diff options
Diffstat (limited to 'src/modules/audio_processing/agc/main')
8 files changed, 3086 insertions, 0 deletions
diff --git a/src/modules/audio_processing/agc/main/interface/gain_control.h b/src/modules/audio_processing/agc/main/interface/gain_control.h new file mode 100644 index 0000000000..2893331faf --- /dev/null +++ b/src/modules/audio_processing/agc/main/interface/gain_control.h @@ -0,0 +1,273 @@ +/* + * Copyright (c) 2011 The WebRTC project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AGC_MAIN_INTERFACE_GAIN_CONTROL_H_ +#define WEBRTC_MODULES_AUDIO_PROCESSING_AGC_MAIN_INTERFACE_GAIN_CONTROL_H_ + +#include "typedefs.h" + +// Errors +#define AGC_UNSPECIFIED_ERROR 18000 +#define AGC_UNSUPPORTED_FUNCTION_ERROR 18001 +#define AGC_UNINITIALIZED_ERROR 18002 +#define AGC_NULL_POINTER_ERROR 18003 +#define AGC_BAD_PARAMETER_ERROR 18004 + +// Warnings +#define AGC_BAD_PARAMETER_WARNING 18050 + +enum +{ + kAgcModeUnchanged, + kAgcModeAdaptiveAnalog, + kAgcModeAdaptiveDigital, + kAgcModeFixedDigital +}; + +enum +{ + kAgcFalse = 0, + kAgcTrue +}; + +typedef struct +{ + WebRtc_Word16 targetLevelDbfs; // default 3 (-3 dBOv) + WebRtc_Word16 compressionGaindB; // default 9 dB + WebRtc_UWord8 limiterEnable; // default kAgcTrue (on) +} WebRtcAgc_config_t; + +#if defined(__cplusplus) +extern "C" +{ +#endif + +/* + * This function processes a 10/20ms frame of far-end speech to determine + * if there is active speech. Far-end speech length can be either 10ms or + * 20ms. The length of the input speech vector must be given in samples + * (80/160 when FS=8000, and 160/320 when FS=16000 or FS=32000). + * + * Input: + * - agcInst : AGC instance. + * - inFar : Far-end input speech vector (10 or 20ms) + * - samples : Number of samples in input vector + * + * Return value: + * : 0 - Normal operation. + * : -1 - Error + */ +int WebRtcAgc_AddFarend(void* agcInst, + const WebRtc_Word16* inFar, + WebRtc_Word16 samples); + +/* + * This function processes a 10/20ms frame of microphone speech to determine + * if there is active speech. Microphone speech length can be either 10ms or + * 20ms. The length of the input speech vector must be given in samples + * (80/160 when FS=8000, and 160/320 when FS=16000 or FS=32000). For very low + * input levels, the input signal is increased in level by multiplying and + * overwriting the samples in inMic[]. + * + * This function should be called before any further processing of the + * near-end microphone signal. + * + * Input: + * - agcInst : AGC instance. + * - inMic : Microphone input speech vector (10 or 20 ms) for + * L band + * - inMic_H : Microphone input speech vector (10 or 20 ms) for + * H band + * - samples : Number of samples in input vector + * + * Return value: + * : 0 - Normal operation. + * : -1 - Error + */ +int WebRtcAgc_AddMic(void* agcInst, + WebRtc_Word16* inMic, + WebRtc_Word16* inMic_H, + WebRtc_Word16 samples); + +/* + * This function replaces the analog microphone with a virtual one. + * It is a digital gain applied to the input signal and is used in the + * agcAdaptiveDigital mode where no microphone level is adjustable. + * Microphone speech length can be either 10ms or 20ms. The length of the + * input speech vector must be given in samples (80/160 when FS=8000, and + * 160/320 when FS=16000 or FS=32000). + * + * Input: + * - agcInst : AGC instance. + * - inMic : Microphone input speech vector for (10 or 20 ms) + * L band + * - inMic_H : Microphone input speech vector for (10 or 20 ms) + * H band + * - samples : Number of samples in input vector + * - micLevelIn : Input level of microphone (static) + * + * Output: + * - inMic : Microphone output after processing (L band) + * - inMic_H : Microphone output after processing (H band) + * - micLevelOut : Adjusted microphone level after processing + * + * Return value: + * : 0 - Normal operation. + * : -1 - Error + */ +int WebRtcAgc_VirtualMic(void* agcInst, + WebRtc_Word16* inMic, + WebRtc_Word16* inMic_H, + WebRtc_Word16 samples, + WebRtc_Word32 micLevelIn, + WebRtc_Word32* micLevelOut); + +/* + * This function processes a 10/20ms frame and adjusts (normalizes) the gain + * both analog and digitally. The gain adjustments are done only during + * active periods of speech. The input speech length can be either 10ms or + * 20ms and the output is of the same length. The length of the speech + * vectors must be given in samples (80/160 when FS=8000, and 160/320 when + * FS=16000 or FS=32000). The echo parameter can be used to ensure the AGC will + * not adjust upward in the presence of echo. + * + * This function should be called after processing the near-end microphone + * signal, in any case after any echo cancellation. + * + * Input: + * - agcInst : AGC instance + * - inNear : Near-end input speech vector (10 or 20 ms) for + * L band + * - inNear_H : Near-end input speech vector (10 or 20 ms) for + * H band + * - samples : Number of samples in input/output vector + * - inMicLevel : Current microphone volume level + * - echo : Set to 0 if the signal passed to add_mic is + * almost certainly free of echo; otherwise set + * to 1. If you have no information regarding echo + * set to 0. + * + * Output: + * - outMicLevel : Adjusted microphone volume level + * - out : Gain-adjusted near-end speech vector (L band) + * : May be the same vector as the input. + * - out_H : Gain-adjusted near-end speech vector (H band) + * - saturationWarning : A returned value of 1 indicates a saturation event + * has occurred and the volume cannot be further + * reduced. Otherwise will be set to 0. + * + * Return value: + * : 0 - Normal operation. + * : -1 - Error + */ +int WebRtcAgc_Process(void* agcInst, + const WebRtc_Word16* inNear, + const WebRtc_Word16* inNear_H, + WebRtc_Word16 samples, + WebRtc_Word16* out, + WebRtc_Word16* out_H, + WebRtc_Word32 inMicLevel, + WebRtc_Word32* outMicLevel, + WebRtc_Word16 echo, + WebRtc_UWord8* saturationWarning); + +/* + * This function sets the config parameters (targetLevelDbfs, + * compressionGaindB and limiterEnable). + * + * Input: + * - agcInst : AGC instance + * - config : config struct + * + * Output: + * + * Return value: + * : 0 - Normal operation. + * : -1 - Error + */ +int WebRtcAgc_set_config(void* agcInst, WebRtcAgc_config_t config); + +/* + * This function returns the config parameters (targetLevelDbfs, + * compressionGaindB and limiterEnable). + * + * Input: + * - agcInst : AGC instance + * + * Output: + * - config : config struct + * + * Return value: + * : 0 - Normal operation. + * : -1 - Error + */ +int WebRtcAgc_get_config(void* agcInst, WebRtcAgc_config_t* config); + +/* + * This function creates an AGC instance, which will contain the state + * information for one (duplex) channel. + * + * Return value : AGC instance if successful + * : 0 (i.e., a NULL pointer) if unsuccessful + */ +int WebRtcAgc_Create(void **agcInst); + +/* + * This function frees the AGC instance created at the beginning. + * + * Input: + * - agcInst : AGC instance. + * + * Return value : 0 - Ok + * -1 - Error + */ +int WebRtcAgc_Free(void *agcInst); + +/* + * This function initializes an AGC instance. + * + * Input: + * - agcInst : AGC instance. + * - minLevel : Minimum possible mic level + * - maxLevel : Maximum possible mic level + * - agcMode : 0 - Unchanged + * : 1 - Adaptive Analog Automatic Gain Control -3dBOv + * : 2 - Adaptive Digital Automatic Gain Control -3dBOv + * : 3 - Fixed Digital Gain 0dB + * - fs : Sampling frequency + * + * Return value : 0 - Ok + * -1 - Error + */ +int WebRtcAgc_Init(void *agcInst, + WebRtc_Word32 minLevel, + WebRtc_Word32 maxLevel, + WebRtc_Word16 agcMode, + WebRtc_UWord32 fs); + +/* + * This function returns a text string containing the version. + * + * Input: + * - length : Length of the char array pointed to by version + * Output: + * - version : Pointer to a char array of to which the version + * : string will be copied. + * + * Return value : 0 - OK + * -1 - Error + */ +int WebRtcAgc_Version(WebRtc_Word8 *versionStr, WebRtc_Word16 length); + +#if defined(__cplusplus) +} +#endif + +#endif // WEBRTC_MODULES_AUDIO_PROCESSING_AGC_MAIN_INTERFACE_GAIN_CONTROL_H_ diff --git a/src/modules/audio_processing/agc/main/matlab/getGains.m b/src/modules/audio_processing/agc/main/matlab/getGains.m new file mode 100644 index 0000000000..e0234b8593 --- /dev/null +++ b/src/modules/audio_processing/agc/main/matlab/getGains.m @@ -0,0 +1,32 @@ +% Outputs a file for testing purposes. +% +% Adjust the following parameters to suit. Their purpose becomes more clear on +% viewing the gain plots. +% MaxGain: Max gain in dB +% MinGain: Min gain at overload (0 dBov) in dB +% CompRatio: Compression ratio, essentially determines the slope of the gain +% function between the max and min gains +% Knee: The smoothness of the transition to max gain (smaller is smoother) +MaxGain = 5; MinGain = 0; CompRatio = 3; Knee = 1; + +% Compute gains +zeros = 0:31; lvl = 2.^(1-zeros); +A = -10*log10(lvl) * (CompRatio - 1) / CompRatio; +B = MaxGain - MinGain; +gains = round(2^16*10.^(0.05 * (MinGain + B * ( log(exp(-Knee*A)+exp(-Knee*B)) - log(1+exp(-Knee*B)) ) / log(1/(1+exp(Knee*B)))))); +fprintf(1, '\t%i, %i, %i, %i,\n', gains); + +% Save gains to file +fid = fopen('gains', 'wb'); +if fid == -1 + error(sprintf('Unable to open file %s', filename)); + return +end +fwrite(fid, gains, 'int32'); +fclose(fid); + +% Plotting +in = 10*log10(lvl); out = 20*log10(gains/65536); +subplot(121); plot(in, out); axis([-60, 0, -5, 30]); grid on; xlabel('Input (dB)'); ylabel('Gain (dB)'); +subplot(122); plot(in, in+out); axis([-60, 0, -60, 10]); grid on; xlabel('Input (dB)'); ylabel('Output (dB)'); +zoom on; diff --git a/src/modules/audio_processing/agc/main/source/Android.mk b/src/modules/audio_processing/agc/main/source/Android.mk new file mode 100644 index 0000000000..e045839147 --- /dev/null +++ b/src/modules/audio_processing/agc/main/source/Android.mk @@ -0,0 +1,49 @@ +# This file is generated by gyp; do not edit. This means you! + +LOCAL_PATH := $(call my-dir) + +include $(CLEAR_VARS) + +LOCAL_ARM_MODE := arm +LOCAL_MODULE_CLASS := STATIC_LIBRARIES +LOCAL_MODULE := libwebrtc_agc +LOCAL_MODULE_TAGS := optional +LOCAL_GENERATED_SOURCES := +LOCAL_SRC_FILES := analog_agc.c \ + digital_agc.c + +# Flags passed to both C and C++ files. +MY_CFLAGS := +MY_CFLAGS_C := +MY_DEFS := '-DNO_TCMALLOC' \ + '-DNO_HEAPCHECKER' \ + '-DWEBRTC_TARGET_PC' \ + '-DWEBRTC_LINUX' \ + '-DWEBRTC_THREAD_RR' +ifeq ($(TARGET_ARCH),arm) +MY_DEFS += \ + '-DWEBRTC_ANDROID' \ + '-DANDROID' +endif +LOCAL_CFLAGS := $(MY_CFLAGS_C) $(MY_CFLAGS) $(MY_DEFS) + +# Include paths placed before CFLAGS/CPPFLAGS +LOCAL_C_INCLUDES := $(LOCAL_PATH)/../../../../.. \ + $(LOCAL_PATH)/../interface \ + $(LOCAL_PATH)/../../../../../common_audio/signal_processing_library/main/interface + +# Flags passed to only C++ (and not C) files. +LOCAL_CPPFLAGS := +LOCAL_LDFLAGS := + +LOCAL_STATIC_LIBRARIES := +# Duplicate the static libraries to fix circular references +LOCAL_STATIC_LIBRARIES += $(LOCAL_STATIC_LIBRARIES) + +LOCAL_SHARED_LIBRARIES := libcutils \ + libdl \ + libstlport +LOCAL_ADDITIONAL_DEPENDENCIES := + +include external/stlport/libstlport.mk +include $(BUILD_STATIC_LIBRARY) diff --git a/src/modules/audio_processing/agc/main/source/agc.gyp b/src/modules/audio_processing/agc/main/source/agc.gyp new file mode 100644 index 0000000000..e28a4c8c68 --- /dev/null +++ b/src/modules/audio_processing/agc/main/source/agc.gyp @@ -0,0 +1,43 @@ +# Copyright (c) 2011 The WebRTC project authors. All Rights Reserved. +# +# Use of this source code is governed by a BSD-style license +# that can be found in the LICENSE file in the root of the source +# tree. An additional intellectual property rights grant can be found +# in the file PATENTS. All contributing project authors may +# be found in the AUTHORS file in the root of the source tree. + +{ + 'includes': [ + '../../../../../common_settings.gypi', # Common settings + ], + 'targets': [ + { + 'target_name': 'agc', + 'type': '<(library)', + 'dependencies': [ + '../../../../../common_audio/signal_processing_library/main/source/spl.gyp:spl', + ], + 'include_dirs': [ + '../interface', + ], + 'direct_dependent_settings': { + 'include_dirs': [ + '../interface', + ], + }, + 'sources': [ + '../interface/gain_control.h', + 'analog_agc.c', + 'analog_agc.h', + 'digital_agc.c', + 'digital_agc.h', + ], + }, + ], +} + +# Local Variables: +# tab-width:2 +# indent-tabs-mode:nil +# End: +# vim: set expandtab tabstop=2 shiftwidth=2: diff --git a/src/modules/audio_processing/agc/main/source/analog_agc.c b/src/modules/audio_processing/agc/main/source/analog_agc.c new file mode 100644 index 0000000000..dfb7adc621 --- /dev/null +++ b/src/modules/audio_processing/agc/main/source/analog_agc.c @@ -0,0 +1,1700 @@ +/* + * Copyright (c) 2011 The WebRTC project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +/* analog_agc.c + * + * Using a feedback system, determines an appropriate analog volume level + * given an input signal and current volume level. Targets a conservative + * signal level and is intended for use with a digital AGC to apply + * additional gain. + * + */ + +#include <assert.h> +#include <stdlib.h> +#ifdef AGC_DEBUG //test log +#include <stdio.h> +#endif +#include "analog_agc.h" + +/* The slope of in Q13*/ +static const WebRtc_Word16 kSlope1[8] = {21793, 12517, 7189, 4129, 2372, 1362, 472, 78}; + +/* The offset in Q14 */ +static const WebRtc_Word16 kOffset1[8] = {25395, 23911, 22206, 20737, 19612, 18805, 17951, + 17367}; + +/* The slope of in Q13*/ +static const WebRtc_Word16 kSlope2[8] = {2063, 1731, 1452, 1218, 1021, 857, 597, 337}; + +/* The offset in Q14 */ +static const WebRtc_Word16 kOffset2[8] = {18432, 18379, 18290, 18177, 18052, 17920, 17670, + 17286}; + +static const WebRtc_Word16 kMuteGuardTimeMs = 8000; +static const WebRtc_Word16 kInitCheck = 42; + +/* Default settings if config is not used */ +#define AGC_DEFAULT_TARGET_LEVEL 3 +#define AGC_DEFAULT_COMP_GAIN 9 +/* This is the target level for the analog part in ENV scale. To convert to RMS scale you + * have to add OFFSET_ENV_TO_RMS. + */ +#define ANALOG_TARGET_LEVEL 11 +#define ANALOG_TARGET_LEVEL_2 5 // ANALOG_TARGET_LEVEL / 2 +/* Offset between RMS scale (analog part) and ENV scale (digital part). This value actually + * varies with the FIXED_ANALOG_TARGET_LEVEL, hence we should in the future replace it with + * a table. + */ +#define OFFSET_ENV_TO_RMS 9 +/* The reference input level at which the digital part gives an output of targetLevelDbfs + * (desired level) if we have no compression gain. This level should be set high enough not + * to compress the peaks due to the dynamics. + */ +#define DIGITAL_REF_AT_0_COMP_GAIN 4 +/* Speed of reference level decrease. + */ +#define DIFF_REF_TO_ANALOG 5 + +#ifdef MIC_LEVEL_FEEDBACK +#define NUM_BLOCKS_IN_SAT_BEFORE_CHANGE_TARGET 7 +#endif +/* Size of analog gain table */ +#define GAIN_TBL_LEN 32 +/* Matlab code: + * fprintf(1, '\t%i, %i, %i, %i,\n', round(10.^(linspace(0,10,32)/20) * 2^12)); + */ +/* Q12 */ +static const WebRtc_UWord16 kGainTableAnalog[GAIN_TBL_LEN] = {4096, 4251, 4412, 4579, 4752, + 4932, 5118, 5312, 5513, 5722, 5938, 6163, 6396, 6638, 6889, 7150, 7420, 7701, 7992, + 8295, 8609, 8934, 9273, 9623, 9987, 10365, 10758, 11165, 11587, 12025, 12480, 12953}; + +/* Gain/Suppression tables for virtual Mic (in Q10) */ +static const WebRtc_UWord16 kGainTableVirtualMic[128] = {1052, 1081, 1110, 1141, 1172, 1204, + 1237, 1271, 1305, 1341, 1378, 1416, 1454, 1494, 1535, 1577, 1620, 1664, 1710, 1757, + 1805, 1854, 1905, 1957, 2010, 2065, 2122, 2180, 2239, 2301, 2364, 2428, 2495, 2563, + 2633, 2705, 2779, 2855, 2933, 3013, 3096, 3180, 3267, 3357, 3449, 3543, 3640, 3739, + 3842, 3947, 4055, 4166, 4280, 4397, 4517, 4640, 4767, 4898, 5032, 5169, 5311, 5456, + 5605, 5758, 5916, 6078, 6244, 6415, 6590, 6770, 6956, 7146, 7341, 7542, 7748, 7960, + 8178, 8402, 8631, 8867, 9110, 9359, 9615, 9878, 10148, 10426, 10711, 11004, 11305, + 11614, 11932, 12258, 12593, 12938, 13292, 13655, 14029, 14412, 14807, 15212, 15628, + 16055, 16494, 16945, 17409, 17885, 18374, 18877, 19393, 19923, 20468, 21028, 21603, + 22194, 22801, 23425, 24065, 24724, 25400, 26095, 26808, 27541, 28295, 29069, 29864, + 30681, 31520, 32382}; +static const WebRtc_UWord16 kSuppressionTableVirtualMic[128] = {1024, 1006, 988, 970, 952, + 935, 918, 902, 886, 870, 854, 839, 824, 809, 794, 780, 766, 752, 739, 726, 713, 700, + 687, 675, 663, 651, 639, 628, 616, 605, 594, 584, 573, 563, 553, 543, 533, 524, 514, + 505, 496, 487, 478, 470, 461, 453, 445, 437, 429, 421, 414, 406, 399, 392, 385, 378, + 371, 364, 358, 351, 345, 339, 333, 327, 321, 315, 309, 304, 298, 293, 288, 283, 278, + 273, 268, 263, 258, 254, 249, 244, 240, 236, 232, 227, 223, 219, 215, 211, 208, 204, + 200, 197, 193, 190, 186, 183, 180, 176, 173, 170, 167, 164, 161, 158, 155, 153, 150, + 147, 145, 142, 139, 137, 134, 132, 130, 127, 125, 123, 121, 118, 116, 114, 112, 110, + 108, 106, 104, 102}; + +/* Table for target energy levels. Values in Q(-7) + * Matlab code + * targetLevelTable = fprintf('%d,\t%d,\t%d,\t%d,\n', round((32767*10.^(-(0:63)'/20)).^2*16/2^7) */ + +static const WebRtc_Word32 kTargetLevelTable[64] = {134209536, 106606424, 84680493, 67264106, + 53429779, 42440782, 33711911, 26778323, 21270778, 16895980, 13420954, 10660642, + 8468049, 6726411, 5342978, 4244078, 3371191, 2677832, 2127078, 1689598, 1342095, + 1066064, 846805, 672641, 534298, 424408, 337119, 267783, 212708, 168960, 134210, + 106606, 84680, 67264, 53430, 42441, 33712, 26778, 21271, 16896, 13421, 10661, 8468, + 6726, 5343, 4244, 3371, 2678, 2127, 1690, 1342, 1066, 847, 673, 534, 424, 337, 268, + 213, 169, 134, 107, 85, 67}; + +int WebRtcAgc_AddMic(void *state, WebRtc_Word16 *in_mic, WebRtc_Word16 *in_mic_H, + WebRtc_Word16 samples) +{ + WebRtc_Word32 nrg, max_nrg, sample, tmp32; + WebRtc_Word32 *ptr; + WebRtc_UWord16 targetGainIdx, gain; + WebRtc_Word16 i, n, L, M, subFrames, tmp16, tmp_speech[16]; + Agc_t *stt; + stt = (Agc_t *)state; + + //default/initial values corresponding to 10ms for wb and swb + M = 10; + L = 16; + subFrames = 160; + + if (stt->fs == 8000) + { + if (samples == 80) + { + subFrames = 80; + M = 10; + L = 8; + } else if (samples == 160) + { + subFrames = 80; + M = 20; + L = 8; + } else + { +#ifdef AGC_DEBUG //test log + fprintf(stt->fpt, + "AGC->add_mic, frame %d: Invalid number of samples\n\n", + (stt->fcount + 1)); +#endif + return -1; + } + } else if (stt->fs == 16000) + { + if (samples == 160) + { + subFrames = 160; + M = 10; + L = 16; + } else if (samples == 320) + { + subFrames = 160; + M = 20; + L = 16; + } else + { +#ifdef AGC_DEBUG //test log + fprintf(stt->fpt, + "AGC->add_mic, frame %d: Invalid number of samples\n\n", + (stt->fcount + 1)); +#endif + return -1; + } + } else if (stt->fs == 32000) + { + /* SWB is processed as 160 sample for L and H bands */ + if (samples == 160) + { + subFrames = 160; + M = 10; + L = 16; + } else + { +#ifdef AGC_DEBUG + fprintf(stt->fpt, + "AGC->add_mic, frame %d: Invalid sample rate\n\n", + (stt->fcount + 1)); +#endif + return -1; + } + } + + /* Check for valid pointers based on sampling rate */ + if ((stt->fs == 32000) && (in_mic_H == NULL)) + { + return -1; + } + /* Check for valid pointer for low band */ + if (in_mic == NULL) + { + return -1; + } + + /* apply slowly varying digital gain */ + if (stt->micVol > stt->maxAnalog) + { + /* Q1 */ + tmp16 = (WebRtc_Word16)(stt->micVol - stt->maxAnalog); + tmp32 = WEBRTC_SPL_MUL_16_16(GAIN_TBL_LEN - 1, tmp16); + tmp16 = (WebRtc_Word16)(stt->maxLevel - stt->maxAnalog); + targetGainIdx = (WebRtc_UWord16)WEBRTC_SPL_DIV(tmp32, tmp16); + assert(targetGainIdx < GAIN_TBL_LEN); + + /* Increment through the table towards the target gain. + * If micVol drops below maxAnalog, we allow the gain + * to be dropped immediately. */ + if (stt->gainTableIdx < targetGainIdx) + { + stt->gainTableIdx++; + } else if (stt->gainTableIdx > targetGainIdx) + { + stt->gainTableIdx--; + } + + /* Q12 */ + gain = kGainTableAnalog[stt->gainTableIdx]; + + for (i = 0; i < samples; i++) + { + // For lower band + tmp32 = WEBRTC_SPL_MUL_16_U16(in_mic[i], gain); + sample = WEBRTC_SPL_RSHIFT_W32(tmp32, 12); + if (sample > 32767) + { + in_mic[i] = 32767; + } else if (sample < -32768) + { + in_mic[i] = -32768; + } else + { + in_mic[i] = (WebRtc_Word16)sample; + } + + // For higher band + if (stt->fs == 32000) + { + tmp32 = WEBRTC_SPL_MUL_16_U16(in_mic_H[i], gain); + sample = WEBRTC_SPL_RSHIFT_W32(tmp32, 12); + if (sample > 32767) + { + in_mic_H[i] = 32767; + } else if (sample < -32768) + { + in_mic_H[i] = -32768; + } else + { + in_mic_H[i] = (WebRtc_Word16)sample; + } + } + } + } else + { + stt->gainTableIdx = 0; + } + + /* compute envelope */ + if ((M == 10) && (stt->inQueue > 0)) + { + ptr = stt->env[1]; + } else + { + ptr = stt->env[0]; + } + + for (i = 0; i < M; i++) + { + /* iterate over samples */ + max_nrg = 0; + for (n = 0; n < L; n++) + { + nrg = WEBRTC_SPL_MUL_16_16(in_mic[i * L + n], in_mic[i * L + n]); + if (nrg > max_nrg) + { + max_nrg = nrg; + } + } + ptr[i] = max_nrg; + } + + /* compute energy */ + if ((M == 10) && (stt->inQueue > 0)) + { + ptr = stt->Rxx16w32_array[1]; + } else + { + ptr = stt->Rxx16w32_array[0]; + } + + for (i = 0; i < WEBRTC_SPL_RSHIFT_W16(M, 1); i++) + { + if (stt->fs == 16000) + { + WebRtcSpl_DownsampleBy2(&in_mic[i * 32], 32, tmp_speech, stt->filterState); + } else + { + memcpy(tmp_speech, &in_mic[i * 16], 16 * sizeof(short)); + } + /* Compute energy in blocks of 16 samples */ + ptr[i] = WebRtcSpl_DotProductWithScale(tmp_speech, tmp_speech, 16, 4); + } + + /* update queue information */ + if ((stt->inQueue == 0) && (M == 10)) + { + stt->inQueue = 1; + } else + { + stt->inQueue = 2; + } + + /* call VAD (use low band only) */ + for (i = 0; i < samples; i += subFrames) + { + WebRtcAgc_ProcessVad(&stt->vadMic, &in_mic[i], subFrames); + } + + return 0; +} + +int WebRtcAgc_AddFarend(void *state, const WebRtc_Word16 *in_far, WebRtc_Word16 samples) +{ + WebRtc_Word32 errHandle = 0; + WebRtc_Word16 i, subFrames; + Agc_t *stt; + stt = (Agc_t *)state; + + if (stt == NULL) + { + return -1; + } + + if (stt->fs == 8000) + { + if ((samples != 80) && (samples != 160)) + { +#ifdef AGC_DEBUG //test log + fprintf(stt->fpt, + "AGC->add_far_end, frame %d: Invalid number of samples\n\n", + stt->fcount); +#endif + return -1; + } + subFrames = 80; + } else if (stt->fs == 16000) + { + if ((samples != 160) && (samples != 320)) + { +#ifdef AGC_DEBUG //test log + fprintf(stt->fpt, + "AGC->add_far_end, frame %d: Invalid number of samples\n\n", + stt->fcount); +#endif + return -1; + } + subFrames = 160; + } else if (stt->fs == 32000) + { + if ((samples != 160) && (samples != 320)) + { +#ifdef AGC_DEBUG //test log + fprintf(stt->fpt, + "AGC->add_far_end, frame %d: Invalid number of samples\n\n", + stt->fcount); +#endif + return -1; + } + subFrames = 160; + } else + { +#ifdef AGC_DEBUG //test log + fprintf(stt->fpt, + "AGC->add_far_end, frame %d: Invalid sample rate\n\n", + stt->fcount + 1); +#endif + return -1; + } + + for (i = 0; i < samples; i += subFrames) + { + errHandle += WebRtcAgc_AddFarendToDigital(&stt->digitalAgc, &in_far[i], subFrames); + } + + return errHandle; +} + +int WebRtcAgc_VirtualMic(void *agcInst, WebRtc_Word16 *in_near, WebRtc_Word16 *in_near_H, + WebRtc_Word16 samples, WebRtc_Word32 micLevelIn, + WebRtc_Word32 *micLevelOut) +{ + WebRtc_Word32 tmpFlt, micLevelTmp, gainIdx; + WebRtc_UWord16 gain; + WebRtc_Word16 ii; + Agc_t *stt; + + WebRtc_UWord32 nrg; + WebRtc_Word16 sampleCntr; + WebRtc_UWord32 frameNrg = 0; + WebRtc_UWord32 frameNrgLimit = 5500; + WebRtc_Word16 numZeroCrossing = 0; + const WebRtc_Word16 kZeroCrossingLowLim = 15; + const WebRtc_Word16 kZeroCrossingHighLim = 20; + + stt = (Agc_t *)agcInst; + + /* + * Before applying gain decide if this is a low-level signal. + * The idea is that digital AGC will not adapt to low-level + * signals. + */ + if (stt->fs != 8000) + { + frameNrgLimit = frameNrgLimit << 1; + } + + frameNrg = WEBRTC_SPL_MUL_16_16(in_near[0], in_near[0]); + for (sampleCntr = 1; sampleCntr < samples; sampleCntr++) + { + + // increment frame energy if it is less than the limit + // the correct value of the energy is not important + if (frameNrg < frameNrgLimit) + { + nrg = WEBRTC_SPL_MUL_16_16(in_near[sampleCntr], in_near[sampleCntr]); + frameNrg += nrg; + } + + // Count the zero crossings + numZeroCrossing += ((in_near[sampleCntr] ^ in_near[sampleCntr - 1]) < 0); + } + + if ((frameNrg < 500) || (numZeroCrossing <= 5)) + { + stt->lowLevelSignal = 1; + } else if (numZeroCrossing <= kZeroCrossingLowLim) + { + stt->lowLevelSignal = 0; + } else if (frameNrg <= frameNrgLimit) + { + stt->lowLevelSignal = 1; + } else if (numZeroCrossing >= kZeroCrossingHighLim) + { + stt->lowLevelSignal = 1; + } else + { + stt->lowLevelSignal = 0; + } + + micLevelTmp = WEBRTC_SPL_LSHIFT_W32(micLevelIn, stt->scale); + /* Set desired level */ + gainIdx = stt->micVol; + if (stt->micVol > stt->maxAnalog) + { + gainIdx = stt->maxAnalog; + } + if (micLevelTmp != stt->micRef) + { + /* Something has happened with the physical level, restart. */ + stt->micRef = micLevelTmp; + stt->micVol = 127; + *micLevelOut = 127; + stt->micGainIdx = 127; + gainIdx = 127; + } + /* Pre-process the signal to emulate the microphone level. */ + /* Take one step at a time in the gain table. */ + if (gainIdx > 127) + { + gain = kGainTableVirtualMic[gainIdx - 128]; + } else + { + gain = kSuppressionTableVirtualMic[127 - gainIdx]; + } + for (ii = 0; ii < samples; ii++) + { + tmpFlt = WEBRTC_SPL_RSHIFT_W32(WEBRTC_SPL_MUL_16_U16(in_near[ii], gain), 10); + if (tmpFlt > 32767) + { + tmpFlt = 32767; + gainIdx--; + if (gainIdx >= 127) + { + gain = kGainTableVirtualMic[gainIdx - 127]; + } else + { + gain = kSuppressionTableVirtualMic[127 - gainIdx]; + } + } + if (tmpFlt < -32768) + { + tmpFlt = -32768; + gainIdx--; + if (gainIdx >= 127) + { + gain = kGainTableVirtualMic[gainIdx - 127]; + } else + { + gain = kSuppressionTableVirtualMic[127 - gainIdx]; + } + } + in_near[ii] = (WebRtc_Word16)tmpFlt; + if (stt->fs == 32000) + { + tmpFlt = WEBRTC_SPL_MUL_16_U16(in_near_H[ii], gain); + tmpFlt = WEBRTC_SPL_RSHIFT_W32(tmpFlt, 10); + if (tmpFlt > 32767) + { + tmpFlt = 32767; + } + if (tmpFlt < -32768) + { + tmpFlt = -32768; + } + in_near_H[ii] = (WebRtc_Word16)tmpFlt; + } + } + /* Set the level we (finally) used */ + stt->micGainIdx = gainIdx; +// *micLevelOut = stt->micGainIdx; + *micLevelOut = WEBRTC_SPL_RSHIFT_W32(stt->micGainIdx, stt->scale); + /* Add to Mic as if it was the output from a true microphone */ + if (WebRtcAgc_AddMic(agcInst, in_near, in_near_H, samples) != 0) + { + return -1; + } + return 0; +} + +void WebRtcAgc_UpdateAgcThresholds(Agc_t *stt) +{ + + WebRtc_Word16 tmp16; +#ifdef MIC_LEVEL_FEEDBACK + int zeros; + + if (stt->micLvlSat) + { + /* Lower the analog target level since we have reached its maximum */ + zeros = WebRtcSpl_NormW32(stt->Rxx160_LPw32); + stt->targetIdxOffset = WEBRTC_SPL_RSHIFT_W16((3 * zeros) - stt->targetIdx - 2, 2); + } +#endif + + /* Set analog target level in envelope dBOv scale */ + tmp16 = (DIFF_REF_TO_ANALOG * stt->compressionGaindB) + ANALOG_TARGET_LEVEL_2; + tmp16 = WebRtcSpl_DivW32W16ResW16((WebRtc_Word32)tmp16, ANALOG_TARGET_LEVEL); + stt->analogTarget = DIGITAL_REF_AT_0_COMP_GAIN + tmp16; + if (stt->analogTarget < DIGITAL_REF_AT_0_COMP_GAIN) + { + stt->analogTarget = DIGITAL_REF_AT_0_COMP_GAIN; + } + if (stt->agcMode == kAgcModeFixedDigital) + { + /* Adjust for different parameter interpretation in FixedDigital mode */ + stt->analogTarget = stt->compressionGaindB; + } +#ifdef MIC_LEVEL_FEEDBACK + stt->analogTarget += stt->targetIdxOffset; +#endif + /* Since the offset between RMS and ENV is not constant, we should make this into a + * table, but for now, we'll stick with a constant, tuned for the chosen analog + * target level. + */ + stt->targetIdx = ANALOG_TARGET_LEVEL + OFFSET_ENV_TO_RMS; +#ifdef MIC_LEVEL_FEEDBACK + stt->targetIdx += stt->targetIdxOffset; +#endif + /* Analog adaptation limits */ + /* analogTargetLevel = round((32767*10^(-targetIdx/20))^2*16/2^7) */ + stt->analogTargetLevel = RXX_BUFFER_LEN * kTargetLevelTable[stt->targetIdx]; /* ex. -20 dBov */ + stt->startUpperLimit = RXX_BUFFER_LEN * kTargetLevelTable[stt->targetIdx - 1];/* -19 dBov */ + stt->startLowerLimit = RXX_BUFFER_LEN * kTargetLevelTable[stt->targetIdx + 1];/* -21 dBov */ + stt->upperPrimaryLimit = RXX_BUFFER_LEN * kTargetLevelTable[stt->targetIdx - 2];/* -18 dBov */ + stt->lowerPrimaryLimit = RXX_BUFFER_LEN * kTargetLevelTable[stt->targetIdx + 2];/* -22 dBov */ + stt->upperSecondaryLimit = RXX_BUFFER_LEN * kTargetLevelTable[stt->targetIdx - 5];/* -15 dBov */ + stt->lowerSecondaryLimit = RXX_BUFFER_LEN * kTargetLevelTable[stt->targetIdx + 5];/* -25 dBov */ + stt->upperLimit = stt->startUpperLimit; + stt->lowerLimit = stt->startLowerLimit; +} + +void WebRtcAgc_SaturationCtrl(Agc_t *stt, WebRtc_UWord8 *saturated, WebRtc_Word32 *env) +{ + WebRtc_Word16 i, tmpW16; + + /* Check if the signal is saturated */ + for (i = 0; i < 10; i++) + { + tmpW16 = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(env[i], 20); + if (tmpW16 > 875) + { + stt->envSum += tmpW16; + } + } + + if (stt->envSum > 25000) + { + *saturated = 1; + stt->envSum = 0; + } + + /* stt->envSum *= 0.99; */ + stt->envSum = (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT(stt->envSum, + (WebRtc_Word16)32440, 15); +} + +void WebRtcAgc_ZeroCtrl(Agc_t *stt, WebRtc_Word32 *inMicLevel, WebRtc_Word32 *env) +{ + WebRtc_Word16 i; + WebRtc_Word32 tmp32 = 0; + WebRtc_Word32 midVal; + + /* Is the input signal zero? */ + for (i = 0; i < 10; i++) + { + tmp32 += env[i]; + } + + /* Each block is allowed to have a few non-zero + * samples. + */ + if (tmp32 < 500) + { + stt->msZero += 10; + } else + { + stt->msZero = 0; + } + + if (stt->muteGuardMs > 0) + { + stt->muteGuardMs -= 10; + } + + if (stt->msZero > 500) + { + stt->msZero = 0; + + /* Increase microphone level only if it's less than 50% */ + midVal = WEBRTC_SPL_RSHIFT_W32(stt->maxAnalog + stt->minLevel + 1, 1); + if (*inMicLevel < midVal) + { + /* *inMicLevel *= 1.1; */ + tmp32 = WEBRTC_SPL_MUL(1126, *inMicLevel); + *inMicLevel = WEBRTC_SPL_RSHIFT_W32(tmp32, 10); + /* Reduces risk of a muted mic repeatedly triggering excessive levels due + * to zero signal detection. */ + *inMicLevel = WEBRTC_SPL_MIN(*inMicLevel, stt->zeroCtrlMax); + stt->micVol = *inMicLevel; + } + +#ifdef AGC_DEBUG //test log + fprintf(stt->fpt, + "\t\tAGC->zeroCntrl, frame %d: 500 ms under threshold, micVol:\n", + stt->fcount, stt->micVol); +#endif + + stt->activeSpeech = 0; + stt->Rxx16_LPw32Max = 0; + + /* The AGC has a tendency (due to problems with the VAD parameters), to + * vastly increase the volume after a muting event. This timer prevents + * upwards adaptation for a short period. */ + stt->muteGuardMs = kMuteGuardTimeMs; + } +} + +void WebRtcAgc_SpeakerInactiveCtrl(Agc_t *stt) +{ + /* Check if the near end speaker is inactive. + * If that is the case the VAD threshold is + * increased since the VAD speech model gets + * more sensitive to any sound after a long + * silence. + */ + + WebRtc_Word32 tmp32; + WebRtc_Word16 vadThresh; + + if (stt->vadMic.stdLongTerm < 2500) + { + stt->vadThreshold = 1500; + } else + { + vadThresh = kNormalVadThreshold; + if (stt->vadMic.stdLongTerm < 4500) + { + /* Scale between min and max threshold */ + vadThresh += WEBRTC_SPL_RSHIFT_W16(4500 - stt->vadMic.stdLongTerm, 1); + } + + /* stt->vadThreshold = (31 * stt->vadThreshold + vadThresh) / 32; */ + tmp32 = (WebRtc_Word32)vadThresh; + tmp32 += WEBRTC_SPL_MUL_16_16((WebRtc_Word16)31, stt->vadThreshold); + stt->vadThreshold = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(tmp32, 5); + } +} + +void WebRtcAgc_ExpCurve(WebRtc_Word16 volume, WebRtc_Word16 *index) +{ + // volume in Q14 + // index in [0-7] + /* 8 different curves */ + if (volume > 5243) + { + if (volume > 7864) + { + if (volume > 12124) + { + *index = 7; + } else + { + *index = 6; + } + } else + { + if (volume > 6554) + { + *index = 5; + } else + { + *index = 4; + } + } + } else + { + if (volume > 2621) + { + if (volume > 3932) + { + *index = 3; + } else + { + *index = 2; + } + } else + { + if (volume > 1311) + { + *index = 1; + } else + { + *index = 0; + } + } + } +} + +WebRtc_Word32 WebRtcAgc_ProcessAnalog(void *state, WebRtc_Word32 inMicLevel, + WebRtc_Word32 *outMicLevel, + WebRtc_Word16 vadLogRatio, + WebRtc_Word16 echo, WebRtc_UWord8 *saturationWarning) +{ + WebRtc_UWord32 tmpU32; + WebRtc_Word32 Rxx16w32, tmp32; + WebRtc_Word32 inMicLevelTmp, lastMicVol; + WebRtc_Word16 i; + WebRtc_UWord8 saturated = 0; + Agc_t *stt; + + stt = (Agc_t *)state; + inMicLevelTmp = WEBRTC_SPL_LSHIFT_W32(inMicLevel, stt->scale); + + if (inMicLevelTmp > stt->maxAnalog) + { +#ifdef AGC_DEBUG //test log + fprintf(stt->fpt, "\tAGC->ProcessAnalog, frame %d: micLvl > maxAnalog\n", stt->fcount); +#endif + return -1; + } else if (inMicLevelTmp < stt->minLevel) + { +#ifdef AGC_DEBUG //test log + fprintf(stt->fpt, "\tAGC->ProcessAnalog, frame %d: micLvl < minLevel\n", stt->fcount); +#endif + return -1; + } + + if (stt->firstCall == 0) + { + WebRtc_Word32 tmpVol; + stt->firstCall = 1; + tmp32 = WEBRTC_SPL_RSHIFT_W32((stt->maxLevel - stt->minLevel) * (WebRtc_Word32)51, 9); + tmpVol = (stt->minLevel + tmp32); + + /* If the mic level is very low at start, increase it! */ + if ((inMicLevelTmp < tmpVol) && (stt->agcMode == kAgcModeAdaptiveAnalog)) + { + inMicLevelTmp = tmpVol; + } + stt->micVol = inMicLevelTmp; + } + + /* Set the mic level to the previous output value if there is digital input gain */ + if ((inMicLevelTmp == stt->maxAnalog) && (stt->micVol > stt->maxAnalog)) + { + inMicLevelTmp = stt->micVol; + } + + /* If the mic level was manually changed to a very low value raise it! */ + if ((inMicLevelTmp != stt->micVol) && (inMicLevelTmp < stt->minOutput)) + { + tmp32 = WEBRTC_SPL_RSHIFT_W32((stt->maxLevel - stt->minLevel) * (WebRtc_Word32)51, 9); + inMicLevelTmp = (stt->minLevel + tmp32); + stt->micVol = inMicLevelTmp; +#ifdef MIC_LEVEL_FEEDBACK + //stt->numBlocksMicLvlSat = 0; +#endif +#ifdef AGC_DEBUG //test log + fprintf(stt->fpt, + "\tAGC->ProcessAnalog, frame %d: micLvl < minLevel by manual decrease, raise vol\n", + stt->fcount); +#endif + } + + if (inMicLevelTmp != stt->micVol) + { + // Incoming level mismatch; update our level. + // This could be the case if the volume is changed manually, or if the + // sound device has a low volume resolution. + stt->micVol = inMicLevelTmp; + } + + if (inMicLevelTmp > stt->maxLevel) + { + // Always allow the user to raise the volume above the maxLevel. + stt->maxLevel = inMicLevelTmp; + } + + // Store last value here, after we've taken care of manual updates etc. + lastMicVol = stt->micVol; + + /* Checks if the signal is saturated. Also a check if individual samples + * are larger than 12000 is done. If they are the counter for increasing + * the volume level is set to -100ms + */ + WebRtcAgc_SaturationCtrl(stt, &saturated, stt->env[0]); + + /* The AGC is always allowed to lower the level if the signal is saturated */ + if (saturated == 1) + { + /* Lower the recording level + * Rxx160_LP is adjusted down because it is so slow it could + * cause the AGC to make wrong decisions. */ + /* stt->Rxx160_LPw32 *= 0.875; */ + stt->Rxx160_LPw32 = WEBRTC_SPL_MUL(WEBRTC_SPL_RSHIFT_W32(stt->Rxx160_LPw32, 3), 7); + + stt->zeroCtrlMax = stt->micVol; + + /* stt->micVol *= 0.903; */ + tmp32 = inMicLevelTmp - stt->minLevel; + tmpU32 = WEBRTC_SPL_UMUL(29591, (WebRtc_UWord32)(tmp32)); + stt->micVol = (WebRtc_Word32)WEBRTC_SPL_RSHIFT_U32(tmpU32, 15) + stt->minLevel; + if (stt->micVol > lastMicVol - 2) + { + stt->micVol = lastMicVol - 2; + } + inMicLevelTmp = stt->micVol; + +#ifdef AGC_DEBUG //test log + fprintf(stt->fpt, + "\tAGC->ProcessAnalog, frame %d: saturated, micVol = %d\n", + stt->fcount, stt->micVol); +#endif + + if (stt->micVol < stt->minOutput) + { + *saturationWarning = 1; + } + + /* Reset counter for decrease of volume level to avoid + * decreasing too much. The saturation control can still + * lower the level if needed. */ + stt->msTooHigh = -100; + + /* Enable the control mechanism to ensure that our measure, + * Rxx160_LP, is in the correct range. This must be done since + * the measure is very slow. */ + stt->activeSpeech = 0; + stt->Rxx16_LPw32Max = 0; + + /* Reset to initial values */ + stt->msecSpeechInnerChange = kMsecSpeechInner; + stt->msecSpeechOuterChange = kMsecSpeechOuter; + stt->changeToSlowMode = 0; + + stt->muteGuardMs = 0; + + stt->upperLimit = stt->startUpperLimit; + stt->lowerLimit = stt->startLowerLimit; +#ifdef MIC_LEVEL_FEEDBACK + //stt->numBlocksMicLvlSat = 0; +#endif + } + + /* Check if the input speech is zero. If so the mic volume + * is increased. On some computers the input is zero up as high + * level as 17% */ + WebRtcAgc_ZeroCtrl(stt, &inMicLevelTmp, stt->env[0]); + + /* Check if the near end speaker is inactive. + * If that is the case the VAD threshold is + * increased since the VAD speech model gets + * more sensitive to any sound after a long + * silence. + */ + WebRtcAgc_SpeakerInactiveCtrl(stt); + + for (i = 0; i < 5; i++) + { + /* Computed on blocks of 16 samples */ + + Rxx16w32 = stt->Rxx16w32_array[0][i]; + + /* Rxx160w32 in Q(-7) */ + tmp32 = WEBRTC_SPL_RSHIFT_W32(Rxx16w32 - stt->Rxx16_vectorw32[stt->Rxx16pos], 3); + stt->Rxx160w32 = stt->Rxx160w32 + tmp32; + stt->Rxx16_vectorw32[stt->Rxx16pos] = Rxx16w32; + + /* Circular buffer */ + stt->Rxx16pos = stt->Rxx16pos++; + if (stt->Rxx16pos == RXX_BUFFER_LEN) + { + stt->Rxx16pos = 0; + } + + /* Rxx16_LPw32 in Q(-4) */ + tmp32 = WEBRTC_SPL_RSHIFT_W32(Rxx16w32 - stt->Rxx16_LPw32, kAlphaShortTerm); + stt->Rxx16_LPw32 = (stt->Rxx16_LPw32) + tmp32; + + if (vadLogRatio > stt->vadThreshold) + { + /* Speech detected! */ + + /* Check if Rxx160_LP is in the correct range. If + * it is too high/low then we set it to the maximum of + * Rxx16_LPw32 during the first 200ms of speech. + */ + if (stt->activeSpeech < 250) + { + stt->activeSpeech += 2; + + if (stt->Rxx16_LPw32 > stt->Rxx16_LPw32Max) + { + stt->Rxx16_LPw32Max = stt->Rxx16_LPw32; + } + } else if (stt->activeSpeech == 250) + { + stt->activeSpeech += 2; + tmp32 = WEBRTC_SPL_RSHIFT_W32(stt->Rxx16_LPw32Max, 3); + stt->Rxx160_LPw32 = WEBRTC_SPL_MUL(tmp32, RXX_BUFFER_LEN); + } + + tmp32 = WEBRTC_SPL_RSHIFT_W32(stt->Rxx160w32 - stt->Rxx160_LPw32, kAlphaLongTerm); + stt->Rxx160_LPw32 = stt->Rxx160_LPw32 + tmp32; + + if (stt->Rxx160_LPw32 > stt->upperSecondaryLimit) + { + stt->msTooHigh += 2; + stt->msTooLow = 0; + stt->changeToSlowMode = 0; + + if (stt->msTooHigh > stt->msecSpeechOuterChange) + { + stt->msTooHigh = 0; + + /* Lower the recording level */ + /* Multiply by 0.828125 which corresponds to decreasing ~0.8dB */ + tmp32 = WEBRTC_SPL_RSHIFT_W32(stt->Rxx160_LPw32, 6); + stt->Rxx160_LPw32 = WEBRTC_SPL_MUL(tmp32, 53); + + /* Reduce the max gain to avoid excessive oscillation + * (but never drop below the maximum analog level). + * stt->maxLevel = (15 * stt->maxLevel + stt->micVol) / 16; + */ + tmp32 = (15 * stt->maxLevel) + stt->micVol; + stt->maxLevel = WEBRTC_SPL_RSHIFT_W32(tmp32, 4); + stt->maxLevel = WEBRTC_SPL_MAX(stt->maxLevel, stt->maxAnalog); + + stt->zeroCtrlMax = stt->micVol; + + /* 0.95 in Q15 */ + tmp32 = inMicLevelTmp - stt->minLevel; + tmpU32 = WEBRTC_SPL_UMUL(31130, (WebRtc_UWord32)(tmp32)); + stt->micVol = (WebRtc_Word32)WEBRTC_SPL_RSHIFT_U32(tmpU32, 15) + stt->minLevel; + if (stt->micVol > lastMicVol - 1) + { + stt->micVol = lastMicVol - 1; + } + inMicLevelTmp = stt->micVol; + + /* Enable the control mechanism to ensure that our measure, + * Rxx160_LP, is in the correct range. + */ + stt->activeSpeech = 0; + stt->Rxx16_LPw32Max = 0; +#ifdef MIC_LEVEL_FEEDBACK + //stt->numBlocksMicLvlSat = 0; +#endif +#ifdef AGC_DEBUG //test log + fprintf(stt->fpt, + "\tAGC->ProcessAnalog, frame %d: measure > 2ndUpperLim, micVol = %d, maxLevel = %d\n", + stt->fcount, stt->micVol, stt->maxLevel); +#endif + } + } else if (stt->Rxx160_LPw32 > stt->upperLimit) + { + stt->msTooHigh += 2; + stt->msTooLow = 0; + stt->changeToSlowMode = 0; + + if (stt->msTooHigh > stt->msecSpeechInnerChange) + { + /* Lower the recording level */ + stt->msTooHigh = 0; + /* Multiply by 0.828125 which corresponds to decreasing ~0.8dB */ + tmp32 = WEBRTC_SPL_RSHIFT_W32(stt->Rxx160_LPw32, 6); + stt->Rxx160_LPw32 = WEBRTC_SPL_MUL(tmp32, 53); + + /* Reduce the max gain to avoid excessive oscillation + * (but never drop below the maximum analog level). + * stt->maxLevel = (15 * stt->maxLevel + stt->micVol) / 16; + */ + tmp32 = (15 * stt->maxLevel) + stt->micVol; + stt->maxLevel = WEBRTC_SPL_RSHIFT_W32(tmp32, 4); + stt->maxLevel = WEBRTC_SPL_MAX(stt->maxLevel, stt->maxAnalog); + + stt->zeroCtrlMax = stt->micVol; + + /* 0.965 in Q15 */ + tmp32 = inMicLevelTmp - stt->minLevel; + tmpU32 = WEBRTC_SPL_UMUL(31621, (WebRtc_UWord32)(inMicLevelTmp - stt->minLevel)); + stt->micVol = (WebRtc_Word32)WEBRTC_SPL_RSHIFT_U32(tmpU32, 15) + stt->minLevel; + if (stt->micVol > lastMicVol - 1) + { + stt->micVol = lastMicVol - 1; + } + inMicLevelTmp = stt->micVol; + +#ifdef MIC_LEVEL_FEEDBACK + //stt->numBlocksMicLvlSat = 0; +#endif +#ifdef AGC_DEBUG //test log + fprintf(stt->fpt, + "\tAGC->ProcessAnalog, frame %d: measure > UpperLim, micVol = %d, maxLevel = %d\n", + stt->fcount, stt->micVol, stt->maxLevel); +#endif + } + } else if (stt->Rxx160_LPw32 < stt->lowerSecondaryLimit) + { + stt->msTooHigh = 0; + stt->changeToSlowMode = 0; + stt->msTooLow += 2; + + if (stt->msTooLow > stt->msecSpeechOuterChange) + { + /* Raise the recording level */ + WebRtc_Word16 index, weightFIX; + WebRtc_Word16 volNormFIX = 16384; // =1 in Q14. + + stt->msTooLow = 0; + + /* Normalize the volume level */ + tmp32 = WEBRTC_SPL_LSHIFT_W32(inMicLevelTmp - stt->minLevel, 14); + if (stt->maxInit != stt->minLevel) + { + volNormFIX = (WebRtc_Word16)WEBRTC_SPL_DIV(tmp32, + (stt->maxInit - stt->minLevel)); + } + + /* Find correct curve */ + WebRtcAgc_ExpCurve(volNormFIX, &index); + + /* Compute weighting factor for the volume increase, 32^(-2*X)/2+1.05 */ + weightFIX = kOffset1[index] + - (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT(kSlope1[index], + volNormFIX, 13); + + /* stt->Rxx160_LPw32 *= 1.047 [~0.2 dB]; */ + tmp32 = WEBRTC_SPL_RSHIFT_W32(stt->Rxx160_LPw32, 6); + stt->Rxx160_LPw32 = WEBRTC_SPL_MUL(tmp32, 67); + + tmp32 = inMicLevelTmp - stt->minLevel; + tmpU32 = ((WebRtc_UWord32)weightFIX * (WebRtc_UWord32)(inMicLevelTmp - stt->minLevel)); + stt->micVol = (WebRtc_Word32)WEBRTC_SPL_RSHIFT_U32(tmpU32, 14) + stt->minLevel; + if (stt->micVol < lastMicVol + 2) + { + stt->micVol = lastMicVol + 2; + } + + inMicLevelTmp = stt->micVol; + +#ifdef MIC_LEVEL_FEEDBACK + /* Count ms in level saturation */ + //if (stt->micVol > stt->maxAnalog) { + if (stt->micVol > 150) + { + /* mic level is saturated */ + stt->numBlocksMicLvlSat++; + fprintf(stderr, "Sat mic Level: %d\n", stt->numBlocksMicLvlSat); + } +#endif +#ifdef AGC_DEBUG //test log + fprintf(stt->fpt, + "\tAGC->ProcessAnalog, frame %d: measure < 2ndLowerLim, micVol = %d\n", + stt->fcount, stt->micVol); +#endif + } + } else if (stt->Rxx160_LPw32 < stt->lowerLimit) + { + stt->msTooHigh = 0; + stt->changeToSlowMode = 0; + stt->msTooLow += 2; + + if (stt->msTooLow > stt->msecSpeechInnerChange) + { + /* Raise the recording level */ + WebRtc_Word16 index, weightFIX; + WebRtc_Word16 volNormFIX = 16384; // =1 in Q14. + + stt->msTooLow = 0; + + /* Normalize the volume level */ + tmp32 = WEBRTC_SPL_LSHIFT_W32(inMicLevelTmp - stt->minLevel, 14); + if (stt->maxInit != stt->minLevel) + { + volNormFIX = (WebRtc_Word16)WEBRTC_SPL_DIV(tmp32, + (stt->maxInit - stt->minLevel)); + } + + /* Find correct curve */ + WebRtcAgc_ExpCurve(volNormFIX, &index); + + /* Compute weighting factor for the volume increase, (3.^(-2.*X))/8+1 */ + weightFIX = kOffset2[index] + - (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT(kSlope2[index], + volNormFIX, 13); + + /* stt->Rxx160_LPw32 *= 1.047 [~0.2 dB]; */ + tmp32 = WEBRTC_SPL_RSHIFT_W32(stt->Rxx160_LPw32, 6); + stt->Rxx160_LPw32 = WEBRTC_SPL_MUL(tmp32, 67); + + tmp32 = inMicLevelTmp - stt->minLevel; + tmpU32 = ((WebRtc_UWord32)weightFIX * (WebRtc_UWord32)(inMicLevelTmp - stt->minLevel)); + stt->micVol = (WebRtc_Word32)WEBRTC_SPL_RSHIFT_U32(tmpU32, 14) + stt->minLevel; + if (stt->micVol < lastMicVol + 1) + { + stt->micVol = lastMicVol + 1; + } + + inMicLevelTmp = stt->micVol; + +#ifdef MIC_LEVEL_FEEDBACK + /* Count ms in level saturation */ + //if (stt->micVol > stt->maxAnalog) { + if (stt->micVol > 150) + { + /* mic level is saturated */ + stt->numBlocksMicLvlSat++; + fprintf(stderr, "Sat mic Level: %d\n", stt->numBlocksMicLvlSat); + } +#endif +#ifdef AGC_DEBUG //test log + fprintf(stt->fpt, + "\tAGC->ProcessAnalog, frame %d: measure < LowerLim, micVol = %d\n", + stt->fcount, stt->micVol); +#endif + + } + } else + { + /* The signal is inside the desired range which is: + * lowerLimit < Rxx160_LP/640 < upperLimit + */ + if (stt->changeToSlowMode > 4000) + { + stt->msecSpeechInnerChange = 1000; + stt->msecSpeechOuterChange = 500; + stt->upperLimit = stt->upperPrimaryLimit; + stt->lowerLimit = stt->lowerPrimaryLimit; + } else + { + stt->changeToSlowMode += 2; // in milliseconds + } + stt->msTooLow = 0; + stt->msTooHigh = 0; + + stt->micVol = inMicLevelTmp; + + } +#ifdef MIC_LEVEL_FEEDBACK + if (stt->numBlocksMicLvlSat > NUM_BLOCKS_IN_SAT_BEFORE_CHANGE_TARGET) + { + stt->micLvlSat = 1; + fprintf(stderr, "target before = %d (%d)\n", stt->analogTargetLevel, stt->targetIdx); + WebRtcAgc_UpdateAgcThresholds(stt); + WebRtcAgc_CalculateGainTable(&(stt->digitalAgc.gainTable[0]), + stt->compressionGaindB, stt->targetLevelDbfs, stt->limiterEnable, + stt->analogTarget); + stt->numBlocksMicLvlSat = 0; + stt->micLvlSat = 0; + fprintf(stderr, "target offset = %d\n", stt->targetIdxOffset); + fprintf(stderr, "target after = %d (%d)\n", stt->analogTargetLevel, stt->targetIdx); + } +#endif + } + } + + /* Ensure gain is not increased in presence of echo or after a mute event + * (but allow the zeroCtrl() increase on the frame of a mute detection). + */ + if (echo == 1 || (stt->muteGuardMs > 0 && stt->muteGuardMs < kMuteGuardTimeMs)) + { + if (stt->micVol > lastMicVol) + { + stt->micVol = lastMicVol; + } + } + + /* limit the gain */ + if (stt->micVol > stt->maxLevel) + { + stt->micVol = stt->maxLevel; + } else if (stt->micVol < stt->minOutput) + { + stt->micVol = stt->minOutput; + } + + *outMicLevel = WEBRTC_SPL_RSHIFT_W32(stt->micVol, stt->scale); + if (*outMicLevel > WEBRTC_SPL_RSHIFT_W32(stt->maxAnalog, stt->scale)) + { + *outMicLevel = WEBRTC_SPL_RSHIFT_W32(stt->maxAnalog, stt->scale); + } + + return 0; +} + +int WebRtcAgc_Process(void *agcInst, const WebRtc_Word16 *in_near, + const WebRtc_Word16 *in_near_H, WebRtc_Word16 samples, + WebRtc_Word16 *out, WebRtc_Word16 *out_H, WebRtc_Word32 inMicLevel, + WebRtc_Word32 *outMicLevel, WebRtc_Word16 echo, + WebRtc_UWord8 *saturationWarning) +{ + Agc_t *stt; + WebRtc_Word32 inMicLevelTmp; + WebRtc_Word16 subFrames, i; + WebRtc_UWord8 satWarningTmp = 0; + + stt = (Agc_t *)agcInst; + + // + if (stt == NULL) + { + return -1; + } + // + + + if (stt->fs == 8000) + { + if ((samples != 80) && (samples != 160)) + { +#ifdef AGC_DEBUG //test log + fprintf(stt->fpt, + "AGC->Process, frame %d: Invalid number of samples\n\n", stt->fcount); +#endif + return -1; + } + subFrames = 80; + } else if (stt->fs == 16000) + { + if ((samples != 160) && (samples != 320)) + { +#ifdef AGC_DEBUG //test log + fprintf(stt->fpt, + "AGC->Process, frame %d: Invalid number of samples\n\n", stt->fcount); +#endif + return -1; + } + subFrames = 160; + } else if (stt->fs == 32000) + { + if ((samples != 160) && (samples != 320)) + { +#ifdef AGC_DEBUG //test log + fprintf(stt->fpt, + "AGC->Process, frame %d: Invalid number of samples\n\n", stt->fcount); +#endif + return -1; + } + subFrames = 160; + } else + { +#ifdef AGC_DEBUG// test log + fprintf(stt->fpt, + "AGC->Process, frame %d: Invalid sample rate\n\n", stt->fcount); +#endif + return -1; + } + + /* Check for valid pointers based on sampling rate */ + if (stt->fs == 32000 && in_near_H == NULL) + { + return -1; + } + /* Check for valid pointers for low band */ + if (in_near == NULL) + { + return -1; + } + + *saturationWarning = 0; + //TODO: PUT IN RANGE CHECKING FOR INPUT LEVELS + *outMicLevel = inMicLevel; + inMicLevelTmp = inMicLevel; + + memcpy(out, in_near, samples * sizeof(WebRtc_Word16)); + if (stt->fs == 32000) + { + memcpy(out_H, in_near_H, samples * sizeof(WebRtc_Word16)); + } + +#ifdef AGC_DEBUG//test log + stt->fcount++; +#endif + + for (i = 0; i < samples; i += subFrames) + { + if (WebRtcAgc_ProcessDigital(&stt->digitalAgc, &in_near[i], &in_near_H[i], &out[i], &out_H[i], + stt->fs, stt->lowLevelSignal) == -1) + { +#ifdef AGC_DEBUG//test log + fprintf(stt->fpt, "AGC->Process, frame %d: Error from DigAGC\n\n", stt->fcount); +#endif + return -1; + } + if ((stt->agcMode < kAgcModeFixedDigital) && ((stt->lowLevelSignal == 0) + || (stt->agcMode != kAgcModeAdaptiveDigital))) + { + if (WebRtcAgc_ProcessAnalog(agcInst, inMicLevelTmp, outMicLevel, + stt->vadMic.logRatio, echo, saturationWarning) == -1) + { + return -1; + } + } +#ifdef AGC_DEBUG//test log + fprintf(stt->agcLog, "%5d\t%d\t%d\t%d\n", stt->fcount, inMicLevelTmp, *outMicLevel, stt->maxLevel, stt->micVol); +#endif + + /* update queue */ + if (stt->inQueue > 1) + { + memcpy(stt->env[0], stt->env[1], 10 * sizeof(WebRtc_Word32)); + memcpy(stt->Rxx16w32_array[0], stt->Rxx16w32_array[1], 5 * sizeof(WebRtc_Word32)); + } + + if (stt->inQueue > 0) + { + stt->inQueue--; + } + + /* If 20ms frames are used the input mic level must be updated so that + * the analog AGC does not think that there has been a manual volume + * change. */ + inMicLevelTmp = *outMicLevel; + + /* Store a positive saturation warning. */ + if (*saturationWarning == 1) + { + satWarningTmp = 1; + } + } + + /* Trigger the saturation warning if displayed by any of the frames. */ + *saturationWarning = satWarningTmp; + + return 0; +} + +int WebRtcAgc_set_config(void *agcInst, WebRtcAgc_config_t agcConfig) +{ + Agc_t *stt; + stt = (Agc_t *)agcInst; + + if (stt == NULL) + { + return -1; + } + + if (stt->initFlag != kInitCheck) + { + stt->lastError = AGC_UNINITIALIZED_ERROR; + return -1; + } + + if (agcConfig.limiterEnable != kAgcFalse && agcConfig.limiterEnable != kAgcTrue) + { + stt->lastError = AGC_BAD_PARAMETER_ERROR; + return -1; + } + stt->limiterEnable = agcConfig.limiterEnable; + stt->compressionGaindB = agcConfig.compressionGaindB; + if ((agcConfig.targetLevelDbfs < 0) || (agcConfig.targetLevelDbfs > 31)) + { + stt->lastError = AGC_BAD_PARAMETER_ERROR; + return -1; + } + stt->targetLevelDbfs = agcConfig.targetLevelDbfs; + + if (stt->agcMode == kAgcModeFixedDigital) + { + /* Adjust for different parameter interpretation in FixedDigital mode */ + stt->compressionGaindB += agcConfig.targetLevelDbfs; + } + + /* Update threshold levels for analog adaptation */ + WebRtcAgc_UpdateAgcThresholds(stt); + + /* Recalculate gain table */ + if (WebRtcAgc_CalculateGainTable(&(stt->digitalAgc.gainTable[0]), stt->compressionGaindB, + stt->targetLevelDbfs, stt->limiterEnable, stt->analogTarget) == -1) + { +#ifdef AGC_DEBUG//test log + fprintf(stt->fpt, "AGC->set_config, frame %d: Error from calcGainTable\n\n", stt->fcount); +#endif + return -1; + } + /* Store the config in a WebRtcAgc_config_t */ + stt->usedConfig.compressionGaindB = agcConfig.compressionGaindB; + stt->usedConfig.limiterEnable = agcConfig.limiterEnable; + stt->usedConfig.targetLevelDbfs = agcConfig.targetLevelDbfs; + + return 0; +} + +int WebRtcAgc_get_config(void *agcInst, WebRtcAgc_config_t *config) +{ + Agc_t *stt; + stt = (Agc_t *)agcInst; + + if (stt == NULL) + { + return -1; + } + + if (config == NULL) + { + stt->lastError = AGC_NULL_POINTER_ERROR; + return -1; + } + + if (stt->initFlag != kInitCheck) + { + stt->lastError = AGC_UNINITIALIZED_ERROR; + return -1; + } + + config->limiterEnable = stt->usedConfig.limiterEnable; + config->targetLevelDbfs = stt->usedConfig.targetLevelDbfs; + config->compressionGaindB = stt->usedConfig.compressionGaindB; + + return 0; +} + +int WebRtcAgc_Create(void **agcInst) +{ + Agc_t *stt; + if (agcInst == NULL) + { + return -1; + } + stt = (Agc_t *)malloc(sizeof(Agc_t)); + + *agcInst = stt; + if (stt == NULL) + { + return -1; + } + +#ifdef AGC_DEBUG + stt->fpt = fopen("./agc_test_log.txt", "wt"); + stt->agcLog = fopen("./agc_debug_log.txt", "wt"); + stt->digitalAgc.logFile = fopen("./agc_log.txt", "wt"); +#endif + + stt->initFlag = 0; + stt->lastError = 0; + + return 0; +} + +int WebRtcAgc_Free(void *state) +{ + Agc_t *stt; + + stt = (Agc_t *)state; +#ifdef AGC_DEBUG + fclose(stt->fpt); + fclose(stt->agcLog); + fclose(stt->digitalAgc.logFile); +#endif + free(stt); + + return 0; +} + +/* minLevel - Minimum volume level + * maxLevel - Maximum volume level + */ +int WebRtcAgc_Init(void *agcInst, WebRtc_Word32 minLevel, WebRtc_Word32 maxLevel, + WebRtc_Word16 agcMode, WebRtc_UWord32 fs) +{ + WebRtc_Word32 max_add, tmp32; + WebRtc_Word16 i; + int tmpNorm; + Agc_t *stt; + + /* typecast state pointer */ + stt = (Agc_t *)agcInst; + + if (WebRtcAgc_InitDigital(&stt->digitalAgc, agcMode) != 0) + { + stt->lastError = AGC_UNINITIALIZED_ERROR; + return -1; + } + + /* Analog AGC variables */ + stt->envSum = 0; + + /* mode = 0 - Only saturation protection + * 1 - Analog Automatic Gain Control [-targetLevelDbfs (default -3 dBOv)] + * 2 - Digital Automatic Gain Control [-targetLevelDbfs (default -3 dBOv)] + * 3 - Fixed Digital Gain [compressionGaindB (default 8 dB)] + */ +#ifdef AGC_DEBUG//test log + stt->fcount = 0; + fprintf(stt->fpt, "AGC->Init\n"); +#endif + if (agcMode < kAgcModeUnchanged || agcMode > kAgcModeFixedDigital) + { +#ifdef AGC_DEBUG//test log + fprintf(stt->fpt, "AGC->Init: error, incorrect mode\n\n"); +#endif + return -1; + } + stt->agcMode = agcMode; + stt->fs = fs; + + /* initialize input VAD */ + WebRtcAgc_InitVad(&stt->vadMic); + + /* If the volume range is smaller than 0-256 then + * the levels are shifted up to Q8-domain */ + tmpNorm = WebRtcSpl_NormU32((WebRtc_UWord32)maxLevel); + stt->scale = tmpNorm - 23; + if (stt->scale < 0) + { + stt->scale = 0; + } + // TODO(bjornv): Investigate if we really need to scale up a small range now when we have + // a guard against zero-increments. For now, we do not support scale up (scale = 0). + stt->scale = 0; + maxLevel = WEBRTC_SPL_LSHIFT_W32(maxLevel, stt->scale); + minLevel = WEBRTC_SPL_LSHIFT_W32(minLevel, stt->scale); + + /* Make minLevel and maxLevel static in AdaptiveDigital */ + if (stt->agcMode == kAgcModeAdaptiveDigital) + { + minLevel = 0; + maxLevel = 255; + stt->scale = 0; + } + /* The maximum supplemental volume range is based on a vague idea + * of how much lower the gain will be than the real analog gain. */ + max_add = WEBRTC_SPL_RSHIFT_W32(maxLevel - minLevel, 2); + + /* Minimum/maximum volume level that can be set */ + stt->minLevel = minLevel; + stt->maxAnalog = maxLevel; + stt->maxLevel = maxLevel + max_add; + stt->maxInit = stt->maxLevel; + + stt->zeroCtrlMax = stt->maxAnalog; + + /* Initialize micVol parameter */ + stt->micVol = stt->maxAnalog; + if (stt->agcMode == kAgcModeAdaptiveDigital) + { + stt->micVol = 127; /* Mid-point of mic level */ + } + stt->micRef = stt->micVol; + stt->micGainIdx = 127; +#ifdef MIC_LEVEL_FEEDBACK + stt->numBlocksMicLvlSat = 0; + stt->micLvlSat = 0; +#endif +#ifdef AGC_DEBUG//test log + fprintf(stt->fpt, + "AGC->Init: minLevel = %d, maxAnalog = %d, maxLevel = %d\n", + stt->minLevel, stt->maxAnalog, stt->maxLevel); +#endif + + /* Minimum output volume is 4% higher than the available lowest volume level */ + tmp32 = WEBRTC_SPL_RSHIFT_W32((stt->maxLevel - stt->minLevel) * (WebRtc_Word32)10, 8); + stt->minOutput = (stt->minLevel + tmp32); + + stt->msTooLow = 0; + stt->msTooHigh = 0; + stt->changeToSlowMode = 0; + stt->firstCall = 0; + stt->msZero = 0; + stt->muteGuardMs = 0; + stt->gainTableIdx = 0; + + stt->msecSpeechInnerChange = kMsecSpeechInner; + stt->msecSpeechOuterChange = kMsecSpeechOuter; + + stt->activeSpeech = 0; + stt->Rxx16_LPw32Max = 0; + + stt->vadThreshold = kNormalVadThreshold; + stt->inActive = 0; + + for (i = 0; i < RXX_BUFFER_LEN; i++) + { + stt->Rxx16_vectorw32[i] = (WebRtc_Word32)1000; /* -54dBm0 */ + } + stt->Rxx160w32 = 125 * RXX_BUFFER_LEN; /* (stt->Rxx16_vectorw32[0]>>3) = 125 */ + + stt->Rxx16pos = 0; + stt->Rxx16_LPw32 = (WebRtc_Word32)16284; /* Q(-4) */ + + for (i = 0; i < 5; i++) + { + stt->Rxx16w32_array[0][i] = 0; + } + for (i = 0; i < 20; i++) + { + stt->env[0][i] = 0; + } + stt->inQueue = 0; + +#ifdef MIC_LEVEL_FEEDBACK + stt->targetIdxOffset = 0; +#endif + + WebRtcSpl_MemSetW32(stt->filterState, 0, 8); + + stt->initFlag = kInitCheck; + // Default config settings. + stt->defaultConfig.limiterEnable = kAgcTrue; + stt->defaultConfig.targetLevelDbfs = AGC_DEFAULT_TARGET_LEVEL; + stt->defaultConfig.compressionGaindB = AGC_DEFAULT_COMP_GAIN; + + if (WebRtcAgc_set_config(stt, stt->defaultConfig) == -1) + { + stt->lastError = AGC_UNSPECIFIED_ERROR; + return -1; + } + stt->Rxx160_LPw32 = stt->analogTargetLevel; // Initialize rms value + + stt->lowLevelSignal = 0; + + /* Only positive values are allowed that are not too large */ + if ((minLevel >= maxLevel) || (maxLevel & 0xFC000000)) + { +#ifdef AGC_DEBUG//test log + fprintf(stt->fpt, "minLevel, maxLevel value(s) are invalid\n\n"); +#endif + return -1; + } else + { +#ifdef AGC_DEBUG//test log + fprintf(stt->fpt, "\n"); +#endif + return 0; + } +} + +int WebRtcAgc_Version(WebRtc_Word8 *versionStr, WebRtc_Word16 length) +{ + const WebRtc_Word8 version[] = "AGC 1.7.0"; + const WebRtc_Word16 versionLen = (WebRtc_Word16)strlen(version) + 1; + + if (versionStr == NULL) + { + return -1; + } + + if (versionLen > length) + { + return -1; + } + + strncpy(versionStr, version, versionLen); + return 0; +} diff --git a/src/modules/audio_processing/agc/main/source/analog_agc.h b/src/modules/audio_processing/agc/main/source/analog_agc.h new file mode 100644 index 0000000000..b32ac6581e --- /dev/null +++ b/src/modules/audio_processing/agc/main/source/analog_agc.h @@ -0,0 +1,133 @@ +/* + * Copyright (c) 2011 The WebRTC project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AGC_MAIN_SOURCE_ANALOG_AGC_H_ +#define WEBRTC_MODULES_AUDIO_PROCESSING_AGC_MAIN_SOURCE_ANALOG_AGC_H_ + +#include "typedefs.h" +#include "gain_control.h" +#include "digital_agc.h" + +//#define AGC_DEBUG +//#define MIC_LEVEL_FEEDBACK +#ifdef AGC_DEBUG +#include <stdio.h> +#endif + +/* Analog Automatic Gain Control variables: + * Constant declarations (inner limits inside which no changes are done) + * In the beginning the range is narrower to widen as soon as the measure + * 'Rxx160_LP' is inside it. Currently the starting limits are -22.2+/-1dBm0 + * and the final limits -22.2+/-2.5dBm0. These levels makes the speech signal + * go towards -25.4dBm0 (-31.4dBov). Tuned with wbfile-31.4dBov.pcm + * The limits are created by running the AGC with a file having the desired + * signal level and thereafter plotting Rxx160_LP in the dBm0-domain defined + * by out=10*log10(in/260537279.7); Set the target level to the average level + * of our measure Rxx160_LP. Remember that the levels are in blocks of 16 in + * Q(-7). (Example matlab code: round(db2pow(-21.2)*16/2^7) ) + */ +#define RXX_BUFFER_LEN 10 + +static const WebRtc_Word16 kMsecSpeechInner = 520; +static const WebRtc_Word16 kMsecSpeechOuter = 340; + +static const WebRtc_Word16 kNormalVadThreshold = 400; + +static const WebRtc_Word16 kAlphaShortTerm = 6; // 1 >> 6 = 0.0156 +static const WebRtc_Word16 kAlphaLongTerm = 10; // 1 >> 10 = 0.000977 + +typedef struct +{ + // Configurable parameters/variables + WebRtc_UWord32 fs; // Sampling frequency + WebRtc_Word16 compressionGaindB; // Fixed gain level in dB + WebRtc_Word16 targetLevelDbfs; // Target level in -dBfs of envelope (default -3) + WebRtc_Word16 agcMode; // Hard coded mode (adaptAna/adaptDig/fixedDig) + WebRtc_UWord8 limiterEnable; // Enabling limiter (on/off (default off)) + WebRtcAgc_config_t defaultConfig; + WebRtcAgc_config_t usedConfig; + + // General variables + WebRtc_Word16 initFlag; + WebRtc_Word16 lastError; + + // Target level parameters + // Based on the above: analogTargetLevel = round((32767*10^(-22/20))^2*16/2^7) + WebRtc_Word32 analogTargetLevel; // = RXX_BUFFER_LEN * 846805; -22 dBfs + WebRtc_Word32 startUpperLimit; // = RXX_BUFFER_LEN * 1066064; -21 dBfs + WebRtc_Word32 startLowerLimit; // = RXX_BUFFER_LEN * 672641; -23 dBfs + WebRtc_Word32 upperPrimaryLimit; // = RXX_BUFFER_LEN * 1342095; -20 dBfs + WebRtc_Word32 lowerPrimaryLimit; // = RXX_BUFFER_LEN * 534298; -24 dBfs + WebRtc_Word32 upperSecondaryLimit;// = RXX_BUFFER_LEN * 2677832; -17 dBfs + WebRtc_Word32 lowerSecondaryLimit;// = RXX_BUFFER_LEN * 267783; -27 dBfs + WebRtc_UWord16 targetIdx; // Table index for corresponding target level +#ifdef MIC_LEVEL_FEEDBACK + WebRtc_UWord16 targetIdxOffset; // Table index offset for level compensation +#endif + WebRtc_Word16 analogTarget; // Digital reference level in ENV scale + + // Analog AGC specific variables + WebRtc_Word32 filterState[8]; // For downsampling wb to nb + WebRtc_Word32 upperLimit; // Upper limit for mic energy + WebRtc_Word32 lowerLimit; // Lower limit for mic energy + WebRtc_Word32 Rxx160w32; // Average energy for one frame + WebRtc_Word32 Rxx16_LPw32; // Low pass filtered subframe energies + WebRtc_Word32 Rxx160_LPw32; // Low pass filtered frame energies + WebRtc_Word32 Rxx16_LPw32Max; // Keeps track of largest energy subframe + WebRtc_Word32 Rxx16_vectorw32[RXX_BUFFER_LEN];// Array with subframe energies + WebRtc_Word32 Rxx16w32_array[2][5];// Energy values of microphone signal + WebRtc_Word32 env[2][10]; // Envelope values of subframes + + WebRtc_Word16 Rxx16pos; // Current position in the Rxx16_vectorw32 + WebRtc_Word16 envSum; // Filtered scaled envelope in subframes + WebRtc_Word16 vadThreshold; // Threshold for VAD decision + WebRtc_Word16 inActive; // Inactive time in milliseconds + WebRtc_Word16 msTooLow; // Milliseconds of speech at a too low level + WebRtc_Word16 msTooHigh; // Milliseconds of speech at a too high level + WebRtc_Word16 changeToSlowMode; // Change to slow mode after some time at target + WebRtc_Word16 firstCall; // First call to the process-function + WebRtc_Word16 msZero; // Milliseconds of zero input + WebRtc_Word16 msecSpeechOuterChange;// Min ms of speech between volume changes + WebRtc_Word16 msecSpeechInnerChange;// Min ms of speech between volume changes + WebRtc_Word16 activeSpeech; // Milliseconds of active speech + WebRtc_Word16 muteGuardMs; // Counter to prevent mute action + WebRtc_Word16 inQueue; // 10 ms batch indicator + + // Microphone level variables + WebRtc_Word32 micRef; // Remember ref. mic level for virtual mic + WebRtc_UWord16 gainTableIdx; // Current position in virtual gain table + WebRtc_Word32 micGainIdx; // Gain index of mic level to increase slowly + WebRtc_Word32 micVol; // Remember volume between frames + WebRtc_Word32 maxLevel; // Max possible vol level, incl dig gain + WebRtc_Word32 maxAnalog; // Maximum possible analog volume level + WebRtc_Word32 maxInit; // Initial value of "max" + WebRtc_Word32 minLevel; // Minimum possible volume level + WebRtc_Word32 minOutput; // Minimum output volume level + WebRtc_Word32 zeroCtrlMax; // Remember max gain => don't amp low input + + WebRtc_Word16 scale; // Scale factor for internal volume levels +#ifdef MIC_LEVEL_FEEDBACK + WebRtc_Word16 numBlocksMicLvlSat; + WebRtc_UWord8 micLvlSat; +#endif + // Structs for VAD and digital_agc + AgcVad_t vadMic; + DigitalAgc_t digitalAgc; + +#ifdef AGC_DEBUG + FILE* fpt; + FILE* agcLog; + WebRtc_Word32 fcount; +#endif + + WebRtc_Word16 lowLevelSignal; +} Agc_t; + +#endif // WEBRTC_MODULES_AUDIO_PROCESSING_AGC_MAIN_SOURCE_ANALOG_AGC_H_ diff --git a/src/modules/audio_processing/agc/main/source/digital_agc.c b/src/modules/audio_processing/agc/main/source/digital_agc.c new file mode 100644 index 0000000000..2966586e48 --- /dev/null +++ b/src/modules/audio_processing/agc/main/source/digital_agc.c @@ -0,0 +1,780 @@ +/* + * Copyright (c) 2011 The WebRTC project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +/* digital_agc.c + * + */ + +#include <string.h> +#ifdef AGC_DEBUG +#include <stdio.h> +#endif +#include "digital_agc.h" +#include "gain_control.h" + +// To generate the gaintable, copy&paste the following lines to a Matlab window: +// MaxGain = 6; MinGain = 0; CompRatio = 3; Knee = 1; +// zeros = 0:31; lvl = 2.^(1-zeros); +// A = -10*log10(lvl) * (CompRatio - 1) / CompRatio; +// B = MaxGain - MinGain; +// gains = round(2^16*10.^(0.05 * (MinGain + B * ( log(exp(-Knee*A)+exp(-Knee*B)) - log(1+exp(-Knee*B)) ) / log(1/(1+exp(Knee*B)))))); +// fprintf(1, '\t%i, %i, %i, %i,\n', gains); +// % Matlab code for plotting the gain and input/output level characteristic (copy/paste the following 3 lines): +// in = 10*log10(lvl); out = 20*log10(gains/65536); +// subplot(121); plot(in, out); axis([-30, 0, -5, 20]); grid on; xlabel('Input (dB)'); ylabel('Gain (dB)'); +// subplot(122); plot(in, in+out); axis([-30, 0, -30, 5]); grid on; xlabel('Input (dB)'); ylabel('Output (dB)'); +// zoom on; + +// Generator table for y=log2(1+e^x) in Q8. +static const WebRtc_UWord16 kGenFuncTable[128] = { + 256, 485, 786, 1126, 1484, 1849, 2217, 2586, + 2955, 3324, 3693, 4063, 4432, 4801, 5171, 5540, + 5909, 6279, 6648, 7017, 7387, 7756, 8125, 8495, + 8864, 9233, 9603, 9972, 10341, 10711, 11080, 11449, + 11819, 12188, 12557, 12927, 13296, 13665, 14035, 14404, + 14773, 15143, 15512, 15881, 16251, 16620, 16989, 17359, + 17728, 18097, 18466, 18836, 19205, 19574, 19944, 20313, + 20682, 21052, 21421, 21790, 22160, 22529, 22898, 23268, + 23637, 24006, 24376, 24745, 25114, 25484, 25853, 26222, + 26592, 26961, 27330, 27700, 28069, 28438, 28808, 29177, + 29546, 29916, 30285, 30654, 31024, 31393, 31762, 32132, + 32501, 32870, 33240, 33609, 33978, 34348, 34717, 35086, + 35456, 35825, 36194, 36564, 36933, 37302, 37672, 38041, + 38410, 38780, 39149, 39518, 39888, 40257, 40626, 40996, + 41365, 41734, 42104, 42473, 42842, 43212, 43581, 43950, + 44320, 44689, 45058, 45428, 45797, 46166, 46536, 46905 +}; + +static const WebRtc_Word16 kAvgDecayTime = 250; // frames; < 3000 + +WebRtc_Word32 WebRtcAgc_CalculateGainTable(WebRtc_Word32 *gainTable, // Q16 + WebRtc_Word16 digCompGaindB, // Q0 + WebRtc_Word16 targetLevelDbfs,// Q0 + WebRtc_UWord8 limiterEnable, + WebRtc_Word16 analogTarget) // Q0 +{ + // This function generates the compressor gain table used in the fixed digital part. + WebRtc_UWord32 tmpU32no1, tmpU32no2, absInLevel, logApprox; + WebRtc_Word32 inLevel, limiterLvl; + WebRtc_Word32 tmp32, tmp32no1, tmp32no2, numFIX, den, y32; + const WebRtc_UWord16 kLog10 = 54426; // log2(10) in Q14 + const WebRtc_UWord16 kLog10_2 = 49321; // 10*log10(2) in Q14 + const WebRtc_UWord16 kLogE_1 = 23637; // log2(e) in Q14 + WebRtc_UWord16 constMaxGain; + WebRtc_UWord16 tmpU16, intPart, fracPart; + const WebRtc_Word16 kCompRatio = 3; + const WebRtc_Word16 kSoftLimiterLeft = 1; + WebRtc_Word16 limiterOffset = 0; // Limiter offset + WebRtc_Word16 limiterIdx, limiterLvlX; + WebRtc_Word16 constLinApprox, zeroGainLvl, maxGain, diffGain; + WebRtc_Word16 i, tmp16, tmp16no1; + int zeros, zerosScale; + + // Constants +// kLogE_1 = 23637; // log2(e) in Q14 +// kLog10 = 54426; // log2(10) in Q14 +// kLog10_2 = 49321; // 10*log10(2) in Q14 + + // Calculate maximum digital gain and zero gain level + tmp32no1 = WEBRTC_SPL_MUL_16_16(digCompGaindB - analogTarget, kCompRatio - 1); + tmp16no1 = analogTarget - targetLevelDbfs; + tmp16no1 += WebRtcSpl_DivW32W16ResW16(tmp32no1 + (kCompRatio >> 1), kCompRatio); + maxGain = WEBRTC_SPL_MAX(tmp16no1, (analogTarget - targetLevelDbfs)); + tmp32no1 = WEBRTC_SPL_MUL_16_16(maxGain, kCompRatio); + zeroGainLvl = digCompGaindB; + zeroGainLvl -= WebRtcSpl_DivW32W16ResW16(tmp32no1 + ((kCompRatio - 1) >> 1), + kCompRatio - 1); + if ((digCompGaindB <= analogTarget) && (limiterEnable)) + { + zeroGainLvl += (analogTarget - digCompGaindB + kSoftLimiterLeft); + limiterOffset = 0; + } + + // Calculate the difference between maximum gain and gain at 0dB0v: + // diffGain = maxGain + (compRatio-1)*zeroGainLvl/compRatio + // = (compRatio-1)*digCompGaindB/compRatio + tmp32no1 = WEBRTC_SPL_MUL_16_16(digCompGaindB, kCompRatio - 1); + diffGain = WebRtcSpl_DivW32W16ResW16(tmp32no1 + (kCompRatio >> 1), kCompRatio); + if (diffGain < 0) + { + return -1; + } + + // Calculate the limiter level and index: + // limiterLvlX = analogTarget - limiterOffset + // limiterLvl = targetLevelDbfs + limiterOffset/compRatio + limiterLvlX = analogTarget - limiterOffset; + limiterIdx = 2 + + WebRtcSpl_DivW32W16ResW16(WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)limiterLvlX, 13), + WEBRTC_SPL_RSHIFT_U16(kLog10_2, 1)); + tmp16no1 = WebRtcSpl_DivW32W16ResW16(limiterOffset + (kCompRatio >> 1), kCompRatio); + limiterLvl = targetLevelDbfs + tmp16no1; + + // Calculate (through table lookup): + // constMaxGain = log2(1+2^(log2(e)*diffGain)); (in Q8) + constMaxGain = kGenFuncTable[diffGain]; // in Q8 + + // Calculate a parameter used to approximate the fractional part of 2^x with a + // piecewise linear function in Q14: + // constLinApprox = round(3/2*(4*(3-2*sqrt(2))/(log(2)^2)-0.5)*2^14); + constLinApprox = 22817; // in Q14 + + // Calculate a denominator used in the exponential part to convert from dB to linear scale: + // den = 20*constMaxGain (in Q8) + den = WEBRTC_SPL_MUL_16_U16(20, constMaxGain); // in Q8 + + for (i = 0; i < 32; i++) + { + // Calculate scaled input level (compressor): + // inLevel = fix((-constLog10_2*(compRatio-1)*(1-i)+fix(compRatio/2))/compRatio) + tmp16 = (WebRtc_Word16)WEBRTC_SPL_MUL_16_16(kCompRatio - 1, i - 1); // Q0 + tmp32 = WEBRTC_SPL_MUL_16_U16(tmp16, kLog10_2) + 1; // Q14 + inLevel = WebRtcSpl_DivW32W16(tmp32, kCompRatio); // Q14 + + // Calculate diffGain-inLevel, to map using the genFuncTable + inLevel = WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)diffGain, 14) - inLevel; // Q14 + + // Make calculations on abs(inLevel) and compensate for the sign afterwards. + absInLevel = (WebRtc_UWord32)WEBRTC_SPL_ABS_W32(inLevel); // Q14 + + // LUT with interpolation + intPart = (WebRtc_UWord16)WEBRTC_SPL_RSHIFT_U32(absInLevel, 14); + fracPart = (WebRtc_UWord16)(absInLevel & 0x00003FFF); // extract the fractional part + tmpU16 = kGenFuncTable[intPart + 1] - kGenFuncTable[intPart]; // Q8 + tmpU32no1 = WEBRTC_SPL_UMUL_16_16(tmpU16, fracPart); // Q22 + tmpU32no1 += WEBRTC_SPL_LSHIFT_U32((WebRtc_UWord32)kGenFuncTable[intPart], 14); // Q22 + logApprox = WEBRTC_SPL_RSHIFT_U32(tmpU32no1, 8); // Q14 + // Compensate for negative exponent using the relation: + // log2(1 + 2^-x) = log2(1 + 2^x) - x + if (inLevel < 0) + { + zeros = WebRtcSpl_NormU32(absInLevel); + zerosScale = 0; + if (zeros < 15) + { + // Not enough space for multiplication + tmpU32no2 = WEBRTC_SPL_RSHIFT_U32(absInLevel, 15 - zeros); // Q(zeros-1) + tmpU32no2 = WEBRTC_SPL_UMUL_32_16(tmpU32no2, kLogE_1); // Q(zeros+13) + if (zeros < 9) + { + tmpU32no1 = WEBRTC_SPL_RSHIFT_U32(tmpU32no1, 9 - zeros); // Q(zeros+13) + zerosScale = 9 - zeros; + } else + { + tmpU32no2 = WEBRTC_SPL_RSHIFT_U32(tmpU32no2, zeros - 9); // Q22 + } + } else + { + tmpU32no2 = WEBRTC_SPL_UMUL_32_16(absInLevel, kLogE_1); // Q28 + tmpU32no2 = WEBRTC_SPL_RSHIFT_U32(tmpU32no2, 6); // Q22 + } + logApprox = 0; + if (tmpU32no2 < tmpU32no1) + { + logApprox = WEBRTC_SPL_RSHIFT_U32(tmpU32no1 - tmpU32no2, 8 - zerosScale); //Q14 + } + } + numFIX = WEBRTC_SPL_LSHIFT_W32(WEBRTC_SPL_MUL_16_U16(maxGain, constMaxGain), 6); // Q14 + numFIX -= WEBRTC_SPL_MUL_32_16((WebRtc_Word32)logApprox, diffGain); // Q14 + + // Calculate ratio + // Shift numFIX as much as possible + zeros = WebRtcSpl_NormW32(numFIX); + numFIX = WEBRTC_SPL_LSHIFT_W32(numFIX, zeros); // Q(14+zeros) + + // Shift den so we end up in Qy1 + tmp32no1 = WEBRTC_SPL_SHIFT_W32(den, zeros - 8); // Q(zeros) + if (numFIX < 0) + { + numFIX -= WEBRTC_SPL_RSHIFT_W32(tmp32no1, 1); + } else + { + numFIX += WEBRTC_SPL_RSHIFT_W32(tmp32no1, 1); + } + y32 = WEBRTC_SPL_DIV(numFIX, tmp32no1); // in Q14 + if (limiterEnable && (i < limiterIdx)) + { + tmp32 = WEBRTC_SPL_MUL_16_U16(i - 1, kLog10_2); // Q14 + tmp32 -= WEBRTC_SPL_LSHIFT_W32(limiterLvl, 14); // Q14 + y32 = WebRtcSpl_DivW32W16(tmp32 + 10, 20); + } + if (y32 > 39000) + { + tmp32 = WEBRTC_SPL_MUL(y32 >> 1, kLog10) + 4096; // in Q27 + tmp32 = WEBRTC_SPL_RSHIFT_W32(tmp32, 13); // in Q14 + } else + { + tmp32 = WEBRTC_SPL_MUL(y32, kLog10) + 8192; // in Q28 + tmp32 = WEBRTC_SPL_RSHIFT_W32(tmp32, 14); // in Q14 + } + tmp32 += WEBRTC_SPL_LSHIFT_W32(16, 14); // in Q14 (Make sure final output is in Q16) + + // Calculate power + if (tmp32 > 0) + { + intPart = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(tmp32, 14); + fracPart = (WebRtc_UWord16)(tmp32 & 0x00003FFF); // in Q14 + if (WEBRTC_SPL_RSHIFT_W32(fracPart, 13)) + { + tmp16 = WEBRTC_SPL_LSHIFT_W16(2, 14) - constLinApprox; + tmp32no2 = WEBRTC_SPL_LSHIFT_W32(1, 14) - fracPart; + tmp32no2 = WEBRTC_SPL_MUL_32_16(tmp32no2, tmp16); + tmp32no2 = WEBRTC_SPL_RSHIFT_W32(tmp32no2, 13); + tmp32no2 = WEBRTC_SPL_LSHIFT_W32(1, 14) - tmp32no2; + } else + { + tmp16 = constLinApprox - WEBRTC_SPL_LSHIFT_W16(1, 14); + tmp32no2 = WEBRTC_SPL_MUL_32_16(fracPart, tmp16); + tmp32no2 = WEBRTC_SPL_RSHIFT_W32(tmp32no2, 13); + } + fracPart = (WebRtc_UWord16)tmp32no2; + gainTable[i] = WEBRTC_SPL_LSHIFT_W32(1, intPart) + + WEBRTC_SPL_SHIFT_W32(fracPart, intPart - 14); + } else + { + gainTable[i] = 0; + } + } + + return 0; +} + +WebRtc_Word32 WebRtcAgc_InitDigital(DigitalAgc_t *stt, WebRtc_Word16 agcMode) +{ + + if (agcMode == kAgcModeFixedDigital) + { + // start at minimum to find correct gain faster + stt->capacitorSlow = 0; + } else + { + // start out with 0 dB gain + stt->capacitorSlow = 134217728; // (WebRtc_Word32)(0.125f * 32768.0f * 32768.0f); + } + stt->capacitorFast = 0; + stt->gain = 65536; + stt->gatePrevious = 0; + stt->agcMode = agcMode; +#ifdef AGC_DEBUG + stt->frameCounter = 0; +#endif + + // initialize VADs + WebRtcAgc_InitVad(&stt->vadNearend); + WebRtcAgc_InitVad(&stt->vadFarend); + + return 0; +} + +WebRtc_Word32 WebRtcAgc_AddFarendToDigital(DigitalAgc_t *stt, const WebRtc_Word16 *in_far, + WebRtc_Word16 nrSamples) +{ + // Check for valid pointer + if (&stt->vadFarend == NULL) + { + return -1; + } + + // VAD for far end + WebRtcAgc_ProcessVad(&stt->vadFarend, in_far, nrSamples); + + return 0; +} + +WebRtc_Word32 WebRtcAgc_ProcessDigital(DigitalAgc_t *stt, const WebRtc_Word16 *in_near, + const WebRtc_Word16 *in_near_H, WebRtc_Word16 *out, + WebRtc_Word16 *out_H, WebRtc_UWord32 FS, + WebRtc_Word16 lowlevelSignal) +{ + // array for gains (one value per ms, incl start & end) + WebRtc_Word32 gains[11]; + + WebRtc_Word32 out_tmp, tmp32; + WebRtc_Word32 env[10]; + WebRtc_Word32 nrg, max_nrg; + WebRtc_Word32 cur_level; + WebRtc_Word32 gain32, delta; + WebRtc_Word16 logratio; + WebRtc_Word16 lower_thr, upper_thr; + WebRtc_Word16 zeros, zeros_fast, frac; + WebRtc_Word16 decay; + WebRtc_Word16 gate, gain_adj; + WebRtc_Word16 k, n; + WebRtc_Word16 L, L2; // samples/subframe + + // determine number of samples per ms + if (FS == 8000) + { + L = 8; + L2 = 3; + } else if (FS == 16000) + { + L = 16; + L2 = 4; + } else if (FS == 32000) + { + L = 16; + L2 = 4; + } else + { + return -1; + } + + memcpy(out, in_near, 10 * L * sizeof(WebRtc_Word16)); + if (FS == 32000) + { + memcpy(out_H, in_near_H, 10 * L * sizeof(WebRtc_Word16)); + } + // VAD for near end + logratio = WebRtcAgc_ProcessVad(&stt->vadNearend, out, L * 10); + + // Account for far end VAD + if (stt->vadFarend.counter > 10) + { + tmp32 = WEBRTC_SPL_MUL_16_16(3, logratio); + logratio = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(tmp32 - stt->vadFarend.logRatio, 2); + } + + // Determine decay factor depending on VAD + // upper_thr = 1.0f; + // lower_thr = 0.25f; + upper_thr = 1024; // Q10 + lower_thr = 0; // Q10 + if (logratio > upper_thr) + { + // decay = -2^17 / DecayTime; -> -65 + decay = -65; + } else if (logratio < lower_thr) + { + decay = 0; + } else + { + // decay = (WebRtc_Word16)(((lower_thr - logratio) + // * (2^27/(DecayTime*(upper_thr-lower_thr)))) >> 10); + // SUBSTITUTED: 2^27/(DecayTime*(upper_thr-lower_thr)) -> 65 + tmp32 = WEBRTC_SPL_MUL_16_16((lower_thr - logratio), 65); + decay = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(tmp32, 10); + } + + // adjust decay factor for long silence (detected as low standard deviation) + // This is only done in the adaptive modes + if (stt->agcMode != kAgcModeFixedDigital) + { + if (stt->vadNearend.stdLongTerm < 4000) + { + decay = 0; + } else if (stt->vadNearend.stdLongTerm < 8096) + { + // decay = (WebRtc_Word16)(((stt->vadNearend.stdLongTerm - 4000) * decay) >> 12); + tmp32 = WEBRTC_SPL_MUL_16_16((stt->vadNearend.stdLongTerm - 4000), decay); + decay = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(tmp32, 12); + } + + if (lowlevelSignal != 0) + { + decay = 0; + } + } +#ifdef AGC_DEBUG + stt->frameCounter++; + fprintf(stt->logFile, "%5.2f\t%d\t%d\t%d\t", (float)(stt->frameCounter) / 100, logratio, decay, stt->vadNearend.stdLongTerm); +#endif + // Find max amplitude per sub frame + // iterate over sub frames + for (k = 0; k < 10; k++) + { + // iterate over samples + max_nrg = 0; + for (n = 0; n < L; n++) + { + nrg = WEBRTC_SPL_MUL_16_16(out[k * L + n], out[k * L + n]); + if (nrg > max_nrg) + { + max_nrg = nrg; + } + } + env[k] = max_nrg; + } + + // Calculate gain per sub frame + gains[0] = stt->gain; + for (k = 0; k < 10; k++) + { + // Fast envelope follower + // decay time = -131000 / -1000 = 131 (ms) + stt->capacitorFast = AGC_SCALEDIFF32(-1000, stt->capacitorFast, stt->capacitorFast); + if (env[k] > stt->capacitorFast) + { + stt->capacitorFast = env[k]; + } + // Slow envelope follower + if (env[k] > stt->capacitorSlow) + { + // increase capacitorSlow + stt->capacitorSlow + = AGC_SCALEDIFF32(500, (env[k] - stt->capacitorSlow), stt->capacitorSlow); + } else + { + // decrease capacitorSlow + stt->capacitorSlow + = AGC_SCALEDIFF32(decay, stt->capacitorSlow, stt->capacitorSlow); + } + + // use maximum of both capacitors as current level + if (stt->capacitorFast > stt->capacitorSlow) + { + cur_level = stt->capacitorFast; + } else + { + cur_level = stt->capacitorSlow; + } + // Translate signal level into gain, using a piecewise linear approximation + // find number of leading zeros + zeros = WebRtcSpl_NormU32((WebRtc_UWord32)cur_level); + if (cur_level == 0) + { + zeros = 31; + } + tmp32 = (WEBRTC_SPL_LSHIFT_W32(cur_level, zeros) & 0x7FFFFFFF); + frac = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(tmp32, 19); // Q12 + tmp32 = WEBRTC_SPL_MUL((stt->gainTable[zeros-1] - stt->gainTable[zeros]), frac); + gains[k + 1] = stt->gainTable[zeros] + WEBRTC_SPL_RSHIFT_W32(tmp32, 12); +#ifdef AGC_DEBUG + if (k == 0) + { + fprintf(stt->logFile, "%d\t%d\t%d\t%d\t%d\n", env[0], cur_level, stt->capacitorFast, stt->capacitorSlow, zeros); + } +#endif + } + + // Gate processing (lower gain during absence of speech) + zeros = WEBRTC_SPL_LSHIFT_W16(zeros, 9) - WEBRTC_SPL_RSHIFT_W16(frac, 3); + // find number of leading zeros + zeros_fast = WebRtcSpl_NormU32((WebRtc_UWord32)stt->capacitorFast); + if (stt->capacitorFast == 0) + { + zeros_fast = 31; + } + tmp32 = (WEBRTC_SPL_LSHIFT_W32(stt->capacitorFast, zeros_fast) & 0x7FFFFFFF); + zeros_fast = WEBRTC_SPL_LSHIFT_W16(zeros_fast, 9); + zeros_fast -= (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(tmp32, 22); + + gate = 1000 + zeros_fast - zeros - stt->vadNearend.stdShortTerm; + + if (gate < 0) + { + stt->gatePrevious = 0; + } else + { + tmp32 = WEBRTC_SPL_MUL_16_16(stt->gatePrevious, 7); + gate = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32((WebRtc_Word32)gate + tmp32, 3); + stt->gatePrevious = gate; + } + // gate < 0 -> no gate + // gate > 2500 -> max gate + if (gate > 0) + { + if (gate < 2500) + { + gain_adj = WEBRTC_SPL_RSHIFT_W16(2500 - gate, 5); + } else + { + gain_adj = 0; + } + for (k = 0; k < 10; k++) + { + if ((gains[k + 1] - stt->gainTable[0]) > 8388608) + { + // To prevent wraparound + tmp32 = WEBRTC_SPL_RSHIFT_W32((gains[k+1] - stt->gainTable[0]), 8); + tmp32 = WEBRTC_SPL_MUL(tmp32, (178 + gain_adj)); + } else + { + tmp32 = WEBRTC_SPL_MUL((gains[k+1] - stt->gainTable[0]), (178 + gain_adj)); + tmp32 = WEBRTC_SPL_RSHIFT_W32(tmp32, 8); + } + gains[k + 1] = stt->gainTable[0] + tmp32; + } + } + + // Limit gain to avoid overload distortion + for (k = 0; k < 10; k++) + { + // To prevent wrap around + zeros = 10; + if (gains[k + 1] > 47453132) + { + zeros = 16 - WebRtcSpl_NormW32(gains[k + 1]); + } + gain32 = WEBRTC_SPL_RSHIFT_W32(gains[k+1], zeros) + 1; + gain32 = WEBRTC_SPL_MUL(gain32, gain32); + // check for overflow + while (AGC_MUL32(WEBRTC_SPL_RSHIFT_W32(env[k], 12) + 1, gain32) + > WEBRTC_SPL_SHIFT_W32((WebRtc_Word32)32767, 2 * (1 - zeros + 10))) + { + // multiply by 253/256 ==> -0.1 dB + if (gains[k + 1] > 8388607) + { + // Prevent wrap around + gains[k + 1] = WEBRTC_SPL_MUL(WEBRTC_SPL_RSHIFT_W32(gains[k+1], 8), 253); + } else + { + gains[k + 1] = WEBRTC_SPL_RSHIFT_W32(WEBRTC_SPL_MUL(gains[k+1], 253), 8); + } + gain32 = WEBRTC_SPL_RSHIFT_W32(gains[k+1], zeros) + 1; + gain32 = WEBRTC_SPL_MUL(gain32, gain32); + } + } + // gain reductions should be done 1 ms earlier than gain increases + for (k = 1; k < 10; k++) + { + if (gains[k] > gains[k + 1]) + { + gains[k] = gains[k + 1]; + } + } + // save start gain for next frame + stt->gain = gains[10]; + + // Apply gain + // handle first sub frame separately + delta = WEBRTC_SPL_LSHIFT_W32(gains[1] - gains[0], (4 - L2)); + gain32 = WEBRTC_SPL_LSHIFT_W32(gains[0], 4); + // iterate over samples + for (n = 0; n < L; n++) + { + // For lower band + tmp32 = WEBRTC_SPL_MUL((WebRtc_Word32)out[n], WEBRTC_SPL_RSHIFT_W32(gain32 + 127, 7)); + out_tmp = WEBRTC_SPL_RSHIFT_W32(tmp32 , 16); + if (out_tmp > 4095) + { + out[n] = (WebRtc_Word16)32767; + } else if (out_tmp < -4096) + { + out[n] = (WebRtc_Word16)-32768; + } else + { + tmp32 = WEBRTC_SPL_MUL((WebRtc_Word32)out[n], WEBRTC_SPL_RSHIFT_W32(gain32, 4)); + out[n] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(tmp32 , 16); + } + // For higher band + if (FS == 32000) + { + tmp32 = WEBRTC_SPL_MUL((WebRtc_Word32)out_H[n], + WEBRTC_SPL_RSHIFT_W32(gain32 + 127, 7)); + out_tmp = WEBRTC_SPL_RSHIFT_W32(tmp32 , 16); + if (out_tmp > 4095) + { + out_H[n] = (WebRtc_Word16)32767; + } else if (out_tmp < -4096) + { + out_H[n] = (WebRtc_Word16)-32768; + } else + { + tmp32 = WEBRTC_SPL_MUL((WebRtc_Word32)out_H[n], + WEBRTC_SPL_RSHIFT_W32(gain32, 4)); + out_H[n] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(tmp32 , 16); + } + } + // + + gain32 += delta; + } + // iterate over subframes + for (k = 1; k < 10; k++) + { + delta = WEBRTC_SPL_LSHIFT_W32(gains[k+1] - gains[k], (4 - L2)); + gain32 = WEBRTC_SPL_LSHIFT_W32(gains[k], 4); + // iterate over samples + for (n = 0; n < L; n++) + { + // For lower band + tmp32 = WEBRTC_SPL_MUL((WebRtc_Word32)out[k * L + n], + WEBRTC_SPL_RSHIFT_W32(gain32, 4)); + out[k * L + n] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(tmp32 , 16); + // For higher band + if (FS == 32000) + { + tmp32 = WEBRTC_SPL_MUL((WebRtc_Word32)out_H[k * L + n], + WEBRTC_SPL_RSHIFT_W32(gain32, 4)); + out_H[k * L + n] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(tmp32 , 16); + } + gain32 += delta; + } + } + + return 0; +} + +void WebRtcAgc_InitVad(AgcVad_t *state) +{ + WebRtc_Word16 k; + + state->HPstate = 0; // state of high pass filter + state->logRatio = 0; // log( P(active) / P(inactive) ) + // average input level (Q10) + state->meanLongTerm = WEBRTC_SPL_LSHIFT_W16(15, 10); + + // variance of input level (Q8) + state->varianceLongTerm = WEBRTC_SPL_LSHIFT_W32(500, 8); + + state->stdLongTerm = 0; // standard deviation of input level in dB + // short-term average input level (Q10) + state->meanShortTerm = WEBRTC_SPL_LSHIFT_W16(15, 10); + + // short-term variance of input level (Q8) + state->varianceShortTerm = WEBRTC_SPL_LSHIFT_W32(500, 8); + + state->stdShortTerm = 0; // short-term standard deviation of input level in dB + state->counter = 3; // counts updates + for (k = 0; k < 8; k++) + { + // downsampling filter + state->downState[k] = 0; + } +} + +WebRtc_Word16 WebRtcAgc_ProcessVad(AgcVad_t *state, // (i) VAD state + const WebRtc_Word16 *in, // (i) Speech signal + WebRtc_Word16 nrSamples) // (i) number of samples +{ + WebRtc_Word32 out, nrg, tmp32, tmp32b; + WebRtc_UWord16 tmpU16; + WebRtc_Word16 k, subfr, tmp16; + WebRtc_Word16 buf1[8]; + WebRtc_Word16 buf2[4]; + WebRtc_Word16 HPstate; + WebRtc_Word16 zeros, dB; + WebRtc_Word16 *buf1_ptr; + + // process in 10 sub frames of 1 ms (to save on memory) + nrg = 0; + buf1_ptr = &buf1[0]; + HPstate = state->HPstate; + for (subfr = 0; subfr < 10; subfr++) + { + // downsample to 4 kHz + if (nrSamples == 160) + { + for (k = 0; k < 8; k++) + { + tmp32 = (WebRtc_Word32)in[2 * k] + (WebRtc_Word32)in[2 * k + 1]; + tmp32 = WEBRTC_SPL_RSHIFT_W32(tmp32, 1); + buf1[k] = (WebRtc_Word16)tmp32; + } + in += 16; + + WebRtcSpl_DownsampleBy2(buf1, 8, buf2, state->downState); + } else + { + WebRtcSpl_DownsampleBy2(in, 8, buf2, state->downState); + in += 8; + } + + // high pass filter and compute energy + for (k = 0; k < 4; k++) + { + out = buf2[k] + HPstate; + tmp32 = WEBRTC_SPL_MUL(600, out); + HPstate = (WebRtc_Word16)(WEBRTC_SPL_RSHIFT_W32(tmp32, 10) - buf2[k]); + tmp32 = WEBRTC_SPL_MUL(out, out); + nrg += WEBRTC_SPL_RSHIFT_W32(tmp32, 6); + } + } + state->HPstate = HPstate; + + // find number of leading zeros + if (!(0xFFFF0000 & nrg)) + { + zeros = 16; + } else + { + zeros = 0; + } + if (!(0xFF000000 & (nrg << zeros))) + { + zeros += 8; + } + if (!(0xF0000000 & (nrg << zeros))) + { + zeros += 4; + } + if (!(0xC0000000 & (nrg << zeros))) + { + zeros += 2; + } + if (!(0x80000000 & (nrg << zeros))) + { + zeros += 1; + } + + // energy level (range {-32..30}) (Q10) + dB = WEBRTC_SPL_LSHIFT_W16(15 - zeros, 11); + + // Update statistics + + if (state->counter < kAvgDecayTime) + { + // decay time = AvgDecTime * 10 ms + state->counter++; + } + + // update short-term estimate of mean energy level (Q10) + tmp32 = (WEBRTC_SPL_MUL_16_16(state->meanShortTerm, 15) + (WebRtc_Word32)dB); + state->meanShortTerm = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(tmp32, 4); + + // update short-term estimate of variance in energy level (Q8) + tmp32 = WEBRTC_SPL_RSHIFT_W32(WEBRTC_SPL_MUL_16_16(dB, dB), 12); + tmp32 += WEBRTC_SPL_MUL(state->varianceShortTerm, 15); + state->varianceShortTerm = WEBRTC_SPL_RSHIFT_W32(tmp32, 4); + + // update short-term estimate of standard deviation in energy level (Q10) + tmp32 = WEBRTC_SPL_MUL_16_16(state->meanShortTerm, state->meanShortTerm); + tmp32 = WEBRTC_SPL_LSHIFT_W32(state->varianceShortTerm, 12) - tmp32; + state->stdShortTerm = (WebRtc_Word16)WebRtcSpl_Sqrt(tmp32); + + // update long-term estimate of mean energy level (Q10) + tmp32 = WEBRTC_SPL_MUL_16_16(state->meanLongTerm, state->counter) + (WebRtc_Word32)dB; + state->meanLongTerm = WebRtcSpl_DivW32W16ResW16(tmp32, + WEBRTC_SPL_ADD_SAT_W16(state->counter, 1)); + + // update long-term estimate of variance in energy level (Q8) + tmp32 = WEBRTC_SPL_RSHIFT_W32(WEBRTC_SPL_MUL_16_16(dB, dB), 12); + tmp32 += WEBRTC_SPL_MUL(state->varianceLongTerm, state->counter); + state->varianceLongTerm = WebRtcSpl_DivW32W16(tmp32, + WEBRTC_SPL_ADD_SAT_W16(state->counter, 1)); + + // update long-term estimate of standard deviation in energy level (Q10) + tmp32 = WEBRTC_SPL_MUL_16_16(state->meanLongTerm, state->meanLongTerm); + tmp32 = WEBRTC_SPL_LSHIFT_W32(state->varianceLongTerm, 12) - tmp32; + state->stdLongTerm = (WebRtc_Word16)WebRtcSpl_Sqrt(tmp32); + + // update voice activity measure (Q10) + tmp16 = WEBRTC_SPL_LSHIFT_W16(3, 12); + tmp32 = WEBRTC_SPL_MUL_16_16(tmp16, (dB - state->meanLongTerm)); + tmp32 = WebRtcSpl_DivW32W16(tmp32, state->stdLongTerm); + tmpU16 = WEBRTC_SPL_LSHIFT_U16((WebRtc_UWord16)13, 12); + tmp32b = WEBRTC_SPL_MUL_16_U16(state->logRatio, tmpU16); + tmp32 += WEBRTC_SPL_RSHIFT_W32(tmp32b, 10); + + state->logRatio = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(tmp32, 6); + + // limit + if (state->logRatio > 2048) + { + state->logRatio = 2048; + } + if (state->logRatio < -2048) + { + state->logRatio = -2048; + } + + return state->logRatio; // Q10 +} diff --git a/src/modules/audio_processing/agc/main/source/digital_agc.h b/src/modules/audio_processing/agc/main/source/digital_agc.h new file mode 100644 index 0000000000..240b220661 --- /dev/null +++ b/src/modules/audio_processing/agc/main/source/digital_agc.h @@ -0,0 +1,76 @@ +/* + * Copyright (c) 2011 The WebRTC project authors. All Rights Reserved. + * + * Use of this source code is governed by a BSD-style license + * that can be found in the LICENSE file in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AGC_MAIN_SOURCE_DIGITAL_AGC_H_ +#define WEBRTC_MODULES_AUDIO_PROCESSING_AGC_MAIN_SOURCE_DIGITAL_AGC_H_ + +#ifdef AGC_DEBUG +#include <stdio.h> +#endif +#include "typedefs.h" +#include "signal_processing_library.h" + +// the 32 most significant bits of A(19) * B(26) >> 13 +#define AGC_MUL32(A, B) (((B)>>13)*(A) + ( ((0x00001FFF & (B))*(A)) >> 13 )) +// C + the 32 most significant bits of A * B +#define AGC_SCALEDIFF32(A, B, C) ((C) + ((B)>>16)*(A) + ( ((0x0000FFFF & (B))*(A)) >> 16 )) + +typedef struct +{ + WebRtc_Word32 downState[8]; + WebRtc_Word16 HPstate; + WebRtc_Word16 counter; + WebRtc_Word16 logRatio; // log( P(active) / P(inactive) ) (Q10) + WebRtc_Word16 meanLongTerm; // Q10 + WebRtc_Word32 varianceLongTerm; // Q8 + WebRtc_Word16 stdLongTerm; // Q10 + WebRtc_Word16 meanShortTerm; // Q10 + WebRtc_Word32 varianceShortTerm; // Q8 + WebRtc_Word16 stdShortTerm; // Q10 +} AgcVad_t; // total = 54 bytes + +typedef struct +{ + WebRtc_Word32 capacitorSlow; + WebRtc_Word32 capacitorFast; + WebRtc_Word32 gain; + WebRtc_Word32 gainTable[32]; + WebRtc_Word16 gatePrevious; + WebRtc_Word16 agcMode; + AgcVad_t vadNearend; + AgcVad_t vadFarend; +#ifdef AGC_DEBUG + FILE* logFile; + int frameCounter; +#endif +} DigitalAgc_t; + +WebRtc_Word32 WebRtcAgc_InitDigital(DigitalAgc_t *digitalAgcInst, WebRtc_Word16 agcMode); + +WebRtc_Word32 WebRtcAgc_ProcessDigital(DigitalAgc_t *digitalAgcInst, const WebRtc_Word16 *inNear, + const WebRtc_Word16 *inNear_H, WebRtc_Word16 *out, + WebRtc_Word16 *out_H, WebRtc_UWord32 FS, + WebRtc_Word16 lowLevelSignal); + +WebRtc_Word32 WebRtcAgc_AddFarendToDigital(DigitalAgc_t *digitalAgcInst, const WebRtc_Word16 *inFar, + WebRtc_Word16 nrSamples); + +void WebRtcAgc_InitVad(AgcVad_t *vadInst); + +WebRtc_Word16 WebRtcAgc_ProcessVad(AgcVad_t *vadInst, // (i) VAD state + const WebRtc_Word16 *in, // (i) Speech signal + WebRtc_Word16 nrSamples); // (i) number of samples + +WebRtc_Word32 WebRtcAgc_CalculateGainTable(WebRtc_Word32 *gainTable, // Q16 + WebRtc_Word16 compressionGaindB, // Q0 (in dB) + WebRtc_Word16 targetLevelDbfs,// Q0 (in dB) + WebRtc_UWord8 limiterEnable, WebRtc_Word16 analogTarget); + +#endif // WEBRTC_MODULES_AUDIO_PROCESSING_AGC_MAIN_SOURCE_ANALOG_AGC_H_ |