[Asterisk-Users] T.38 fax summary

[Steve Underwood]

Lee Howard wrote:

> On 2005.02.27 09:30 Martijn van Oosterhout wrote:
>
>> On Sun, Feb 27, 2005 at 09:10:48AM -0800, Lee Howard wrote:
>> > Fax cannot handle a one-second delay.  As Steve mentions in the
>> > article, per-spec fax has some timings (particularly silence in
>> > direction "switching") set at 75 ms +/- 20 ms.  So if the delay gets
>> > much larger than 75 ms, then there's likely to be trouble.  Now,
>> > some fax machines may tolerate larger delays, but that tolerance is
>> > beyond the spec, and thus should not be used as a gauge.
>>
>> Something's not right here.
>
>
> Quite right.  I'm sorry to have misled.
>
> What happens is this (as an example scenario):
>
> The receiver will, for an example, receive the post-page message.  The 
> sender expects a response to this.  The receiver, however, is required 
> to wait between 55 and 95 ms before transmitting the response.  The 
> sender will likely be looking for the post-page response immediately 
> after transmitting the post-page message.  Per spec the sender will 
> only wait about 3 seconds (per-spec between 2550 and 3450 ms) before 
> giving up wating and retransmitting the post-page message (and then 
> re-expecting the response).

This is also slightly wrong. The gaps in the audio stream are specified 
as 75+-20ms. The response is specified as occuring a *minimum* of 75ms 
after the received carrier has ceased.

>
> So if there is a steady 1000 ms lag between the sender and the 
> receiver (both ways, meaning we assume that both ends could have the 
> 1-second jitter buffer), what will happen is this:
>
> The sender will finish transmitting the post-page message.  One second 
> later the receiver finishes getting it.  The receiver will introduce 
> its own required pause, and add to that the overhead of any processing 
> required, and then it will return the signal.  The sender will not get 
> that signal for another 1000 ms.  That means that for the total 
> processing of that to occur the 2550 ms danger-zone time is nearly 
> reached.  Add to that buffer-time the latency time, and I'd say that 
> you'd be looking at a signal failure quite certainly.
>
> In real-world action, however, the 2550-3450 ms danger-zone time is 
> practically never reached.  In normal use that time is often very 
> close to 400 ms.
>
> So yes, 75 ms latency is not accurate for a command-response 
> interaction between two fax machines.  And, per-spec the response 
> could, in theory, sustain a 1000 ms lag.  However, that would 
> far-exceed normal behavior, and I'd be surprised if it would not prove 
> fatal to most fax communications.

I think this still allows significant buffering - say 500ms - without 
causing trouble. Extreme buffer would, however, be troublesome. 500ms, 
less the rollover time needed for the FEC, should give pretty good 
jitter buffering.

Regards,
Steve