Lectrosonics SPNCWB Technical Data Manual page 4

Advanced Acoustic Echo
Cancellation
Conventional AEC algorithms face a trade-off between
convergence rate and depth. A fast convergence time
adapts quickly when a new conversation begins or when
a change occurs in the acoustic space, but the cancella-
tion depth is limited. Deeper cancellation requires more
time, so an echo may be heard at the far end until the
AEC achieves a fairly deep convergence.
30.0
24.0
Far side audio (blue)
18.0
12.0
6.0
0.0
0.0 2.5 5.0
Mostly far
side audio
Example of AEC activity over a 30 second period.
Local audio (tan)
7.5 10.0 12.5 15.0 17.5
time [s]
Path
change
Double talk
An ideal AEC would react very quickly in the beginning
and then start applying more calculations over longer
time intervals to achieve a deeper cancellation as the
conference progresses.
The ideal echo canceller would also maintain con-
vergence regardless of signal types or levels. This is
precisely what the ASPEN echo canceller does. It is
designed to handle multi-site bridging and any number
of microphones simultaneously, and it works with the
gain proportional mixing algorithm perfectly.
AEC cancellation
depth (dB)
20.0 22.5 25.0 27.5
Mostly far
side audio
An extremely difficult environ-
ment for an AEC is when local
sound reinforcement is being
used. Far side audio enters the
local microphones and mixes
with local audio and noise,
which makes it extremely diffi-
cult for the AEC to identify and
cancel the echo.
The example shown here is
a 30 second recording of a
conference with local and far
side audio activity, plus a local
sound reinforcement system.
0-1.5 seconds:
2-9.5 seconds:
At 10 seconds:
The AEC will not diverge (lose
convergence) unless some-
30.0
thing changes in the local
acoustic environment, such as
a microphone moving. When
this happens, it will converge
again and adapt to the new
echo path. These are usually
very subtle changes and go
completely unnoticed by the
conference participants.