Hans Johnson wrote the following to August Abolins, and I quote (in part):
  AA> But what is the problem with Supra's implementation of v34+ and
  AA> why can't we connect with USRs beyond 31200?
 HJ> There is nothing wrong with supra's implementation of v.34.  The
 HJ> supra folows the specs to the letter, where as the USR isn't quite
 HJ> standard thus we have this problem.  The problem should dissapear
 HJ> as soon as USR releases a new flasher for thier modems.
Not quite the case:
------------------------------------------------------------------------------
From: dan@supra.com (Dan Moore)
Subject: Re: Anyone had a 33.6 connection with the new supra flasher
Organization: Communications Division of Diamond Multimedia Systems, Inc.
In article  JNavas@NavasGrp.Dublin.CA.US
(John Navas) writes:
>jcrosby@erie.net (Jeff Crosby) wrote:
>>John Navas  wrote:
>>>> dan@supra.com (Dan Moore) wrote:
>>>>
>>>> >I assume your calling a Courier with your Supra, if so have the
>>>> >Courier call the Supra.  This will allow a 3429 symbol rate and a
>>>> >33.6Kbps connection if the line conditions permit it.  If your using a
>>>> >Sportster 336 you should be getting a 3429 symbol rate when either
>>>> >originating or answering a call to the Supra assuming the line 
conditions
>>>> >allow it.
>>>>
>>>> So far for a courier being better than a sportster. I hope USR will
>>>> fix this soon. Dream on dream on ... ;-))
>>
>>>This is a Supra/Rockwell problem, not a USR problem.
>>
>>Actually, its a USR/Courier problem.  From what I've heard, the was a
>>problem with an early Hayes modem connecting with the 3429 symbol rate
>>to the courier.  Courier disables that symbol rate to fix the problem.
>>Rockwell/Supra/etc have solved the problem a long time ago
>>( > 1 year>. The problem has since been with the courier.  USR still
>>makes the couriers to disable the 3429 symbol rate of incoming
>>rockwells to fix a problem that doesn't even exist anymore.
>With all due respect, you are misinformed.  The basic problems are that
>Supra used an initial negotiation sequence that indicates a maximum speed
>of 28.8 (rather than 33.6), and that insufficient interoperability testing
>was performed before the Supra product was released.  Regardless, it's
>pretty much a tempest in a teapot, because 33.6 connections are very rare.
        As John states, this issue is pretty much a "tempest in a teapot"
since 33.6 connections only are possible on a small percentage of the
network.  If your going to get a 31.2Kbps connection then it doesn't really
matter if your using a 3200 symbol rate or a 3429 symbol rate.
        I had hoped to ignore this until after our next firmware release,
unfortunately I don't believe I can do that.  There are people giving lots of
different speculated causes, including people repeating variations of some
statements I had made in the past.  I know better than to comment prior to
knowing all the facts.  Instead I made a comment, that wasn't well phrased,
that appeared to blame USR.  I had intended to say that I thought the fix had
to come from them, since at that point I was not aware of a possible solution
on my side.  So a large part of the blame for the confusion and on going
discussion is mine.
        John and I have exchanged a few emails on the subject, we agree on
most things and disagree on a few.  He feels (with some justification) that
this problem is due to Diamond/Rockwell.  I feel (with different
justifications) that the problem is due to a combination of Diamond/Rockwell
and USR actions.
        To be honest I was rather pissed when I finally realized that each
company had very carefully done the correct thing (or at least one of the
possible correct things) in order for us to all shoot ourselves in the foot.
If any of the involved parties had done things differently there never would
have been a problem.  "The road to hell is paved with good intentions,"   I
believe that pretty much summarizes this whole issue.
         When originating a V.34 call Diamond modems, with v1.440-17
and earlier firmware, send INFO1C sequences that project a maximum data rate
of 28.8Kbps for both the 3200 and 3429 symbol rates.   When answering a call
the USR modem looks at the projected maximum data rate for the symbol rates
that are possible with the bandwidth on the line and chooses the symbol rate
with the highest projection.  If the 3200 and 3429 symbol rates project the
same maximum data rate they choose the 3200 symbol rate.  (I believe they 
nly
do this when the 2743 symbol rate is disabled.  This is based on limited
testing so it's possible there are other factors that I'm not aware of.)
         What follows is part of an email I sent to John on the subject.
(It's the part I wrote, I won't post his emails.)
        ===================================================
        The revised version of recommendation V.34 states that bit 25 of the
INFO0 sequence indicates the support of up to 1664 point constellation.  In
the definition of the MP0 and MP1 sequences it states that "data rates 
reater
than 12 shall only be indicated when the remote modem supports up to 1664
point signal constellations".  There is no statement that the data rates
enabled in the MP sequence must be lower than the projected maximum data rate
for the current symbol rate that was in the INFO1 sequence.
        The INFO1C and INFO1A sequences both include statements that
"projected maximum data rates greater than 12 shall only be indicated when 
he
remote modem supports up to 1664 point signal constellations".  There is no
statement that the projected maximum data rate for any symbol rate will be 
he
maximum possible for that symbol rate, or that data rates higher than the
projected maximum data rate are not available.  There are modems (such as 
urs
with our current firmware, as well as some non-Rockwell based products) that
project a maximum data rate of 28.8 for the 3200 and 3429 symbol rates, even
if 31.2Kbps or 33.6Kbps data rate is possible.  There are modems (such as 
SR)
that put the maximum possible data rate for a symbol rate in the projected
maximum data rate field.  There also are modems that put an actual projection
based on the results of the line probe in the projected maximum data rate
field (ie. they may put 21.6Kbps in the field and then make a 31.2Kbps
connection).
        Our putting 28.8Kbps in the projected maximum data rate fields of the
INFO1C sequence does appear to conform with the V.34 recommendation.
        USR chooses the 3200 symbol rate when answering a call from a modem
that doesn't support the 2743 symbol rate and that indicates the same maximum
projected data rate for both the 3200 and 3429 symbol rates.  This does not
match what some other V.34 implementations do.  By choosing the 3200 symbol
rate instead of the 3429 symbol rate they are giving up about 2dB of SNR.  On
a lot of lines that is the same as giving up one data rate.  Every other V.34
implementation I've tested chooses the highest symbol rate that is supported
by the available bandwidth, simply because the higher symbol rates have 
etter
noise immunity than the lower ones.  As I stated in my last email, I suspect
USR chooses the 3200 symbol rate in order to avoid problems with very old
Rockwell products.  This problem was fixed well over a year ago.  I can't
blame USR for making the choice they did since there are a lot of modems that
still have the problem.  (We're finding customers with modems with flash that
are running the original V.34 firmware we shipped.  They had never upgraded
until they heard about the 33.6Kbps upgrade.)
        Is USR wrong to not choose the highest possible symbol rate?  They
aren't violating the V.34 recommendation by doing so.  They are causing a lot
of people to get lower data rates when calling USR modems, even if we ignore
the 31.2Kbps and 33.6Kbps data rates.  They also are preventing problems with
a lot of users of old Rockwell based modems.
        The revised version of recommendation V.34 only states one 
equirement
for supporting the 31.2Kbps and 33.6Kbps data rates, that the modems must
support signal constellations with up to 1664 points.  Everything else is 
eft
up to the implementation.  Both Diamond/Rockwell and USR made choices that
conform with the V.34 recommendation, the combination of those choices 
revent
the selection of a 3429 symbol rate when calling from our modems to USR 
modems.
        ===================================================
--
Dan Moore
Diamond Multimedia
------------------------------------------------------------------------------
-
Beyond a few misconceptions about how the USR V.34 implmentation actually 
works, Dan expresses a great deal of the gist of the problem.
Regards....
Craig
aka: cford@ix.netcom.com
--- timEd/2 1.10+
---------------