[Asterisk-Users] Fun with Echo

[Rich Adamson]

Eric "ManxPower" Wieling wrote:
> The number of taps the EC has to deal with is the delay on the PSTN 
> side.  I can't imagine echo is more than a few ms in modern TDM 
> networks.  This latency has NOTHING to do with VoIP latency, since the 
> echo must be canceled BEFORE it gets to the VoIP side of things.

Not so sure about that.

We've got four analog pstn lines from Alltel; three from one CO and a 
fourth from a second CO (Centrex). Both CO's are relatively new switches 
(last ten years), and both use some form of remote line concentrator 
connected via fiber.

When using the X100P (and later the TDM04b), Mark had to implement 
echotraining=800 as the pstn line had not settled down in the first 200 
milliseconds to allow the EC to train. Seemed rather unusual at the 
time, but he was actually logged into our system and played with that 
delay (before it became a part of the code).

After the echotraining=800, things were significantly better and I 
played with each of the ECs (KB, MG2, Aggressive, etc) to find something 
that would train up in a reasonable fashion.

Now, when using a A200D and sangamo's utilities, I see the echo tail is 
roughly 38 milliseconds. All four lines (from two COs) appear the same.

I'm not sure what Alltel is using for a remote line concentrator, but it 
almost appears they are doing multiple analog-digital conversions before 
hitting the actual CO.

R.