00001 /* 00002 * SpanDSP - a series of DSP components for telephony 00003 * 00004 * echo.h - An echo cancellor, suitable for electrical and acoustic 00005 * cancellation. This code does not currently comply with 00006 * any relevant standards (e.g. G.164/5/7/8). 00007 * 00008 * Written by Steve Underwood <steveu@coppice.org> 00009 * 00010 * Copyright (C) 2001 Steve Underwood 00011 * 00012 * Based on a bit from here, a bit from there, eye of toad, 00013 * ear of bat, etc - plus, of course, my own 2 cents. 00014 * 00015 * All rights reserved. 00016 * 00017 * This program is free software; you can redistribute it and/or modify 00018 * it under the terms of the GNU General Public License as published by 00019 * the Free Software Foundation; either version 2 of the License, or 00020 * (at your option) any later version. 00021 * 00022 * This program is distributed in the hope that it will be useful, 00023 * but WITHOUT ANY WARRANTY; without even the implied warranty of 00024 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 00025 * GNU General Public License for more details. 00026 * 00027 * You should have received a copy of the GNU General Public License 00028 * along with this program; if not, write to the Free Software 00029 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. 00030 * 00031 * $Id: echo.h,v 1.7 2005/12/01 12:43:37 steveu Exp $ 00032 */ 00033 00034 /*! \file */ 00035 00036 #if !defined(_ECHO_H_) 00037 #define _ECHO_H_ 00038 00039 /*! \page echo_can_page Line echo cancellation for voice 00040 00041 \section echo_can_page_sec_1 What does it do? 00042 This module aims to provide G.168-2002 compliant echo cancellation, to remove 00043 electrical echoes (e.g. from 2-4 wire hybrids) from voice calls. 00044 00045 \section echo_can_page_sec_2 How does it work? 00046 The heart of the echo cancellor is FIR filter. This is adapted to match the echo 00047 impulse response of the telephone line. It must be long enough to adequately cover 00048 the duration of that impulse response. The signal transmitted to the telephone line 00049 is passed through the FIR filter. Once the FIR is properly adapted, the resulting 00050 output is an estimate of the echo signal received from the line. This is subtracted 00051 from the received signal. The result is an estimate of the signal which originated 00052 at the far end of the line, free from echos of our own transmitted signal. 00053 00054 The least mean squares (LMS) algorithm is attributed to Widrow and Hoff, and was 00055 introduced in 1960. It is the commonest form of filter adaption used in things 00056 like modem line equalisers and line echo cancellers. There it works very well. 00057 However, it only works well for signals of constant amplitude. It works very poorly 00058 for things like speech echo cancellation, where the signal level varies widely. 00059 This is quite easy to fix. If the signal level is normalised - similar to applying 00060 AGC - LMS can work as well for a signal of varying amplitude as it does for a modem 00061 signal. This normalised least mean squares (NLMS) algorithm is the commonest one used 00062 for speech echo cancellation. Many other algorithms exist - e.g. RLS (essentially 00063 the same as Kalman filtering), FAP, etc. Some perform significantly better than NLMS. 00064 However, factors such as computational complexity and patents favour the use of NLMS. 00065 00066 A simple refinement to NLMS can improve its performance with speech. NLMS tends 00067 to adapt best to the strongest parts of a signal. If the signal is white noise, 00068 the NLMS algorithm works very well. However, speech has more low frequency than 00069 high frequency content. Pre-whitening (i.e. filtering the signal to flatten 00070 its spectrum) the echo signal improves the adapt rate for speech, and ensures the 00071 final residual signal is not heavily biased towards high frequencies. A very low 00072 complexity filter is adequate for this, so pre-whitening adds little to the 00073 compute requirements of the echo canceller. 00074 00075 An FIR filter adapted using pre-whitened NLMS performs well, provided certain 00076 conditions are met: 00077 00078 - The transmitted signal has poor self-correlation. 00079 - There is no signal being generated within the environment being cancelled. 00080 00081 The difficulty is that neither of these can be guaranteed. 00082 00083 If the adaption is performed while transmitting noise (or something fairly noise 00084 like, such as voice) the adaption works very well. If the adaption is performed 00085 while transmitting something highly correlative (typically narrow band energy 00086 such as signalling tones or DTMF), the adaption can go seriously wrong. The reason 00087 is there is only one solution for the adaption on a near random signal - the impulse 00088 response of the line. For a repetitive signal, there are any number of solutions 00089 which converge the adaption, and nothing guides the adaption to choose the generalised 00090 one. Allowing an untrained canceller to converge on this kind of narrowband 00091 energy probably a good thing, since at least it cancels the tones. Allowing a well 00092 converged canceller to continue converging on such energy is just a way to ruin 00093 its generalised adaption. A narrowband detector is needed, so adapation can be 00094 suspended at appropriate times. 00095 00096 The adaption process is based on trying to eliminate the received signal. When 00097 there is any signal from within the environment being cancelled it may upset the 00098 adaption process. Similarly, if the signal we are transmitting is small, noise 00099 may dominate and disturb the adaption process. If we can ensure that the 00100 adaption is only performed when we are transmitting a significant signal level, 00101 and the environment is not, things will be OK. Clearly, it is easy to tell when 00102 we are sending a significant signal. Telling, if the environment is generating a 00103 significant signal, and doing it with sufficient speed that the adaption will 00104 not have diverged too much more we stop it, is a little harder. 00105 00106 The key problem in detecting when the environment is sourcing significant energy 00107 is that we must do this very quickly. Given a reasonably long sample of the 00108 received signal, there are a number of strategies which may be used to assess 00109 whether that signal contains a strong far end component. However, by the time 00110 that assessment is complete the far end signal will have already caused major 00111 mis-convergence in the adaption process. An assessment algorithm is needed which 00112 produces a fairly accurate result from a very short burst of far end energy. 00113 00114 \section echo_can_page_sec_3 How do I use it? 00115 The echo cancellor processes both the transmit and receive streams sample by 00116 sample. The processing function is not declared inline. Unfortunately, 00117 cancellation requires many operations per sample, so the call overhead is only a 00118 minor burden. 00119 */ 00120 00121 #include "fir.h" 00122 00123 #define NONUPDATE_DWELL_TIME 600 /* 600 samples, or 75ms */ 00124 00125 /* Mask bits for the adaption mode */ 00126 #define ECHO_CAN_USE_NLP 0x01 00127 #define ECHO_CAN_USE_SUPPRESSOR 0x02 00128 #define ECHO_CAN_USE_CNG 0x04 00129 #define ECHO_CAN_USE_ADAPTION 0x08 00130 00131 /*! 00132 G.168 echo canceller descriptor. This defines the working state for a line 00133 echo canceller. 00134 */ 00135 typedef struct 00136 { 00137 int tx_power[4]; 00138 int rx_power[3]; 00139 int clean_rx_power; 00140 00141 int rx_power_threshold; 00142 int nonupdate_dwell; 00143 00144 fir16_state_t fir_state; 00145 /*! Echo FIR taps (16 bit version) */ 00146 int16_t *fir_taps16[4]; 00147 /*! Echo FIR taps (32 bit version) */ 00148 int32_t *fir_taps32; 00149 00150 int curr_pos; 00151 00152 int taps; 00153 int tap_mask; 00154 int adaption_mode; 00155 00156 int32_t supp_test1; 00157 int32_t supp_test2; 00158 int32_t supp1; 00159 int32_t supp2; 00160 int vad; 00161 int cng; 00162 /* Parameters for the Hoth noise generator */ 00163 int cng_level; 00164 int cng_rndnum; 00165 int cng_filter; 00166 00167 int16_t geigel_max; 00168 int geigel_lag; 00169 int dtd_onset; 00170 int tap_set; 00171 int tap_rotate_counter; 00172 00173 int32_t latest_correction; /* Indication of the magnitude of the latest 00174 adaption, or a code to indicate why adaption 00175 was skipped, for test purposes */ 00176 int32_t last_acf[28]; 00177 int narrowband_count; 00178 int narrowband_score; 00179 } echo_can_state_t; 00180 00181 /*! Create a voice echo canceller context. 00182 \param len The length of the canceller, in samples. 00183 \return The new canceller context, or NULL if the canceller could not be created. 00184 */ 00185 echo_can_state_t *echo_can_create(int len, int adaption_mode); 00186 00187 /*! Free a voice echo canceller context. 00188 \param ec The echo canceller context. 00189 */ 00190 void echo_can_free(echo_can_state_t *ec); 00191 00192 /*! Flush (reinitialise) a voice echo canceller context. 00193 \param ec The echo canceller context. 00194 */ 00195 void echo_can_flush(echo_can_state_t *ec); 00196 00197 /*! Set the adaption mode of a voice echo canceller context. 00198 \param ec The echo canceller context. 00199 \param adapt The mode. 00200 */ 00201 void echo_can_adaption_mode(echo_can_state_t *ec, int adaption_mode); 00202 00203 /*! Process a sample through a voice echo canceller. 00204 \param ec The echo canceller context. 00205 \param tx The transmitted audio sample. 00206 \param rx The received audio sample. 00207 \return The clean (echo cancelled) received sample. 00208 */ 00209 int16_t echo_can_update(echo_can_state_t *ec, int16_t tx, int16_t rx); 00210 00211 #endif 00212 /*- End of file ------------------------------------------------------------*/