VCP回声调试参数说明

最新推荐文章于 2022-12-09 16:27:35 发布

elomio

最新推荐文章于 2022-12-09 16:27:35 发布

阅读量2.3k

点赞数 1

分类专栏： echo

本文为博主原创文章，未经博主允许不得转载!

本文链接：https://blog.csdn.net/elimio/article/details/82717362

版权

echo 专栏收录该内容

2 篇文章 0 订阅

订阅专栏

VCP回声调试参数说明

1. VCP介绍

Alango VCP可能大家知道的比较少，鉴于网站上公开资料较少，又不同于webrtc和speech，我就把我理解过的一些内容写出来了。不过，个人建议您去看webrtc！
本文参数是基于VCP工具翻译并整理。大家随意参考了。

2. AF模块

AF即自适应滤波（adaptive filter）。以下都以AF代替。
AF模块是回声消除的主要模块，主要用来消除线性回声。

AF和ES对应，而ES就是后面要介绍的残留回声消除了。

2.1 General AF setting

2.1.1 TX_AF_BFF

AF（自适应滤波频率）下限，即小于这个频率，都不会经过AF处理。

2.1.2 TX_AF_UDF

Upper duplex (filtering) frequency
全双工滤波频率上限。高于这个设定的频率，通通都是半双工处理（不会经过AF处理）。
以下为英文说明，可以不看：

Defines Upper Duplex Frequency (UDF). In frequency bands above the UDF frequency, VCP ES operates in half-duplex mode: any echo is closed above UDF). No adaptive filtering is performed in TF is the transfer function estimated based on AF coefficients in frequencies within the frequency rage: [UDF/2, UDF].

2.1.3 TX_AF_AFUF

1st adaptive filter upper frequency
第一级 AF自适应滤波上限频率，也就是第二级滤波频率下限了。
结合2.1和2.2,即
第一级AF频率范围：TX_AF_BFF -> TX_AF_AFUF
第二级AF频率范围：TX_AF_AFUF -> TX_AF_UDF
举例如下图：第一级为0-3000HZ，第二级为3000-4000HZ.
AF滤波频率分级（1，2两级）

2.1.4 TX_AF_AFATHR

Minimal signal level for AF adaptation
AF自适应的最小信号强度

Adaptation of AF is not performed if AEC Reference full-band signal level is lower than this threshold.
Adaptation is not performed in the sub-band if AEC Reference sub-band signal level is lower than: AFTHR - 20*log10(sqrt(bands)), where bands = {128 @ Fs=8 KHz | 256 @ Fs = 16 KHz}. Allows saving processing power and improve AF performance.

2.1.5 TX_AF_AFFTHR

Minimal signal level for AF filtering
2.4节强调adaptation（自适应），这节强调filtering（滤波）. 更详细的内容Alango未说明。
AF minimal signal threshold

2.2 AF length and adaptation

先给张图：
AF length and adaptation

2.2.1 TX_AF_ARF

Adaptation rate factor
自适应速度因子，值越大，收敛越快，但是稳定性越差。

The effective adaptation rate is scaled by the correlation between Reference and AF output signal (refer to AFCCA and FLCFV parameters).
Dependancies:
TX_AF_AFCCA

Adaptive filter correlation-controlled adaptation [flag]
TX_AF_FLCFV

Correlation function value limit [frac]

2.2.2 TX_AF_AFAH

Reference activity hold
正常情况下,自适应滤波器只在参考信号信号活动期间工作.使能这个标志后,活动时间会延长到AF滤波器相对应的AFL时间. 这个标志一般会改善滤波效果.

2.2.3 TX_AF_AFCCA

Adaptive filter correlation-controlled adaptation (勾选就对了!)
可以通过参考信号和AF输出之间的相关度来控制滤波器自适应.
高相关度代表次优滤波器性能(因声学或其他外部因素变化导致非收敛滤波器),低相关度表示良好的滤波性能. 相关性的计算是计算密集型任务。
关闭此标志, 可以节省MIPS, 代价是损失滤波质量和性能;
关闭此相关度计算,AF收敛速度是由ARF定义并由FLCFV来缩放;
打开此相关度计算,AF收敛速度是由ARF和计算的相关度(由FLCFV初始化);这提供更优化的滤波表现.

Filter adaptation can be controlled by degree of correlation between the Reference signal and adaptive filter output. High correlation indicates about suboptimal filtering performance (not converged filter due to change of acoustics or other external factors), while low correlation indicates about good filtering performance. Calculation of the correlation is a computationally intensive task. This flag allows enabling or disabling the computation of correlation to help saving MIPS at expense of some loss of filtering and/or quality performance. If calculation of correlation is switched off, adaptive filter convergence speed is defined by ARF and is scaled by FLCFV. If calculation of correlation is switched on, adaptive filter convergence speed is defined by ARF and degree of the computed correlation (limited by FLCFV), which provides more optimal filtering performance.

2.2.4 TX_AF_FLCFV

Correlation function value limit
如果AFCCA打开,这个参数设置了动态相关度计算最大限制值. 动态相关度用于缩放AF自适应滤波器的自适应速度ARF.
如果AFCCA关闭, 相关度就不会被计算, 这个参数定义了ARF因子的缩放常数.
####2.2.5 TX_AF_AFL1
Effective length of 1st adaptive filter
Defines effective length of the 1st adaptive filter operating in the 0 - AFUF (Hz) frequency region.
长滤波器可消除长的回声尾部, 但导致较慢的收敛速度并增加所需计算资源.

2.2.6 TX_AF_AFL2

Effective length of 2nd adaptive filter (above AFUF)
Defines effective length of the 2nd adaptive filter operating in the AFUF - UDF (Hz) frequency region.

2.2.7 TX_AF_FOST

Filter output sample type, off=min, on=last
定义自适应滤波器输出的采样类型.(TSF打开才有效). 0是选择两个过滤阶段的最小采样值, 1是两个阶段的最后一个值

Defines output sample type of adaptive filter when two-stage filtering (TSF) is enabled. 0 chooses minimal sample value of two filtering stages; 1 chooses last value after 2 stages.

3. ES模块

主要用来消除残留回声(包括非线性回声)

3.1 General Es settings

General Es settings

3.1.1 TX_ES_ESTHR

Minimal reference signal level that triggers ES
假设AEC参考信号幅度小于ESTHR, ES模块就会被设置为bypasss.
Note: Improper setting may spoil echo cancellation performance.

3.1.2 TX_ES_UDFTHR

Reference signal activity level threshold for bands above UDF
如果参考信号幅度小于UDFTHR threshold, 则不管回声活动检测情况如何,都不会执行回声抑制.

If the Reference signal level is lower than UDFTHR threshold, then the echo suppression above the UDF frequency is not performed irrespective(无关的,不考虑的) to the echo activity detection.

3.1.3 TX_ES_REFTHR

Reference signal activity level threshold
如果全频带AEC参考信号幅度低于REFTHR，则不应用ESOTHR和PSL机制。

3.1.4 TX_ES_UTFF

UDF Transfer Function Factor

Bands above UDF are getting closed in case [Ref.level * TransferFunction * UTFF >= Mic level]. UTFF defines a "safety margin". The greater the value of UTFF, the more “aggressive” the suppression above UDF.

3.2 Initial Convergence setting

Initial Convergence setting

3.2.1 TX_ES_HDTIME

Half-duplex ref. activity time after VCP init.
VCP启动后, 一个特殊的far-talk活动持续时间. 在这个时间内, VCP工作在半双工模式下. 这个特性使得VCP可以在自适应滤波器没有收敛前, 完全压缩初始的回声.

3.2.2 TX_ES_GICTV

Global initial convergence target value
定义了在回到全双工之前, 对应的各个频带内, 自适应滤波器必须完成的目标幅度衰减值总和.
在目标值完成之前, ES模块在整个频带内工作在半双工模式.

GICTV is cancelled automatically in the following conditions:
(i) 75% of frequency bands reached the target attenuation value defined by ES_LICTV .
(ii) full-band AF attenuation reached the ES_GICTV target.
(iii)full-band Reference activity (calculated as sub-band activity frame-counter x number of bands x frames per second x seconds) is registered for 7 seconds 0 dB setting cancels this mechanism.

3.2.3 TX_ES_LICTV

Local initial convergence target value
和GICTV类似,不过这个是工作在子频带内.
LICTV is cancelled automatically when ES_GICTV target is reached. 0 dB setting cancels this mechanism.

3.3 Comfort Noise substitution

Comfort Noise substitution
因为Comfort Noise较为简单,因此略过.

3.4 Band cancellation&Global cancellation

ES_cancellation

3.4.1 TX_ES_BCT

ES band cancellation threshold
定义了自适应滤波器输出和输入最小比值, 如果小于BCT, 子频带会被关闭.
逻辑是: 自适应滤波衰减幅度越大, 那就越可能是残留回声, 所以对应子频带越应该关闭. 所以, BCT值越大, 频带越可能会被关闭,对应回声就越小了.
举例说,如果BCT=-6, 并且AFout/AFin < 0.5, 子频带就会被关闭. 反之, 就会打开.
如果RX没有语音活动, BCT机制通常是无效的.

3.4.2 TX_ES_GCT

ES global cancellation threshold
和BCT类似,只是这个是在全带宽内工作.
举例说,如果整理带宽内的ESout/AFin> GCT, 那么整体带宽被关闭,这样基本上就是工作在半双工模式下了.

3.4.3 TX_ES_ESOTHR

Minimal ES output signal level allowed during RX activity
在RX活动期间,ES模块输出的最小电平. 如果ES输出低于该阈值,则全频带自动关闭(用舒适噪音代替). 这个机制可移除低值残留回声斑点. 并且这个机制是在BCT和GCT之后, 在ES的最后阶段.

3.5 Non-linear distortions

ES_NL
待补充

3.6 Auxiliary

Auxiliary

3.6.1 TX_ES_RRT

Room reverberation time (use only if adaptive filter is not long enough!)
房间混响时间（仅在自适应滤波器不够长时使用！）
非零值启用一种特殊机制，该机制在BCT计算中考虑房间混响时间。 RRT定义回波衰减到-60 dB所需的时间。通过人为地延长回声的“持续时间”，在BCT公式中考虑RRT。当回声tail较长时，并且自适应滤波器长度有限且由于MIPS /存储器约束而无法增加时，应使用RRT。
Notes: Affects BCT behavior. RRT block should not be used in high-quality hands-free communication systems.

3.6.2 TX_ES_PSL

TX input saturation level (to close ES bands during Far End activity)
TX过饱和会引入非线性失真.
PSL定义了一个TX input的阈值, 高于这个阈值被认为是过饱和, 如果这样的过饱和发生在RXout活动上, 则全频带TX output会被mute并被舒适噪声填充.
如果downlink没有活动, PSL是没有效果的. 如果设置为0db, PSL机制关闭.