Line data    Source code

       1             : /******************************************************************************************************
       2             : 
       3             :    (C) 2022-2025 IVAS codec Public Collaboration with portions copyright Dolby International AB, Ericsson AB,
       4             :    Fraunhofer-Gesellschaft zur Foerderung der angewandten Forschung e.V., Huawei Technologies Co. LTD.,
       5             :    Koninklijke Philips N.V., Nippon Telegraph and Telephone Corporation, Nokia Technologies Oy, Orange,
       6             :    Panasonic Holdings Corporation, Qualcomm Technologies, Inc., VoiceAge Corporation, and other
       7             :    contributors to this repository. All Rights Reserved.
       8             : 
       9             :    This software is protected by copyright law and by international treaties.
      10             :    The IVAS codec Public Collaboration consisting of Dolby International AB, Ericsson AB,
      11             :    Fraunhofer-Gesellschaft zur Foerderung der angewandten Forschung e.V., Huawei Technologies Co. LTD.,
      12             :    Koninklijke Philips N.V., Nippon Telegraph and Telephone Corporation, Nokia Technologies Oy, Orange,
      13             :    Panasonic Holdings Corporation, Qualcomm Technologies, Inc., VoiceAge Corporation, and other
      14             :    contributors to this repository retain full ownership rights in their respective contributions in
      15             :    the software. This notice grants no license of any kind, including but not limited to patent
      16             :    license, nor is any license granted by implication, estoppel or otherwise.
      17             : 
      18             :    Contributors are required to enter into the IVAS codec Public Collaboration agreement before making
      19             :    contributions.
      20             : 
      21             :    This software is provided "AS IS", without any express or implied warranties. The software is in the
      22             :    development stage. It is intended exclusively for experts who have experience with such software and
      23             :    solely for the purpose of inspection. All implied warranties of non-infringement, merchantability
      24             :    and fitness for a particular purpose are hereby disclaimed and excluded.
      25             : 
      26             :    Any dispute, controversy or claim arising under or in relation to providing this software shall be
      27             :    submitted to and settled by the final, binding jurisdiction of the courts of Munich, Germany in
      28             :    accordance with the laws of the Federal Republic of Germany excluding its conflict of law rules and
      29             :    the United Nations Convention on Contracts on the International Sales of Goods.
      30             : 
      31             : *******************************************************************************************************/
      32             : 
      33             : /*====================================================================================
      34             :     EVS Codec 3GPP TS26.443 Nov 04, 2021. Version 12.14.0 / 13.10.0 / 14.6.0 / 15.4.0 / 16.3.0
      35             :   ====================================================================================*/
      36             : 
      37             : #include <stdint.h>
      38             : #include "options.h"
      39             : #ifdef DEBUGGING
      40             : #include "debug.h"
      41             : #endif
      42             : #include "cnst.h"
      43             : #include "prot.h"
      44             : #include "rom_com.h"
      45             : #include "wmc_auto.h"
      46             : 
      47             : /*------------------------------------------------------------------*
      48             :  * Local constants
      49             :  *------------------------------------------------------------------*/
      50             : 
      51             : #define RATEWIN 600 /* length of the rate control window. This is 600 active speech frames. This equals roughly 12s of active speech */
      52             : 
      53             : /*------------------------------------------------------------------*
      54             :  * update_average_rate()
      55             :  *
      56             :  * SC-VBR update average data rate
      57             :  *------------------------------------------------------------------*/
      58             : 
      59        1403 : void update_average_rate(
      60             :     SC_VBR_ENC_HANDLE hSC_VBR, /* i/o: SC-VBR state structure  */
      61             :     const int32_t core_brate   /* i  : core bitrate            */
      62             : )
      63             : {
      64             :     float avratetarg; /* target rate for next RATEWIN active frames */
      65             :     float target;     /* target set by VBR_ADR_MAX_TARGET*RATEWIN*10 */
      66             : 
      67        1403 :     if ( hSC_VBR->numactive == RATEWIN ) /* goes into rate control only the numactive ==RATEWIN. So rate control is triggered after each RATEWIN avtive frames */
      68             :     {
      69             :         /* after 1000 blocks of RATEWIN frames, we change the way we control the average rate by using
      70             :            st->global_avr_rate=0.99*st->global_avr_rate+0.01*st->sum_of_rates. This will avoid
      71             :            veriables growing indefinitely while providing a good long term average rate */
      72             : 
      73           2 :         if ( hSC_VBR->global_frame_cnt < 1000 )
      74             :         {
      75           2 :             hSC_VBR->global_frame_cnt++;
      76           2 :             hSC_VBR->global_avr_rate = ( hSC_VBR->global_avr_rate * ( hSC_VBR->global_frame_cnt - 1 ) + hSC_VBR->sum_of_rates ) / hSC_VBR->global_frame_cnt;
      77             :         }
      78             :         else
      79             :         {
      80           0 :             hSC_VBR->global_avr_rate = 0.01f * hSC_VBR->sum_of_rates + 0.99f * hSC_VBR->global_avr_rate;
      81             :         }
      82             : 
      83           2 :         if ( hSC_VBR->sum_of_rates == 0 )
      84             :         {
      85           0 :             hSC_VBR->sum_of_rates = (float) ( RATEWIN * VBR_ADR_MAX_TARGET * 10 );
      86             :         }
      87             : 
      88           2 :         target = VBR_ADR_MAX_TARGET * 10 * RATEWIN;
      89             : 
      90           2 :         if ( target < hSC_VBR->global_avr_rate ) /* Action is taken to reduce the averge rate. Only initiated if the global rate > target rate */
      91             :         {
      92             :             /* Check the VAD SNR values to table the noisey/not noisey decision */
      93             : 
      94           2 :             if ( hSC_VBR->SNR_THLD < 67 ) /* Currently in QFF mode. The bumpup thresholds are slightly relaxed for noisy speech. */
      95             :             {
      96             :                 /*  Increase the threshold so the the bumpup procedure is done using the noisy thresholds.
      97             :                     Use 3.5 steps to quickly ramp up the rate control to reduce the settling time */
      98           0 :                 hSC_VBR->SNR_THLD += 3.5f;
      99             :             }
     100           2 :             else if ( hSC_VBR->mode_QQF == 0 && hSC_VBR->sum_of_rates > target ) /* Now SNR_THLD is in the max allowed. Sill the global average is higher and
     101             :                                                                           last RATEWIN frames have a higher agerage than the target rate. Now slightly
     102             :                                                                           more aggresive rate control is used by changing the mode to QQF. Still the
     103             :                                                                           same strict bumpups (more bumpups,higher rate) are used. */
     104             :             {
     105             :                 /* Kick in QQF mode */
     106           0 :                 hSC_VBR->mode_QQF = 1;
     107             :             }
     108           2 :             else if ( hSC_VBR->sum_of_rates > target ) /* Actions (1) and (2) are not sufficient to control the rate. Still the last RATEWIN active
     109             :                                                       frames have a higher average rate than the target rate. More aggresive rate control is
     110             :                                                       needed. At this point the rate_control flag is set. This will enable the more relaxed
     111             :                                                       bump up thresholds (less bump ups->reduced rate)*/
     112             :             {
     113             :                 /* Relaxed bump ups are used */
     114           2 :                 hSC_VBR->rate_control = 1;
     115             : 
     116             :                 /* This will be triggered only if the gloabl average rate is considerablly higher than the target rate.
     117             :                    Keep a higher threshold to avoid short term rate increases over the target rate. */
     118           2 :                 if ( hSC_VBR->global_avr_rate > ( target + 420.0f ) ) /* Last resort rate control. This is a safer rate control mechanism by increasing NELPS */
     119             :                 {
     120           2 :                     hSC_VBR->Last_Resort = 1; /* compute based on a larger window as the last resort */
     121             :                 }
     122             :                 else
     123             :                 {
     124           0 :                     hSC_VBR->Last_Resort = 0;
     125             :                 }
     126             :             }
     127           0 :             else if ( hSC_VBR->sum_of_rates < target ) /* If the average rate of last RATEWIN frames is controlled by above actions, disable the most
     128             :                                                      aggresive rate control mechanisms. Still keep QQF mode as the global rate is not under
     129             :                                                      the target rate*/
     130             :             {
     131           0 :                 hSC_VBR->Last_Resort = 0;
     132           0 :                 hSC_VBR->mode_QQF = 1;
     133           0 :                 hSC_VBR->rate_control = 0;
     134             :             }
     135             :         }
     136             :         else
     137             :         {
     138             :             /* floding back to lesser and leser aggresive rate control mechanisms gradually if global rate is under control */
     139           0 :             hSC_VBR->Last_Resort = 0;
     140             : 
     141           0 :             if ( hSC_VBR->rate_control == 1 )
     142             :             {
     143           0 :                 hSC_VBR->rate_control = 0;
     144             :             }
     145           0 :             else if ( hSC_VBR->mode_QQF == 1 ) /* now rate control is not active and still the global rate is below the target. so go to QFF mode */
     146             :             {
     147           0 :                 hSC_VBR->mode_QQF = 0;
     148             :             }
     149             :             else
     150             :             {
     151           0 :                 if ( hSC_VBR->SNR_THLD >= 60 )
     152             :                 {
     153           0 :                     hSC_VBR->SNR_THLD -= 1.5f;
     154             :                 }
     155             :                 else
     156             :                 {
     157           0 :                     hSC_VBR->SNR_THLD = 60.0f;
     158             :                 }
     159             :             }
     160             :         }
     161             : 
     162           2 :         if ( hSC_VBR->global_avr_rate < target - 120 ) /* In QFF mode and global rate is less than target rate-0.2kbps. We can send some Q frames
     163             :                                                     to F frames to improve the quality */
     164             :         {
     165             :             /* kick in bouncing back from Q to F */
     166           0 :             hSC_VBR->Q_to_F = 1;
     167             : 
     168             :             /* average rate for next 600ms = global_rate * 2 - rate of the past RATEWIN active frames */
     169           0 :             avratetarg = (float) ( ( RATEWIN * 10 ) * 2 * VBR_ADR_MAX_TARGET - hSC_VBR->global_avr_rate );
     170             : 
     171             :             /* compute the percentage of frames that needed to be sent to F. st->pattern_m is computed as % val * 1000. eg. if % is 10%, then
     172             :                st->pattern_m=100 . Later this value is used in voiced.enc to bump up 10% of PPP frames to F frames. */
     173           0 :             hSC_VBR->pattern_m = (int16_t) ( 1000 * ( avratetarg - 6.15f * RATEWIN * 10 ) / ( 10 * RATEWIN * 0.1f ) );
     174             : 
     175           0 :             if ( hSC_VBR->pattern_m < 0 )
     176             :             {
     177           0 :                 hSC_VBR->pattern_m = 0; /* no bump up will ever happen */
     178             :             }
     179             : 
     180           0 :             if ( hSC_VBR->pattern_m > 1000 )
     181             :             {
     182           0 :                 hSC_VBR->pattern_m = 1000; /* 10% of bump ups */
     183             :             }
     184             : 
     185           0 :             hSC_VBR->patterncount = 0;
     186             :         }
     187             :         else
     188             :         {
     189           2 :             hSC_VBR->Q_to_F = 0;
     190             :         }
     191             : 
     192           2 :         hSC_VBR->sum_of_rates = 0;
     193           2 :         hSC_VBR->numactive = 0;
     194             :     }
     195             : 
     196        1403 :     hSC_VBR->numactive++;
     197             : 
     198             :     /* sum the total number of bits (in kbytes) * 10 here */
     199        1403 :     hSC_VBR->sum_of_rates += ( core_brate / 1000.0f ) * 10;
     200             : 
     201        1403 :     return;
     202             : }