RE: XAUI IO specs
Ed,
I agree that Ali's example was incorrect, but I am surprised by your
assertions about Vcm. Vcm is, in my understanding, a part of the
measurement of a differential signal just as much as Vdm is (since
I don't have subscripts I'm making cm and dm lower case). It is
at any given moment the average between Vo+ and Vo-. Thus it is
a measured value rather than a fixed reference point. It would be
a constant if you had a perfectly balanced differential output, but
of course we don't ever have prefect balance so it moves around
with respect to a reference such as ground.
Vcm is important to receiver performance so most specs of differential
signaling do specifiy DC and AC amplitude for it.
Pat
-----Original Message-----
From: Ed Grivna [mailto:elg@cypress.com]
Sent: Thursday, April 27, 2000 2:01 PM
To: stds-802-3-hssg@ieee.org; kdemsky@us.ibm.com; t11_2@dpt.co;
jenkins@lsil.com; rtaborek@earthlink.net; Ali.Ghiasi@Eng.sun.com
Cc: Ali.Ghiasi@Eng.sun.com
Subject: Re: XAUI IO specs
Hi Ali,
I hate to say it, but your explaination disagrees with itself, which leaves
me and others just as confused as when you started. Please see my comments
below.
>
> Hi
>
> Several question have been raised about XAUI "Hari" specifications and
> what was some of the intentions:
>
> o. Based on my understanding and everyone who participated during
> conference call the Vp-p differential max defined as 800 mV.
These are differential signals. The only way a differntial signal
should ever be specified is in terms of signal RELATIVE to signal.
A Vp-p differential number identifes the p-p amplitude that would
be seen when probing a differential signal using a differntial
probe and receiving a pulse sequence containing a mix of logic
zeros and logic ones. A Vp measurement documents the difference
in voltage between the true and complement signals when driving
A logic-1 or A logic-0 (not when driving both).
> - This means each side of the driver has a swing of 400 mV.
This statement; i.e., "a swing of 400 mV" means that you are making
a single-ended measurement. Thats OK, but any single-ended measurement
must be made from a reference voltage. If done correctly (i.e., with
a stable reference and clean signaling, the measurement of "swing"
should be the same regardless of the reference.
If your output has swing of 400 mV, then the difference in amplitude
between driving a logic-1 and drving a logic-0 is 400 mV.
>
> - Logic 1 would defined as Vo+ at (VCM+400 mV) and Vo- at (VCM-400
mV)
Here you are making a masurement relative to some arbitrary level called
VCM. In this measurement you state that a when a logic-1 is driven, the
true output (+) measures 400 mV higher than this arbitrary reference.
Exactly where this reference comes from I do not know. Since I
believe you are describing an output you should be able to pick some
arbitrary reference point for your measurement. I would probably pick
ground for ease of measurement, though many CML drivers are internally
refrenced to VCC. Im guessing that if you pick just he right VCM
value, you are assuming that it will be in the middle of the HIGH and LOW
measurements for that outout of the differential driver.
> - Logic 0 would defined as Vo+ at (VCM-400 mV) and Vo- at (VCM+400
mV)
Aha! I was right (actually I peeked ahead). Here you state that when
driving a logic-0, the true output measures -400 mV relative to this
arbitrary reference. Now, if this is the case, then "SWING" is the
difference
between these two. Since one is +400 and the other is -400 (both measured
on the sme pin relative to this arbitrary reference voltage) then the
SWING (logic-0 to a logic-1) is 800 mV.
You also have the exact opposite taking place with the complementary output
of the driver. It also has a SWING of 800 mV. If you were to then
probe these two points with a differential probe, you would measure +800 mV
when driving a logic-1 as shown here.
Vo+ = VCM+400 mV
Vo- = VCM-400 mV
Vdiff = Vo+ - (-Vo)
= (VCM + 400 mV) - (VCM - 400 mV)
= VCM + 400 mV - VCM -(-400 mV)
= VCM - VCM + 400 mV -(-400 mV)
= 0 + 400 mV + 400 mV
= 800 mV
When driving a logic-0, the differential probe would measure a -800 mV.
Vo+ = VCM-400 mV
Vo- = VCM+400 mV
Vdiff = Vo+ - (-Vo)
= (VCM - 400 mV) - (VCM + 400 mV)
= VCM - 400 mV - VCM -(+400 mV)
= VCM - VCM - 400 mV -(+400 mV)
= 0 - 400 mV - 400 mV
= -800 mV
These are the signal amplitudes that you would see on a differential scope
(or on a single-ended scope using two probes with the scope set to display
A-B). When driving a logic-1, the scope would show an amplitude of +800 mV,
and when driving a logic-0 the scope would show an amplitude of -800 mV.
The Vp-p of this differential measurement is 1600 mV. However, your opening
paragraph defines this as being only 800 mV p-p differential.
IT CANNOT BE BOTH.
>
> - If the driver output VDp-p <85 mV you got noise.
>
> VCM- Voltage Common Mode
I do not see where this magical VCM number comes from. These are
differential
signals. In many cases thay cannot be seen from the local noise EXCEPT as
a differential signal. The only way they can be specified or measured is as
a differntial signal. This introduction of an undefined an non-measureable
VCM signal does nothing to improve the measurement, and definately does not
clarify the issue.
> VDp-p Voltage Differential p-p
A measurement of a differential signal when drving only a steady-state
signal only shows you peak voltage. It does not show p-p. This would
show as a horizontal line on the scope. A horizontal line shows
the peak voltage when driving either a logic-1 OR a logic-0. To see
the p-p you need to see BOTH a logic-1 AND a logic-0 (this requires
at least one transition in the data stream).
>
> o. During the time we defined Hari, we explicitly left out reference to
> VCM. This was due to interoperablity of different logic voltage levels
NOT CORRECT. VCM does not belong anywhere with respect to the MEASUREMENT
of a differential signal. A differential signal can and should only
be specified as signal relative to signal. The common mode spec you
are refereing to is a receiver specification indicating just how
much common mode range (or rejection) that receiver has.
> and technologies. For example most of the existing SerDes today, it is
> not possible to connect the driver directly to the same part receiver.
Sure you can. You just need to do it through DC-blocking caps. These
caps remove the DC-content (or offset) from the source signal. This
is exactly what a differntial measurement does; i.e., it subtracts out any
present offset and only shows how one signal changes relative to another
signal.
> At the same time there are some applications such as large
> switches which may be very desirable not to use AC-coupling caps.
You are correct. You also have some PHYs that are powerd by 5V, some at
3.3V, some at 2.5V, 2.2V, 2.0V, and hopefully soon 1.8V. Since many
of these outputs are referenced to VCC (thats how CML works), the DC
content of the signals will vary as the power supply shifts.
> In some of these applications you may end up using several thousands of
> caps on a backplane. In another standard I proposed the followings an is
> incorporated in to the specifications:
>
> - If the transmitter does not provide VCM of 0.75 V, then must add
AC caps
> - If the receiver can not tolerate 0.75 V of VCM, then it must add
AC caps.
Ali, you're acting like this VCM is some fixed voltage. It is anything but.
Probe some of these differential signals once, but do it as two
single-ended measurements. You will see all kinds of hash on the signals.
If you probe these same signals differentially, most (hopefully) of this
hash will dissapear.
All that hash is what the common mode rejection of the receiver is supposed
to take care of. Each receiver will have a center to its common mode
operating range. So long as the received signals remain within that range,
they are received just fine. As soon as they go outside that range, you
may start to see some nasties. This center-point in the common mode
operating range is where PHY vendors either provide internal DC-restoration
or they recommend an external biasing to that same point.
For most receivers, this point is NOT at 0.75V.
> Overall the above definition of VCM would not burden the implementation
> as it is optional and only high applications need it.
I most heartily beg to differ. The specification of a VCM in reference to
a differntial signal is an abomination. It totally confuses the entire
concept of differntial signaling. DIFFERENTIAL SIGNALS ARE NOT MEASURED
RELATIVE TO A REFERENCE. While you may have some desire to allow a
DC-connection between transmitters and receivers, please do not push this
desire into totally unrelated areas such as how to measure or
define a differential signal. Especially when the explinations do not
resolve to the common practive of how these signals are specified in
EITHER NCITS T11 (Vp-p differntial) or IEEE 802.3 (Vp differential)
standards.
> o. I have also had several comments on the drive amplitude. If we decide
> to define copper twin-ax cable of length 10-15 m, we would need to
> increase the amplitude somewhat. IB drive amplitude are higher than Hari.
> One option here might be:
>
> - Drive hari levels pre-emphasis Off
> - Drive IB levels turn pre-emphasis On
Since your present drive amplitiude numbers basicaly match that of the
existing T11 Fibre Channel and IEEE 802.3z Gigabit Ethernet specifications,
and in fact exceed the minimum (1100 mVp-p differential) by 500mV, you
have already specified a 3.2 dB amplitude increase over the present
minimum definitions. This additional 3+ dB will give you a significant
increase in your cable length.
>
> I hope this answers some of your question.
Actually, it doesn't (at least not for me). If possible, could you
please review my responses and comment on same. The area of
differential signaling can at times be confusing.
Regards,
Ed Grivna
Cypress Semiconductor
(952) 851-5046
>
> Thanks,
>
> Ali Ghiasi
> Sun Microsystems
> (650)786-3310
>