[Asterisk-Users] [Asterisk-Dev] Severe ISDN signal distortion in
CVS-HEAD with octoBRI
Elwin Andriol
elwin at heuveltop.nl
Tue Aug 23 06:58:06 MST 2005
BUG/SYMPTOMS:
1.Under certain circumstances, octoBRI (and most likely quadBRI) ISDN
cards (Junghanns/CologneChip) severely distort certain ISDN payload.
2.Although these claims relate to the bristuff patch, the problem might
not be limited to bristuff and in fact be rather asterisk/zaptel/libpri
related.
3.Signal distortion is limited to the use of the CVS version of the
Bristuff patch for asterisk (tested on
Bristuff-0.2.0-RC8f-CVS/Asterisk-CVS-05.29.05). The Stable versions of
the Bristuff patch do not have the problem (tested on
Bristuff-0.2.0-RC8f/Asterisk-1.0.7 and Bristuff-0.2.0-RC8n/Asterisk-1.0.9).
4.Signal distortion is limited to calls with an ISDN Transfer
Capability of 0x00 (SPEECH) or 0x10 (3K1AUDIO). Calls with an ISDN
Transfer Capability of 0x08 (DIGITAL) are not being effected.
5.The amount of signal distortion is of such severity that it is almost
impossible to reliably send a fax call through the ISDN card.
6.The amount of signal distortion also appears to render the echo
cancellation useless.
SOLUTION:
None yet.
NEEDED:
Someone with knowledge of the asterisk/zaptel/libpri internals, to
guide me.
TEST SETUP:
The following hardware and software has been used to test the problem:
- 1.7GHz Intel P4 / 512M
- CologneChip octoBRI ISDN card (not tested with Junghanns.NET versions
yet)
- Debian Sarge / 2.6.12 kernel
- Asterisk (CVS-05.29.05 / 1.0.7 / 1.0.9) + Bristuff (RC8f-CVS / RC8f /
RC8n)
- Elmeg and DeTeWe OpenCom PBX as analog channel banks.
TEST RESULTS:
asterisk + bristuff version: ISDN L1 Bit Error Rate:
05-29-05 + RC8f-CVS >1%
1.0.7 + RC8f 0
1.0.9 + RC8n 0
SITUATION/BACKGROUND:
I'm new to asterisk development and uncertain about reporting this in
mantis. Because the cause of the mentioned symptoms hasn't been pinned
down yet, and therefor might be something else than currently suspected,
I will try to explain the context in which the problem has surfaced and
the rationale by which I came to my assertions. Because of this, and the
fact that I couldn't find any related material on the web, this document
has become quite lengthy. Hopefully a bit of reading won't scare off
most, for the implications of this problem might be related to a large
set of other problems.
For a while I've been deploying asterisk PBXs on a commercial basis.
Most of those sites use asterisk with an octoBRI ISDN card. The ISDN
card is used to hook up to the telco and, if the site already has an
existing telephony infrastructure, to a conventional PBX acting as an
analog channel bank to facilitate the existing infrastructure.
After been plagued by problems with call transfers and some other
issues with BT101s, I decided to switch from the stable
bristuff/asterisk to the CVS-HEAD version. Although this switch has
solved a lot of problems for me, there have been considerably more
complaints about echo since. More frustrating, nothing seems to help to
get rid of the echo.
One month ago, I converted a site that used a hylafax solution. The
hylafax machine was connected to a DeTeWe PBX with an analog (POTS)
line. After the conversion the DeTeWe PBX was merely an analog channel
bank that connected the hylafax machine to the asterisk machine. The
asterisk machine had taken over all the functionality of old PBX. Both
were connected to each other by 4 ISDN interfaces. For reasons unknown
to me at that time, it wasn't possible to consistently receive faxes on
the hylafax machine. Only 1 out of 20 faxes was received without errors.
The strange thing was that sending faxes worked fine. Because I couldn't
find the cause of the problem with receiving faxes, I devised another
solution that is far from perfect and goes beyond the scope of this
document.
A week ago, still annoyed by not knowing what caused the fax problem, I
tried to route some personal calls (including a fax) through the
asterisk machine. To my surprise the fax was mangled as well, which let
me to believe the problem wasn't in the hylafax machine nor the DeTeWe
PBX. Next, I asked a friend to test the asterisk machine with a
professional ISDN testing device. An ISDN (layer 1) Bit Error Rate test
resulted in almost 1% of errors for voice call and fax call (TC 0x00 and
0x10), whereas ISDN should have a BER less than 1 billionth percent. It
explained why the faxes got mangled. The fact that the transmission of
faxes went well, turned out to be because those calls got a DIGITAL TC
(0x08) from the conventional PBX. Those digital ISDN call, as intended
for G4 faxes (e.g.) weren't distorted in any way.
Puzzled by the fact that I couldn't find anything on the wiki, mailing
lists or mantis about the inability to properly send faxes through an
octoBRI, I tried to determine the cause myself. First I installed the
latest stable version of bristuff (0.2.0-RC8n). To my surprise this
version did not distort any calls, indiscriminate of the call's TC.
Since I normally use RC8f-CVS version of bristuff, I also tested the
RC8f version. This version also did not distort calls. This let me to
believe that the problem is not in the bristuff driver (qozap), but
rather in some difference(s) between the stable and the CVS-HEAD version
of asterisk. In fact, a test on an asterisk CVS-HEAD system without
bristuff, but with a TE410P card also showed some signal distortion,
though far less severe.
Because Junghanns.NET appears to have stopped development of the CVS
versions of their bristuff drivers and the last version is a little
outdated (uses CVS 05-29-05), it's impossible to say if the distortion
of fax calls might be something resolved by now in CVS-HEAD. I'm not
very familiar with the inner workings of the bristuff code, therefor I
feel very reluctant to rewrite/tweak the RC8f-CVS patch to make it work
on the current CVS-HEAD version. I also don't think that would solve
anything since mantis didn't have any related issues (that I could find)
since.
From a theoretical viewpoint I can imagine that it might be useful to
put an ISDN voice call (TC 0x00 SPEECH) through a band or low-pass
filter to reduce noise and eliminate inaudible components. If the
measured call distortion isn't a bug, this might be what's happening in
CVS-HEAD. If that is true, apparently the same thing is done for non-G4
fax calls (TC 0x10 3K1AUDIO). This apparently screws the fax data
signal, thus rendering faxing through asterisk practically useless. It
also is dubious if filtering on a voice call might be helpful, for in
theory it can severely disrupt/disable an echo cancellation algorithm.
For anyone thinking of using the SetTransferCapability application in a
dial plan to circumvent the problem by setting the TC of fax calls to
DIGITAL (0x10), this is not an option. As soon as that call will arrive
at another PBX with an analog fax machine attached to it, the call will
be rejected (as per ISDN specification). Therefor it seems the only
solution is to figure out what's going wrong with ISDN calls with a TC
of 0x00 and 0x10 in the CVS-HEAD version of asterisk.
I'm unable to find any code that might cause any (band/low-pass)
filtering of calls. I'm not even sure where to look for it, if any. My
best guess would be in the zaptel or maybe libpri driver, but I'm not
familiar with that level of protocol workings. At least not yet.
I would be very pleased if anyone could point me the right direction
with this one. I'm willing to put quite some effort in resolving this
problem. I'm currently out of options, for I lack the knowledge of the
inner structure and workings of asterisk.
If my assertions are right and there is an issue with the quality of
fax and voice call in the CVS-HEAD, it might better be fixed before
asterisk 1.2 gets out (from a PR standpoint that is). Therefor I realy
hope someone can lend me a hand.
sincerely,
Elwin Andriol
elwin at heuveltop.nl
"There are management solutions to technical problems,
but no technical solutions to management problems"
_______________________________________________
Asterisk-Dev mailing list
Asterisk-Dev at lists.digium.com
http://lists.digium.com/mailman/listinfo/asterisk-dev
To UNSUBSCRIBE or update options visit:
http://lists.digium.com/mailman/listinfo/asterisk-dev
More information about the asterisk-users
mailing list