[asterisk-users] Clarification on DAHDI Fax Detection

[Steve Underwood]

On 01/04/2011 09:53 PM, Kevin P. Fleming wrote:
> On 01/03/2011 07:08 PM, Steve Underwood wrote:
>> On 01/04/2011 04:22 AM, Kevin P. Fleming wrote:
>>>
>>> No. CNG tone is never used to affect the state of an echo canceller.
>>> All G.168 compliant echo cancellers will respond to the CED tone
>>> (generated by the answering endpoint) and will reconfigure the echo
>>> canceller appropriately. Most modern ECs will *not* be disabled, but
>>> will enter a 'linear' mode where they can do some echo suppression but
>>> not complete cancellation. DAHDI will detect CED when most software
>>> echo cancellers are in use and will disable them (since none of the
>>> available software ECs can go into linear mode). The Digium HPEC
>>> software EC will detect CED on its own and enter linear mode.
>> That's not true. Modern echo cancellers normally disable completely. It
>> is arguable whether they should disable completely for FAX, but they
>> need to behave properly for all modems. For any duplex modem, disabling
>> only the NLP is useless. They need to cancel end to end, so they don't
>> get upset by a continuously adapting canceller, and so they can minimise
>> the issues caused by the highly non-linear G.711 channel.
>
> This doesn't match up with what the manufacturers of the two G.168 ECs 
> that Digium distributes have told me personally about their products. 
> Their ECs behave differently for FAX and 'regular' modems, but they do 
> that based on the detection of the V.21 preamble, ANSam and other 
> signals in addition to CED, which seemed to be much more detail than 
> was warranted in my response to the OP :-)
Well, that makes a bit more sense, but I am very skeptical about this. 
The Octasic canceller is highly problematic with various modems and 
tones, so they aren't exactly a reference model for how to do things. 
Reports I here of the other canceller are much more positive. Its 
obvious why they want to keep the canceller alive. Long echoes over VoIP 
channels, combined with slow responding FAX boxes, can lead to a FAX 
machine hearing its own output heavily delayed, and it may mistake this 
for the response from the far end. T.38 largely avoids this kind of issue.

The start of a FAX call doesn't really have a good signal on which to 
train a canceller. They can use the first V.21 burst in each direction 
(The FAX signals for G3 or the V.8 exchange for Super G3), and then lock 
down the canceller, but those signals aren't wide band enough to be 
ideal. The canceller could adapt very oddly. If they continue adapting 
once the wideband signals from the fast modems start, they are likely to 
upset modem operation there. If they just accept that, and rely on the 
fast modem retrying, it will usually step down in speed. I believe I 
have seen this behaviour in setups where the signal looks very clean, 
but the FAXes always exchange at 12000bps.

Steve