80 81 If the adaption is performed while transmitting noise (or something fairly 82 noise like, such as voice) the adaption works very well. If the adaption is 83 performed while transmitting something highly correlative (typically narrow 84 band energy such as signalling tones or DTMF), the adaption can go seriously 85 wrong. The reason is there is only one solution for the adaption on a near 86 random signal - the impulse response of the line. For a repetitive signal, 87 there are any number of solutions which converge the adaption, and nothing 88 guides the adaption to choose the generalised one. Allowing an untrained 89 canceller to converge on this kind of narrowband energy probably a good thing, 90 since at least it cancels the tones. Allowing a well converged canceller to 91 continue converging on such energy is just a way to ruin its generalised 92 adaption. A narrowband detector is needed, so adapation can be suspended at 93 appropriate times. 94 95 The adaption process is based on trying to eliminate the received signal. When 96 there is any signal from within the environment being cancelled it may upset 97 the adaption process. Similarly, if the signal we are transmitting is small, 98 noise may dominate and disturb the adaption process. If we can ensure that the 99 adaption is only performed when we are transmitting a significant signal level, 100 and the environment is not, things will be OK. Clearly, it is easy to tell when 101 we are sending a significant signal. Telling, if the environment is generating 102 a significant signal, and doing it with sufficient speed that the adaption will 103 not have diverged too much more we stop it, is a little harder. 104 105 The key problem in detecting when the environment is sourcing significant 106 energy is that we must do this very quickly. Given a reasonably long sample of 107 the received signal, there are a number of strategies which may be used to 108 assess whether that signal contains a strong far end component. However, by the 109 time that assessment is complete the far end signal will have already caused 110 major mis-convergence in the adaption process. An assessment algorithm is 111 needed which produces a fairly accurate result from a very short burst of far 112 end energy. 113 114 How do I use it? 115 116 The echo cancellor processes both the transmit and receive streams sample by 117 sample. The processing function is not declared inline. Unfortunately, 118 cancellation requires many operations per sample, so the call overhead is only 119 a minor burden. 120 */ 121 122 #include "fir.h" 123 #include "oslec.h" 124 125 /* 126 G.168 echo canceller descriptor. This defines the working state for a line 127 echo canceller. 128 */ 129 struct oslec_state { 130 int16_t tx; 131 int16_t rx; 132 int16_t clean; 133 int16_t clean_nlp; 134 135 int nonupdate_dwell; 136 int curr_pos; 137 int taps; 138 int log2taps; 139 int adaption_mode; 140 141 int cond_met; 142 int32_t pstates; 143 int16_t adapt; 144 int32_t factor; 145 int16_t shift; 146 147 /* Average levels and averaging filter states */ 148 int ltxacc; 149 int lrxacc; 150 int lcleanacc; 151 int lclean_bgacc; 152 int ltx; 153 int lrx; 154 int lclean; 155 int lclean_bg; 156 int lbgn; 157 int lbgn_acc; 158 int lbgn_upper; 159 int lbgn_upper_acc; 160 161 /* foreground and background filter states */ 162 struct fir16_state_t fir_state; 163 struct fir16_state_t fir_state_bg; 164 int16_t *fir_taps16[2]; 165 166 /* DC blocking filter states */ 167 int tx_1; 168 int tx_2; 169 int rx_1; 170 int rx_2; 171 172 /* optional High Pass Filter states */ 173 int32_t xvtx[5]; 174 int32_t yvtx[5]; 175 int32_t xvrx[5]; 176 int32_t yvrx[5]; 177 178 /* Parameters for the optional Hoth noise generator */ 179 int cng_level; 180 int cng_rndnum; 181 int cng_filter; 182 183 /* snapshot sample of coeffs used for development */ 184 int16_t *snapshot; 185 }; 186 187 #endif /* __ECHO_H */ 188