[asterisk-users] HPEC > VPM ?

[Steve Underwood]

Noah Miller wrote:
>>> Next question: does anybody know how to handle extremely long tail
>>> echo that a VPM module cannot?
>>>       
>> How long is 'long' in this case? The VPMs and HPEC (and OSLEC) can
>> handle 128ms echo tails, which is pretty darn long. It's rare to see an
>> echo tail longer than that except on very high latency connections, or
>> when the echo is actually acoustically generated by the far end and not
>> by network effects.
>>     
>
> I haven't done any real measurement on it, but I believe it's actually
> longer than 128ms.  As I go higher and higher with echocancel values,
> the echo does get better, but is never totally eliminated.  At
> echocancel=1024, there is still rather pronounced echo on calls in the
> "local" exchanges.  The calls are also more or less half-duplex at
> that point because the vpm is filtering out so much of the signal as
> echo.
>   
Tails are *never* long. Nobody really builds a 128ms canceller. They 
cancel a few specific sections of delay within a 128ms interval. 
Typically four 8ms sections - one for each end of the local analogue 
link, and one for each end of the far analogue link. Things should not 
get progressively better by gradually lengthening the echo canceller. 
The echo should basically just disappear when the length is sufficient, 
and the algorithm can place and tune an 8ms canceller at the right delay.
> I may just tell the client to look at a partial PRI.  All this echo
> chasing is getting costly for them.
>   
If the echo varies with which number you call, its predominantly from 
the far end. In those cases PRI won't help. You still need to cancel 
that far end echo.

Steve