[Asterisk-Users] Echo Cancellation Feature

[Rich Adamson]

> I do feel the echo cancellation does need some work.
> 
> Currently, other than listening to users, there is no way to benchmark or
> trouble shoot echo problems.

Sure there are, it's just that 99% of the asterisk implementors don't
have the test equipment to do it, and a good share probably wouldn't
know how to do it if they had access to the equipment.

> We find that 2 to 3 out of every 20 calls will experience echo.  While
> echo is a problem that naturally occurs from SIP - PSTN and vice versa, I
> am still baffled by the fact that the cancellation works randomly.
> 
> When doing a zap show channel X, it will also report that the cancellation
> is still on.  We experience the most echo with a T100P --> Adtran TA 750
> FXO modules.  While I understand these do not have impedence matching, I
> wonder to myself why echo cancellation works sometimes, and others not.
> 
> Looking at Network util, processor util, and memory utilization, they do
> not provide any clear indication as to why /when it occurs.

Not likely to have any impact whatsoever.
 
> Is there anything more that can be done to debug echo cancellation, and
> further are other users experiencing this random echo.  I know it was
> discussed before, but the support folks at digium aren't able to offer
> anymore help.

You've probably read enough from previous postings to know there are
several different locations within an end-to-end voice call where echo can
creap into a system. In very general terms, any place where an end-to-end 
channel incures a two-wire to four-wire conversion (whether done in hardware
or software), echo can creap in due to lots of different reasons. Since
asterisk provides us with lots of configuration choices, hardly any two
systems are alike. Therefore, don't know that anyone is going to write
* code anytime soon to correct something that can't be pointed to, etc.

Someone mentioned they have echo on sip to sip calls (presumably the call
was processed by a single * system). If they do, the problem is likely
in the sip phone as there are no echo cancallation needs in that four-wire
end-to-end call from an * perspective.

I've got a fair amount of test equipment (and 20+ years telephony 
background), and am planning to assemble a document identifying some of 
the pstn issues, level settings, and other things impacting a reasonable 
system implementation. Unless someone wants to UPS a transmission test 
set to me quickly, the document won't be completed for several weeks. 
(The only test set I have access to will not be released for a couple 
of weeks due to classes, etc.)

I'm also expecting these tests to point out a number of other transmission
issues within asterisk that we'll get documented with real numbers, etc.

Rich