asterisk users - Oct 2004 - echo cancellation: the never-ending quest for truth

Asterisk apparently has five echo cancellation algorithms: STEVE, STEVE2,
MARK, MARK2 and MARK3.  The current default appears to be MARK2.

My question is, has anyone had any experience with any of the others
(other than MARK2), and is there some conventional wisdom as to when to
use one over another?

TIA

Bruce Komito
High Sierra Networks, Inc.
www.servers-r-us.com
(775) 236-5815

"Bruce Komito" <brucek@bagel.com> spake thusly:
> Asterisk apparently has five echo cancellation algorithms: STEVE, STEVE2,
> MARK, MARK2 and MARK3.  The current default appears to be MARK2.
>
> My question is, has anyone had any experience with any of the others
> (other than MARK2), and is there some conventional wisdom as to when to
> use one over another?
Bruce,

I asked Mark that question at Astricon, since it seems to me that this 
section of the code offers lots of options with little guidance, and I 
haven't seen this matter receive any in-depth treatment on the list lately 
either (I guess most people simply accept the out-of-the-box default, and 
those who do have echo problems for any reason (like us) dig a little 
deeper).

He said that MARK2 is the recommended one, and that some people have found 
the aggressive option help.

FWIW, we run MARK2 in aggressive mode here.

-Darren

--
Darren Nickerson
Senior Sales & Support Engineer
iFax Solutions, Inc. www.ifax.com
darren.nickerson@ifax.com
+1.215.438.4638 x8106
+1.215.243.8335 (fax)

> Asterisk apparently has five echo cancellation algorithms: STEVE, STEVE2,
> MARK, MARK2 and MARK3.  The current default appears to be MARK2.
> 
> My question is, has anyone had any experience with any of the others
> (other than MARK2), and is there some conventional wisdom as to when to
> use one over another?
Most of those were variations that attempted to address echo issues with
the old x100p card. If you look at the source code dates, none of those
have really been touched in well over a year.

According to Steve, the echo cancellation algorithms do not have a very
wide operational range, and as a result its not all that difficult to find pstn
situations were the echo characteristics fall outside the usable ranges.
Its basically a trial & error test to see which one works the best given
"your" pstn line characteristics. Your neighbors line characteristics
may be completely different.

Following Mark's echotraining=800 changes around June 2004, it seems
most echo problems were minimized, and there hasn't been any list
activity relative to those algorithms since then.  Those that still had
objectionable echo found that swapping motherboards cured their issues.
That seemed to be related to interrupt latency and/or pci bus controller
chip problems that are apparently well known to those that work heavily 
with audio applications.

Rich

Bruce Komito wrote:
> Asterisk apparently has five echo cancellation algorithms: STEVE, STEVE2,
> MARK, MARK2 and MARK3.  The current default appears to be MARK2.
> 
> My question is, has anyone had any experience with any of the others
> (other than MARK2), and is there some conventional wisdom as to when to
> use one over another?
> 
> TIA
> 
> Bruce Komito
> High Sierra Networks, Inc.
> www.servers-r-us.com
> (775) 236-5815
> 
Make sure that you check the WIKI. Try starting at:
 
http://www.voip-info.org/tiki-index.php?page=Asterisk%20echo%20cancellation#comments

I downloaded a paper from the web which explained the math, but I can't 
find it now. You should be able to Google "Simple LMS with Doubletalk 
detection" - which is what I did to find it in the first place.

Stephen R. Besch