asterisk users - Jan 2006 - Strange echo phenomenon (double tandem)

I have a strange problem with echo.

My setup includes a Grandstream HT-488 which is both an FXO and a FXS.

I noticed last evening that if I called the FXS through my asterisk box 
from my cell, the resulting connection was fine for me at the cell end, 
  but produced dramatic and conversation ruining echo at the FXS of the 
HT-488.

Today I realized that if I call out from the same FXS to my cell, the 
conversation sounds great at both ends?

At first I wasn't disturbed by this, as I thought perhaps there was a 
different codec being used or something? But cranking up the verbosity 
and checking this theory shows they are both using g729.

I tried a different cell with a different wireless service provider and 
the result was identical.

Any thoughts on why this might be, and ideas on how it might be fixed?

Thanks,
Marty

PS Current asterisk as of 1/30/2006

> I have a strange problem with echo.
> 
> My setup includes a Grandstream HT-488 which is both an FXO and a FXS.
> 
> I noticed last evening that if I called the FXS through my asterisk box 
> from my cell, the resulting connection was fine for me at the cell end, 
>   but produced dramatic and conversation ruining echo at the FXS of the 
> HT-488.
> 
> Today I realized that if I call out from the same FXS to my cell, the 
> conversation sounds great at both ends?
> 
> At first I wasn't disturbed by this, as I thought perhaps there was a 
> different codec being used or something? But cranking up the verbosity 
> and checking this theory shows they are both using g729.
> 
> I tried a different cell with a different wireless service provider and 
> the result was identical.
> 
> Any thoughts on why this might be, and ideas on how it might be fixed?
Pure guess...

The echo canceller (EC) on the 488 is rather limited and the call paths 
for incoming vs outgoing calls are on the edge of the canceller. The
EC training algorithum might also be a little different for incoming
vs outgoing calls.

The only basis for that guess is my playing around with a new 488 running
the latest available firmware on a pstn line that is known to have a
long echo tail. It fails miserably and can't be used on this pstn line
at all.