asterisk users - Jan 2015 - PJSIP / T.38 - Asterisk not passing on v21 preamble and data

Dear list,

happy new year!

I am still trying to make T.38 work. In the meantime, I have upgraded to
Asterisk 13.1.0, and I am using the most recent PJSIP library (compiling that
stuff myself). My local fax software is capable of T.38, as is my ITSP; Asterisk
sits in the middle, of course. Asterisk is in the same private subnet as the
local fax software and talks to the ITSP through a NAT'd connection.

I think I'm almost there, but I am experiencing something strange: Asterisk
is not forwarding the v21 preamble and the other v21 data to the local fax
software. Is there any typical configuration error which could be the cause?

I have captured the call flow and have seen the following so far:

First, the fax software invites Asterisk for G711, then Asterisk invites the
ITSP's SIP gateway for G711. This works, all INVITEs are OK'd and
ACK'd, G711 packets are flowing between Asterisk and the ITSP's media
gateway in both directions, and Asterisk is passing on the appropriate packets
to the local fax software.

Then, like expected, at a certain point the ITSP invites Asterisk for T.38, and
Asterisk passes the INVITE to the local fax software. Again, all INVITEs are
OK'd and ACK'd.

After that, the ITSP's media server sends a v21 preamble and some other v21
data which Asterisk receives correctly, but does *not* pass to the local fax
software.

According to the following document, this T.38 call flow is normal except that
Asterisk should pass on the v21 data to the local fax software:
https://www.escaux.com/docs/DRD_T38Support_AdminGuide.html

Now I am worried regarding multiple questions:

1) Did anybody test T.38 with SPANDSP? If yes, which version of SPANDSP did you
use? Mine is 0.0.6 PRE 20. Should I try to upgrade to PRE 21? Or to one of the
snapshots?

2) Recently, I have sent some questions about similar subjects to this list, and
I have got helpful answers; people told me that I should *not* enable the fax
*gateway* feature if both endpoints are capable of T.38. On the other hand, I
have read (at multiple places) the the *gateway* code is responsible for
detecting the v21 preamble. How does this fit together?

3) Does codec_dahdi play any role in a T.38 scenario? If I load it, I get an
error message relating to /dev/dahdi/transcode, so I disabled it. On the other
hand, I think I'm using dahdi for timing (at least, res_timing_dahdi is
loaded), hence the question.

I first would like to know if there is some sort of typical error which could
prevent Asterisk / PJSIP / SPANDSP from passing the v21 data to the local fax
software. If there is no such error, I'll make the log and configuration
available for download (can't provide them here due to the 40 kB message
size limit).

Thank you very much for any thoughts,

Recursive

Hi,

On Thu, Jan 1, 2015 at 7:09 PM, Recursive <lists at binarus.de>
wrote:
> 1) Did anybody test T.38 with SPANDSP? If yes, which version of SPANDSP did
you use? Mine is 0.0.6 PRE 20. Should I try to upgrade to PRE 21? Or to one of
the snapshots?
I remember that I was having issues with an older libspandsp library.
I simply upgraded to the latest available version which made the issue
go away. We were using 0.0.6 pre 12 before and upgraded to 0.0.6 pre
20. However, I believe (and please someone do correct me if I'm wrong)
that libspandsp is not required when using t38 passthrough, which is
what you are trying to do.

BTW, in the old thread I had asked if you saw anything in the logs
about the SIP message that was being retransmitted. Do you still see
retransmission with asterisk 13 and pjsip and a hangup after 30
seconds?
> 2) Recently, I have sent some questions about similar subjects to this
list, and I have got helpful answers; people told me that I should *not* enable
the fax *gateway* feature if both endpoints are capable of T.38. On the other
hand, I have read (at multiple places) the the *gateway* code is responsible for
detecting the v21 preamble. How does this fit together?
AFAIK, when both ends support T.38, you don't need the gateway and can
turn to passthough mode. In passthrough mode, asterisk does not need
to detect anything. Just take the packet it receives and pass the
payload to the other end of the call.

Cheers,

Frederic

On 06.01.2015 13:44, Frederic Van Espen wrote:
> On Thu, Jan 1, 2015 at 7:09 PM, Recursive <lists at binarus.de>
wrote:
>> 1) Did anybody test T.38 with SPANDSP? If yes, which version of SPANDSP
did you use? Mine is 0.0.6 PRE 20. Should I try to upgrade to PRE 21? Or to one
of the snapshots?
> 
> I remember that I was having issues with an older libspandsp library.
> I simply upgraded to the latest available version which made the issue
> go away. We were using 0.0.6 pre 12 before and upgraded to 0.0.6 pre
> 20.
That's the version I am using. Nevertheless, I'll try to compile and use
PRE 21 and the latest snapshot. I'll report the results.
> However, I believe (and please someone do correct me if I'm wrong)
> that libspandsp is not required when using t38 passthrough, which is
> what you are trying to do.
Then I'll just delete the SPANDSP module from the module directory (or
compile a version without libspandsp it this does not work), test again and
report the results.

Could it make a difference if I use Digium's fax module instead of SPANDSP?
> BTW, in the old thread I had asked if you saw anything in the logs
> about the SIP message that was being retransmitted. Do you still see
> retransmission with asterisk 13 and pjsip and a hangup after 30
> seconds?
Currently I don't see such messages / timeouts, but this may be due to the
fact that I currently don't get a T.38 connection which is lasting long
enough. As far as I can remember, the 32 second timeout in my old chan_sip
configuration startet exactly when the communication switched to T.38.

Since PJSIP (or whoever) currently does not pass on the preamble to the local
fax software, the switch to T.38 still happens, but the local fax software then
says BYE after a few seconds (because obviously it does not detect the other end
due to Asterisk not passing on the preamble packets). I'm not sure if the
timeout would strike again if I had a T.38 "connection" for more than
5 (or so) seconds.
>> 2) Recently, I have sent some questions about similar subjects to this
list, and I have got helpful answers; people told me that I should *not* enable
the fax *gateway* feature if both endpoints are capable of T.38. On the other
hand, I have read (at multiple places) the the *gateway* code is responsible for
detecting the v21 preamble. How does this fit together?
> 
> AFAIK, when both ends support T.38, you don't need the gateway and can
> turn to passthough mode.
This is what I'm trying to do. Gateway is turned off in the dialplan and in
the endpoints' description.
> In passthrough mode, asterisk does not need
> to detect anything. Just take the packet it receives and pass the
> payload to the other end of the call.
Well, that's what I have been thinking as well. But Asterisk does not pass
on the packets which it receives from the ITSP's media gateway.

Thank you very much,

Recursive