[Foip] FoIP-TG: Issues with V.8/V.34/V.152 call flows.

[Steve Underwood]

Paul E. Jones wrote:
> Steve,
>
>   
>> T.38 D.2.2.3 says a call starting in T.38 mode starts with a T.38 invite
>> from the source of the call.
>>     
Some systems do follow this flow for SIP, but they cause trouble because 
too many other systems choke on it. For reliable operation I've found it 
necessary for a T.38 terminal (i.e. a box that never does anything but 
FAX) to start in voice and stay that way for several seconds before 
requesting T.38 operation.

> H.323 Annex D has similar options for H.323, but nobody ever supported that
> flow, as far as I know.  The "fax only" scenarios will present problems in
> most cases going to the PSTN.
>  
>   
>> T.38 D.2.2.4 says that if a call starts in voice mode, the receiving
>> gateway is the one which should initiate a transition to T.38 is if
>> detects a FAX machine.
>>     
>
> Given that either device may first detect a fax on the line, why should the
> spec impose this requirement as to which GW initiates a voice/fax
> transition?  What was written in H.323 Annex D was this:
> " Note that in some implementations the presence of either CNG or CED is
> optional. Therefore, both endpoints should take an active role in order to
> properly detect fax."
>   
I think there are several things here.

- A *lot* of boxes choke on glare conditions. If both ends decide to 
re-invite for a switch to FAX at the same moment, the call falls over. 
Obviously that shouldn't happen, but the industry standard for SIP 
implementations is pretty awful. Specifying that one particular end 
initiates T.38 reduces this problem.

- CNG is "optional". CED is "optional". What is certain to happen is the 
initial signal from the called FAX machine. That makes the T.38 gateway 
on the called side the obvious choice for initiating the switchover. I 
find a large number of switch-over attempts are initiated from my own 
T.38 gateway software when it declares V.21 preamble to be present.

- When a FAX machine calls a voice line by mistake, is it appropriate 
for the calling end to be able to initiate a switch-over to T.38? Maybe 
this is harmless, as the call sure isn't going anywhere useful. It 
doesn't feel right, though.

As well as the issue of who initiates the switchover, there is the issue 
of what happens to the audio during negotiation. A number of systems 
mute it. A number let it through. If they let it through it may be let 
through cleanly, or a low bit rate codec may seriously corrupt it. If 
the signal gets through cleanly, and negotiation is slow, the two FAX 
machines may have exchanged DIS and DCS by the time T.38 operation 
begins, and the 4 or more boxes in the chain can get confused. This area 
is not handled well in the real world today.

Regards,
Steve