| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
|
Hi Charles, All for KX. The
6 I looked at had a problems with KX. They are: Peter T1, T20,
Tyco 1, 6, Molex in H2, Molex out h2 I have no
doubt that these channels will really work for KX. … Rich From: Charles Moore [mailto:charles.moore@avagotech.com] rich, When I talked to Charles I said this seemed to have
promise because the Goergen line has as strong sqrt(f) dependence and low
frequencies. It has basis in the skin loss physics. This would mostly affect KX
which we really didn't spend much time on. After all we just changed the ICR
fit line. However, I made these comments without really taking a close look at
our data. Isn't theory grand... J Since then I did look at the data. I had Steve
Krooswyk alter the spreadsheet s-parameter analysis tool (SCAT) to look at this
new fit. What I discovered was that it actually makes many channels appear much
worse. After further review of the data I think we could do much better by just
changing the confidence region lines for KX only. However, I think the data
needs to be presented and then we, as working group, can determine a plan of
action. …Rich Mellitz, Intel Corporation 803-216-2160 (o) 803-873-7343 (m) -----Original Message----- guys, Rich Mellitz and i have been discussing
changing the channel model, to use a fit to Amax, as we are now doing for the
interference tolerance test channel (see moore_03_1105.pdf), instead of a fit
to a straight line. This would have several advantages: 1. It would use a common method for
both the test channel and the working channel 2. It would allow us to replace
three graphical tests: i.
That the linear fit is better than Amax ii. That
the insertion loss is always better than Amax + something iii. That the
ripple is better than something with one numeric
test, that mc < 1.0, and the graphical ripple test. We
may need an additional limit on bc. 3. Since the fit to Amax will be substantially
better than the fit to a straight
line, the ripple will be smaller and more meaningful. This will reduce
the number of false fails. We should be able to shrink the
ripple bounds (Rich: will you figure out by how much?) which may
also reduce false passes. 4. Since the fit will be especially
improved at low frequency, we will be able to
reduce f1, especially for 10GBASE-KR, which actually uses low
frequencies more than either 1000BASE-KX or 10GBASE-KX4. Does anyone see any problems with
this proposal?
charles -- |--------------------------------------------------------------------| | Charles Moore | Agilent
Technologies | ASIC Products
Division | (970) 288-4561 |--------------------------------------------------------------------|
-- |--------------------------------------------------------------------|| Charles Moore | Avago Technologies| ASIC Products Division| charles_moore@agilent.com| (970) 288-4561|--------------------------------------------------------------------| |