asterisk users - Jan 2007 - Sangoma A102 w/ EC module gets intermittent echo /audio artifacts

I've replaced 2XTE110 with an A102 with echo cancellation specifically to
deal with echo problems. However, user feedback has indicated to me that on
some calls (not a lot, but some) the call is unusable, with audio
artifiacts, described by one user, as: "very bad phasing reverb &
feedback
(from my rock & roll days)". This is quite intermittent, as in most
cases,
the user says, it's a one-time thing, they hang up, the problem caller calls
back, everythings good. It's as if the Sangoma is trying too hard? I
personally have not heard this, but I have to trust what the users say. 

Some ideas I'd like to bounce:

1. tx and rxgains - this card is plugged into an Atlas 550 which seems to
run a little hot on the gains. 
2. Timing - the card takes it's sync from the Atlas, which in turn syncs
from the PRI. Maybe have one port on the card take it's timing from the
other port? Don't see how it would be relevant, but hey, that's all
I've
got. 
3. Taps? Does an A102 even care about taps or just echocancel=yes?

Running * 1.0.9, Zaptel 1.0.9, FC2 yum-updated to current, quad Xeon.
Production box, handles 2-6 thousand calls a day, snom 360 handsets w/
latest firmware, load average never goes over 2.0 

My conf files:

wanpipe1.conf:

[devices]
wanpipe1 = WAN_AFT_TE1, Comment

[interfaces]
w1g1 = wanpipe1, , TDM_VOICE, Comment

[wanpipe1]
CARD_TYPE 	= AFT
S514CPU 	= A
CommPort 	= PRI
AUTO_PCISLOT 	= NO
PCISLOT 	= 4
PCIBUS  	= 1
FE_MEDIA	= T1
FE_LCODE	= B8ZS
FE_FRAME	= ESF
FE_LINE		= 1
TE_CLOCK 	= NORMAL
TE_REF_CLOCK 	= 0
TE_HIGHIMPEDANCE	= NO
LBO 		= 0DB
FE_TXTRISTATE	= NO
MTU 		= 1500
UDPPORT 	= 9000
TTL		= 255
IGNORE_FRONT_END = NO
TDMV_SPAN	= 1
TDMV_DCHAN	= 0

[w1g1]
ACTIVE_CH	= ALL
TDMV_ECHO_OFF	= NO
TDMV_HWEC	= YES

wanpipe2.conf:

[devices]
wanpipe2 = WAN_AFT_TE1, Comment

[interfaces]
w2g1 = wanpipe2, , TDM_VOICE, Comment

[wanpipe2]
CARD_TYPE 	= AFT
S514CPU 	= A
CommPort 	= PRI
AUTO_PCISLOT 	= NO
PCISLOT 	= 4
PCIBUS  	= 1
FE_MEDIA	= T1
FE_LCODE	= B8ZS
FE_FRAME	= ESF
FE_LINE		= 2
TE_CLOCK 	= NORMAL
TE_REF_CLOCK 	= 0
TE_HIGHIMPEDANCE	= NO
LBO 		= 0DB
FE_TXTRISTATE	= NO
MTU 		= 1500
UDPPORT 	= 9000
TTL		= 255
IGNORE_FRONT_END = NO
TDMV_SPAN	= 2
TDMV_DCHAN	= 0

[w2g1]
ACTIVE_CH	= ALL
TDMV_ECHO_OFF	= NO
TDMV_HWEC	= YES

Zaptel.conf:

loadzone = us
span=1,1,0,esf,b8zs
span=2,1,0,esf,b8zs
bchan=1-23
dchan=24
bchan=25-47
dchan=48

Zapata.conf:

[channels]
language=en
context=from-pstn
switchtype=national
pridialplan=unknown
signalling=pri_cpe
faxdetect=incoming
usecallerid=yes
echocancel=yes
echocancelwhenbridged=yes
txgain=-2.0
rxgain=-2.0
group=0
channel => 1-23
channel => 25-47

< Zaptel.conf:

< loadzone = us
< span=1,1,0,esf,b8zs
< span=2,1,0,esf,b8zs
< bchan=1-23
< dchan=24
< bchan=25-47
< dchan=48

Just a quick thought in looking at the settings above, it appears that you
have set both spans as the primary timing source.  I am pretty sure that
only one span should be the primary timing source.  The other span should
either be at 0 (not used as a timing source) or set as a secondary timing
source.

Hope this helps.

Michael L. Young

Colin Anderson wrote:
> I've replaced 2XTE110 with an A102 with echo cancellation specifically
to
> deal with echo problems. However, user feedback has indicated to me that on
> some calls (not a lot, but some) the call is unusable, with audio
> artifiacts, described by one user, as: "very bad phasing reverb &
feedback
> (from my rock & roll days)". This is quite intermittent, as in
most cases,
> the user says, it's a one-time thing, they hang up, the problem caller
calls
> back, everythings good. It's as if the Sangoma is trying too hard? I
> personally have not heard this, but I have to trust what the users say. 
> 
> Some ideas I'd like to bounce:
> 
> 1. tx and rxgains - this card is plugged into an Atlas 550 which seems to
> run a little hot on the gains. 
> 2. Timing - the card takes it's sync from the Atlas, which in turn
syncs
> from the PRI. Maybe have one port on the card take it's timing from the
> other port? Don't see how it would be relevant, but hey, that's all
I've
> got. 
> 3. Taps? Does an A102 even care about taps or just echocancel=yes?
I have fixed similar problems by reducing the gains.  I call it "ECFO"
Echo Cancel Freak Out.