| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
|
Hi
Rich,
It
appears you are right onto one of the issues that highlights why GroupDelay is
not the proper quantity of interest ... namely "you can't add
derivatives".
Looking at your PhaseDelay plot, it looks like you used
wrapped phase (the result has a sawtooth shape). To calculate phase delay
you need to use unwrapped phase.
Attached is a set of Magnitude, PhaseDelay an
PhaseDelayDistortion plots for a series of (mathematically modeled) lenghts of
low-grade FR-4 (6", 13", 18", 24" & 30"). Here PhaseDelay grows linear
with distance as you would expect so as you desire, 'the whole appears to be the
sum of the parts'. Also when normalized for PhaseDelayDistortion (the true
quantity of interest), the distortion grows linearly with distance. These
results are just using a model so need to be verified with
measurement. The model also does not have significant reflections.
Significant reflections may do interesting things to the PhaseDelayDistortion at
specific frequencies.
The
values for PhaseDelayDistortion are in the 10's to 100 ps. This is
significant relative to our data rates and jitter budgets. The actual
mapping to jitter is a complex function. Thankfully, equalizers can
equalize out PhaseDelayDistortion. The amount of PhaseDelayDistortion
encountered in real channels will effect the complexity of equalizer needed (a
future group discussion). The immediate issue is seeing if switching from
GroupDelay to PhaseDelay will eliminate the disagreements over measurement
BW/averaging and allow us to make faster channel
characterizations.
Regards,
Brian
Brunn
Channel Adhoc attendee
-----Original Message-----
From: Mellitz, Richard [mailto:richard.mellitz@intel.com] Sent: Wednesday, August 04, 2004 5:48 PM To: Brian Brunn; STDS-802-3-BLADE@listserv.ieee.org Subject: RE: [BP] Wed August 4th Agenda: 802.3ap Channel Model Adhoc Conference Call Hi
Brian I under
that different group delays at different critical frequencies directly effects
jitter (eye closure and distortion) on a specific channel. It is also a
sensitive measure of one of component of jitter. I tied to make this work before
for compliance channel on another project but discovered you can’t add
derivatives. I can loosely bound a system loss by combination of losses of
the components. However, I found many cases where the group delay of chain of
channels wasn’t even close to a sum of the component channels. So I agree the
Group delay should be deemed questionable at best. Have you
done a similar analysis for Phase Delay? Is the sum of the parts equal to the
whole? Just for
kicks I plotted Phase delay (-phi/omega) for two radically different channels.
You can see the mag il response of the two channels in the attachment as well as
the PD. I was trying to get feel how we used group delay. Could you help
me here? … Rich
Mellitz, Intel Corporation Ad hoc
meeting attendee From:
owner-stds-802-3-blade@listserv.ieee.org
[mailto:owner-stds-802-3-blade@listserv.ieee.org] On Behalf Of Brian Brunn Hi
All, Attached is my
recommendation that we look at changing our quantity of interest from Group
Delay to Phase Delay. The criticality of
delay distortion is debatable. However, I want to get this out
quickly because switching to Phase Delay may be the right thing to do *and* help
eliminate some of the repeatability problems people are seeing when taking
channel measurements. Someone on the call mentioned that magnitude
repeatability was fine and that the phase was the
problem.. Regards, Brian
Brunn |
PhaseDelayDistortion_040805.pdf