|Thread Links||Date Links|
|Thread Prev||Thread Next||Thread Index||Date Prev||Date Next||Date Index|
(hopefully this will make it through the filter, we think we’ve found the offending words…”
This is the first part of Bill Jones’ & Xiaopeng’s discussion, which part 2 made it through earlier.
From: William Jones
Sent: Wednesday, January 15, 2003 4:08 PM
Subject: channel model (part 1)
I agree with you that the frequency domain model is not sufficient to
determine the practically achievable performance of the receiver. However,
the frequency model can give us a quite good estimation of the best
achievable performance of the best receiver. The real-world implementation
can only perform worse than the theoretical limit. Since the phase
information is not available right now to my knowledge, by using available
frequency-domain model only, we can at least do some channel throughtput
estimation instead of waiting. For example, the question about whether the
throughtput of 10Gbps can be supported by 100m CAT-5E channel based on a
well-accepted channel/noise model has not been answered yet.
My suggestion is to put the time-domain model aside for our discussion now,
and make the frequency-domain model clearly clarified and available to all
people, then we can determine what is the best thing we can do in this
standard study group.
I would not support the use of channel and noise models based on the DSL
spectrum management standard. In the case of the insertion loss, the RLCG
model parameters sited in table A.1 are for 26-AWG cable. Cat5 cable is
24-AWG. The Next and Elfext crosstalk models were developed for the
(DSL) loop with many pairs in the binder and would not be appropriate
for the in cable crosstalk we are considering. In particular, there is no
justification for n^.6 which was derived empirically from measurements in
that application. If any thing, it may be applicable to the alien
crosstalk. Finally, these types of models would have use limited to
frequency domain calculations. To determine the practically achievable
performance of finite length equalizers and cancellers, we will need time
domain models. In this regard, I agree with Chris that the models should
be derived from amplitude and phase measurements. Phase information is
very important in capturing group delay distortion and related effects.
For the transformer model, I propose a first order high pass filter. For
the low frequency 3 dB cutoff frequency, I suggest 12.5 MHz.