
specific tones. If those tones do not stimulate the system’s nonlinearity then the distortion cannot be measured properly. Stenger et al. [15] have proposed measuring the impulse response of a LEM with cheap analog equipment in the presence of nonlinear distortion and background noise by modeling nonlinear distortion as noise and then estimating the linear part using the Wiener solution. There is an absence of any clear definitive metric that defines the nonlinear distortion.
Non-Coherent Distortion (NCD) is a nonlinear distortion measure that provides a continuous frequency domain distortion curve for multi-tone signals such as speech or music [16]. A single
nonlinear distortion measure called Total Non-Coherent Distortion (TNCD) is obtained by power summing the NCD curve across the
Figure 1. Nonlinear echo path model. frequency band from DC up to the Nyquist frequency. It is the
ratio of the uncorrelated noise plus the nonlinear distortion power through creation of harmonics and inter-modulation products by to the total output power. TNCD is calculated as the soft clipping of large amplitude values. Enclosure vibration due to mechanical coupling of the loudspeaker components with the GNN(Z)¦case of the device especially at the lower voice frequencies causes (1) Z
O significant nonlinear distortion to be picked up by the microphone. Gyy()Finally the loudspeaker itself is a major source of nonlinear Z
distortion. The nonlinearities in the loudspeaker could be due to where acoustic, electromagnetic, and mechanical reasons [12]. If we 2
ªº
«1 GXY(jZ)could capture the signal present immediately after the section » GYY(Z) GNN(Z) (2)
shown in Figure 1 by a dotted box, then we should be able to GXX()GYY()»«¬¼capture most of the nonlinear distortion and the acoustic path could
then be estimated using existing AEC techniques.
with GXX( ) and GYY( ) being the real and positive autospectra of
the input and output signals respectively. GXY(j ) is the complex
One way is to capture the loudspeaker acoustic signal by
input-output cross spectrum. Figure 2 shows the TNCD against
placing a microphone just next to the loudspeaker. But this method
different sound pressure levels (SPL) for a small loudspeaker. It is
is not very useful due to it also sensing the near-end speech and the
observed that the TNCD increases very sharply as the loudspeaker
background noise. An optical sensor can be used to measure the
volume is increased.
loudspeaker’s cone displacement [13]. Experiments were
performed using a laser position sensitive detector (PSD) sensor to

capture the motion of cone of the loudspeaker and to obtain the echo source signal. However, this method requires sophisticated hardware that drives up the cost of real-time device.
An accelerometer is an electromechanical device that measures acceleration. It can be used to measure vibrations precisely.
Experiments with an accelerometer mounted on the frame, the cone, and the magnet of a loudspeaker were performed.
Unfortunately, the frequency response of the loudspeaker changes
when an accelerometer is mounted on the cone due to the low mass of the cone of a small loudspeaker. An accelerometer mounted on the loudspeaker frame or magnet gives similar signals, and mounting an accelerometer at these positions does not affect the loudspeaker’s frequency response. An accelerometer could also be placed on the device enclosure to sense the vibrations.
Nonlinear distortion depends highly on the stimulus signal’s
time-domain amplitude and spectral content. Total harmonic Sound Pressure Level (dB) distortion (THD) measurement quantifies harmonic distortion due
to single tones, but it cannot quantify intermodulation distortion (IMD) in a multitone signal caused by the nonlinear behavior of a Figure 2. TNCD vs. Sound pressure level on an arbitrary scale. system. To keep distortion metrics meaningful, efficient measures that can quantify nonlinear distortion due to multi-tone signals, 3. PROPOSED AEC ARCHITECTURE such as speech, are required.
ITU-T ecommendation O.42 [14] measures nonlinear Figure 3 shows the proposed configuration for nonlinear acoustic distortion using a 4-tone inter-modulation method, but it is echo cancellation where x(n) denotes the far end received signal designed for telephony network equipment and uses only four that is fed to the loudspeaker.
Total NonCoherent Distortion(%)
Suppression – up to 12dB Superior acoustic echo cancellation performance with linear and patented proprietary non-linear echo filtering Features and ...
- Big echo cancellation on/off flag, 1: on, 0: off. - Reserved. - Non-linear AEC post-filter on/off flag, 1: on, 0: off. - Linear AEC ...
Non-Linear Echo Cancellation - SAM technology may cancel non-linear echo resulting from - imperfect audio design and speaker - overdriven speaker - ...
non-linear Use non-linear processing during echo cancellation operation Specify the operation of the E&M signal output Configure output attenuation for voice...
such as resize/rotate * Linear and non-linear filtering operation on an ...recursion, Linear Prediction Analysis & Quantization, Echo Cancellation, ...
Non Linear transform:非线性变换开关。在 loudspeaker 模式下使用,用于应对 扬声...在 echo cancellation 处理的最后会加入舒 适噪声来获得较好的听觉效果同时掩蔽...
and has sig nificantly impr oved results of non linear echo suppression. Keywords: acoust ic echo cancellation; no ise cancellation; voice activ e ...
(应用增益 18.0 分贝) ; 2、Acoustic Echo cancellation Settings(声学回声消除...(主增益) ; Linear (线性) PEQ Presets(预设) Show All(显示全部) Graphic ...
and has sig nificantly impr oved results of non linear echo suppression. Keywords: acoust ic echo cancellation; no ise cancellation; voice activ e ...
Acoustic Echo Cancellation – Up to 60dB by single microphone and non-linear echo cancellation ? Noise Suppression – Stationary Noise up to 12 dB and ...
热门文档