Polycom C16 - Manual (Page 475)

Designing Audio Conferencing Systems
B - 13
AEC Reference
The AEC reference provides the AEC with the information of what signals it
should cancel from the room (see the following figure). The echo canceller
reference is usually a combination of audio from the remote sites including
telephone and video conferencing audio and also any program audio sources.
Microphone signals are typically not included in the acoustic echo canceller
reference signal.
In order to cancel echoes, the original source signal (the remote audio) must be
part of the echo canceller reference. Generally the reference is the original
signal that is played out of the loudspeaker system (as shown in the following
figure) before it generates echo in the local room.
If the reference signal is significantly different from signal played out of the
loudspeaker (due to dynamics processing, for instance), the echo canceller
may not be able to cancel the echo signal because it is not recognized as being
the same as the reference signal, causing a persistent residual echo to be sent
to the remote site as the local echo canceller treats this signal as a local talker.
Do not apply dynamics processing or other non-linear signal processing on the
loudspeaker signal as that will distort the echo signal substantially from the
signal the echo canceller is expecting to see for the echo canceller reference. If
dynamics processing is required, process the signal before it is used as the echo
canceller reference.
Tail Time
The tail time of an echo canceller is the amount of time the echo canceller can
still recognize a signal as an echo from the remote talkers' speech and not
interpret that signal as speech from a local talker.
All echo cancellers have some upper limit, called the tail time, after which they
do not recognize the echo signal as a version of the reference signal and
consequently can not remove the echo from the room. The tail time is
measured in milliseconds (although can be interpreted in terms of distance)
and should be greater than 100 msec for medium sized rooms and greater than
A
Filter
AEC
Far Out
Far In
NLP
Near In
Near Out