1 /* SPDX-License-Identifier: GPL-2.0-only */ 2 /* 3 * SpanDSP - a series of DSP components for telephony 4 * 5 * echo.c - A line echo canceller. This code is being developed 6 * against and partially complies with G168. 7 * 8 * Written by Steve Underwood <steveu@coppice.org> 9 * and David Rowe <david_at_rowetel_dot_com> 10 * 11 * Copyright (C) 2001 Steve Underwood and 2007 David Rowe 12 * 13 * All rights reserved. 14 */ 15 16 #ifndef __ECHO_H 17 #define __ECHO_H 18 19 /* 20 Line echo cancellation for voice 21 22 What does it do? 23 24 This module aims to provide G.168-2002 compliant echo cancellation, to remove 25 electrical echoes (e.g. from 2-4 wire hybrids) from voice calls. 26 27 How does it work? 28 29 The heart of the echo cancellor is FIR filter. This is adapted to match the 30 echo impulse response of the telephone line. It must be long enough to 31 adequately cover the duration of that impulse response. The signal transmitted 32 to the telephone line is passed through the FIR filter. Once the FIR is 33 properly adapted, the resulting output is an estimate of the echo signal 34 received from the line. This is subtracted from the received signal. The result 35 is an estimate of the signal which originated at the far end of the line, free 36 from echos of our own transmitted signal. 37 38 The least mean squares (LMS) algorithm is attributed to Widrow and Hoff, and 39 was introduced in 1960. It is the commonest form of filter adaption used in 40 things like modem line equalisers and line echo cancellers. There it works very 41 well. However, it only works well for signals of constant amplitude. It works 42 very poorly for things like speech echo cancellation, where the signal level 43 varies widely. This is quite easy to fix. If the signal level is normalised - 44 similar to applying AGC - LMS can work as well for a signal of varying 45 amplitude as it does for a modem signal. This normalised least mean squares 46 (NLMS) algorithm is the commonest one used for speech echo cancellation. Many 47 other algorithms exist - e.g. RLS (essentially the same as Kalman filtering), 48 FAP, etc. Some perform significantly better than NLMS. However, factors such 49 as computational complexity and patents favour the use of NLMS. 50 51 A simple refinement to NLMS can improve its performance with speech. NLMS tends 52 to adapt best to the strongest parts of a signal. If the signal is white noise, 53 the NLMS algorithm works very well. However, speech has more low frequency than 54 high frequency content. Pre-whitening (i.e. filtering the signal to flatten its 55 spectrum) the echo signal improves the adapt rate for speech, and ensures the 56 final residual signal is not heavily biased towards high frequencies. A very 57 low complexity filter is adequate for this, so pre-whitening adds little to the 58 compute requirements of the echo canceller. 59 60 An FIR filter adapted using pre-whitened NLMS performs well, provided certain 61 conditions are met: 62 63 - The transmitted signal has poor self-correlation. 64 - There is no signal being generated within the environment being 65 cancelled. 66 67 The difficulty is that neither of these can be guaranteed. 68 69 If the adaption is performed while transmitting noise (or something fairly 70 noise like, such as voice) the adaption works very well. If the adaption is 71 performed while transmitting something highly correlative (typically narrow 72 band energy such as signalling tones or DTMF), the adaption can go seriously 73 wrong. The reason is there is only one solution for the adaption on a near 74 random signal - the impulse response of the line. For a repetitive signal, 75 there are any number of solutions which converge the adaption, and nothing 76 guides the adaption to choose the generalised one. Allowing an untrained 77 canceller to converge on this kind of narrowband energy probably a good thing, 78 since at least it cancels the tones. Allowing a well converged canceller to 79 continue converging on such energy is just a way to ruin its generalised 80 adaption. A narrowband detector is needed, so adapation can be suspended at 81 appropriate times. 82 83 The adaption process is based on trying to eliminate the received signal. When 84 there is any signal from within the environment being cancelled it may upset 85 the adaption process. Similarly, if the signal we are transmitting is small, 86 noise may dominate and disturb the adaption process. If we can ensure that the 87 adaption is only performed when we are transmitting a significant signal level, 88 and the environment is not, things will be OK. Clearly, it is easy to tell when 89 we are sending a significant signal. Telling, if the environment is generating 90 a significant signal, and doing it with sufficient speed that the adaption will 91 not have diverged too much more we stop it, is a little harder. 92 93 The key problem in detecting when the environment is sourcing significant 94 energy is that we must do this very quickly. Given a reasonably long sample of 95 the received signal, there are a number of strategies which may be used to 96 assess whether that signal contains a strong far end component. However, by the 97 time that assessment is complete the far end signal will have already caused 98 major mis-convergence in the adaption process. An assessment algorithm is 99 needed which produces a fairly accurate result from a very short burst of far 100 end energy. 101 102 How do I use it? 103 104 The echo cancellor processes both the transmit and receive streams sample by 105 sample. The processing function is not declared inline. Unfortunately, 106 cancellation requires many operations per sample, so the call overhead is only 107 a minor burden. 108 */ 109 110 #include "fir.h" 111 #include "oslec.h" 112 113 /* 114 G.168 echo canceller descriptor. This defines the working state for a line 115 echo canceller. 116 */ 117 struct oslec_state { 118 int16_t tx; 119 int16_t rx; 120 int16_t clean; 121 int16_t clean_nlp; 122 123 int nonupdate_dwell; 124 int curr_pos; 125 int taps; 126 int log2taps; 127 int adaption_mode; 128 129 int cond_met; 130 int32_t pstates; 131 int16_t adapt; 132 int32_t factor; 133 int16_t shift; 134 135 /* Average levels and averaging filter states */ 136 int ltxacc; 137 int lrxacc; 138 int lcleanacc; 139 int lclean_bgacc; 140 int ltx; 141 int lrx; 142 int lclean; 143 int lclean_bg; 144 int lbgn; 145 int lbgn_acc; 146 int lbgn_upper; 147 int 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; 156 int tx_2; 157 int rx_1; 158 int rx_2; 159 160 /* optional High Pass Filter states */ 161 int32_t xvtx[5]; 162 int32_t yvtx[5]; 163 int32_t xvrx[5]; 164 int32_t yvrx[5]; 165 166 /* Parameters for the optional Hoth noise generator */ 167 int cng_level; 168 int cng_rndnum; 169 int cng_filter; 170 171 /* snapshot sample of coeffs used for development */ 172 int16_t *snapshot; 173 }; 174 175 #endif /* __ECHO_H */ 176