User Manual

Acoustic Echo Canceller White Paper
processing device with high computational capabilities. In
recent years, the advancement of signal processing
technology and DSPs (digital signal processors) has made
it possible to create sophisticated echo cancellers.
2. Adaptive Digital Filter
An ADF is a digital filter that can learn. ADFs analyze
input and output signals in real time and identify unknown
transfer systems
[3].
[]
nx
[]
ny
~
[]
ne
[] kg []
kh
[
]
ny
+
−
[
]
nd
Fig.2 Acoustic Echo Canceller Using ADF
An acoustic echo canceller that uses an ADF is configured
as in
Fig.2. The ADF is used to simulate the transfer
system from the
loudspeaker to the microphone,
[
] kg
,
using a filter, . To remove acoustic echo
[] kh
[
] ny
, the
echo canceller obtains pseudo echo signal
[
] ny
~
by
sending the input signal from the far end, , to filter
and subtracting the pseudo echo signal from the
microphone signal. The signal with the acoustic echo
removed from it, , is not just sent to the far end; the
ADF also learns from it. The ADF learns how to minimize
the errors, , in its filtering. The algorithms that are
used for this learning are referred to as adaptive algorithms.
There are many different types of these algorithms. Factors
such as the estimation accuracy, convergence rate, and
amount of computation vary depending on the algorithm
that is used. One of the best known adaptive algorithms is
the NLMS (normalized least mean square) algorithm. It
updates the filter in accordance with the following formula.
[] nx
[] kh
[]
ne
[]
ne
[3]
[] [] [][ neknxkhkh
x
nn
⋅−⋅−=
+
2
1
σ
3. Implementation
In an actual teleconference, the microphone signal doesn't
just include acoustic echo; it also includes external noise
[
] nd
, which includes background noise and the sound of
the near-end speaker. External noise interferes with the
learning of the filter and reduces the filter's estimation
accuracy. Also, time variation and nonlinearity in the
transfer system,
[
] kg
, also reduce the estimation
accuracy and result in residual echo. To supplement the
functionality of the ADF, an echo suppressor for
suppressing residual echo is connected in the later stage of
the ADF. (
Fig.3)
Fig.3 Acoustic Echo Canceller
The acoustic echo canceller shown in
Fig.3 is implemented
in a Yamaha DSP that is specially designed for audio signal
processing, the YSS950 DAP1 (32-bit floating-point;
shown in
Fig.4). With a sampling rate of 48 kHz, this DSP
makes it possible to obtain acoustic echo cancelling with
both high quality sound and high precision.
Fig.4 Yamaha YSS950 DAP1
]
μ
2