Thread Links | Date Links | ||||
---|---|---|---|---|---|
Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
Shannon, I think the entire group is under the realization that we need to entertain all possible solutions to make this work. My voting no at the last straw poll was related to the undefined changes for the rest of the curve. As I have stated before, I am of the opinion that too much is being taken away from the lower frequency in the proposed curves from Charles, which will make the channel more difficult to meet and hold down there.
As far as your perspective, you guys are definitely the closest to it. No doubt about that. However, I will point out that the presentations you have pointed to from what I understand are not necessarily from KR solutions. Transmitters were either test equipment or 2 tap. That was observed and changed in the EIT test spec, right? I think Joe’s work is much more convincing at this point.
As far as your comments regarding the channels, there are n+1 different sides to that story, isn’t there? Channels submitted were in accordance with what was felt could be fabricated and assembled while trying to maintain as broad a list of possible materials as possible. In addition a range of channels were provided to try and test various test conditions.
Regarding improvements to be had, the same thing could be said regarding the packaging. I look at the return loss of the file sent out by Charles and what Joe showed in his presentation, and think,”Wow the package is having a huge impact on self interference.” The package is a difficult issue though isn’t it? Well so is a system, given the vast array of compromises that need to be made, and by the way trying to maximize the total system density while at the same time trying to keep cost to a minimum as well. I believe I have heard similar comments made regarding the issues related to putting this stuff into a chip right?
Depending on what side of the fence you are sitting on, the compromises seen from that perspective will always seem like the most insurmountable. I will say this though – making a single channel work to meet further constrained channels is far more simplistic than doing it in a system environment.
My comment regarding the 0.75m channel was intended as an observation of your -20dB channel for 1m. I believe from the blade server people we saw 31” to 34” for most applications and some claiming to need up to the 1m. I do not believe we should consider changing the objective yet. There are things still being considered by some. From a channel perspective -24 is better than -23 is better than -22.5… etc. And don’t forget the lower end as well.
My two cents.
John
-----Original
Message-----
John, et al,
I’m not going to be in San Diego, so I thought I’d better clarify my position on the proposed channel insertion loss values. My intent was to show support for reducing channel insertion loss. Not necessarily all the way to 20dB@5.16GHz, but less than 26dB, for the following reasons.
In my opinion, Charles, Matt, and I have the most accurate perspective of the problem since we’ve actually performed and published EIT testing on real transceivers. For KX, KX4 and KR testing, Charles and I used a channel that closely matched the Goergen channel (26dB@5.16GHz) and saw crosstalk tolerances of 450, 180, and ~2mV@BER 1E-12 respectively. The power summed crosstalk from Tyco, Molex, and Intel NEXT, FEXT shows 20-100mVp@BER 1E-12 aggression to the thru signal, so the crosstalk tolerance limits are in the ballpark. Matt found that crosstalk tolerance results on a ~23dB@5.16GHz channel were ~30mV below the 20dB channel for KR operation. EIT test results documents are here: http://grouper.ieee.org/groups/802/3/ap/public/jun05/sawyer_02_0605.pdf http://grouper.ieee.org/groups/802/3/ap/public/sep05/sawyer_01_0905.pdf http://grouper.ieee.org/groups/802/3/ap/public/sep05/brown_01_0905.pdf
Real measured data is a compelling argument, and so far we have seen, 4 out of 4 vendors (2 unpublished) that are either borderline or failing EIT (last Fall’s definition) testing for KR. The 802.3ap channels’ PCBs and connectors were designed years ago for ~XAUI data rates, and appear to be the best opportunity for improvement (larger signal with reduced PNA, hence reduced DCD) in this system to enable KR operation.
Is there a compelling argument against a 0.75m channel?
-Shannon
From: DAmbrosia, John F
[mailto:john.dambrosia@TYCOELECTRONICS.COM]
Shannon, Technically speaking only Cases 1,2,and 3 fail the 20dB spec at Nyquist.
You need to be careful in making a statement like going to a 20dB channel. As you pointed out many channels failed that number, but I believe those are all the 1m channels or stub limited channels that are failing it. The channels that are meeting that the number in general are the 0.75m. The potential ramifications could be that to meet the 20dB channel we consider changing our reach objective to 0.75m.
John
-----Original
Message-----
Charles,
The difference (no pun intended) shows up in the differential vs single ended S-Parameters. I plotted SDD21 and S21 for modITTC23withCoupler.s4p, and there is 1.428dB difference at 5.160GHz. See attached.
Sorry I had to be at another meeting at noon, but I like the idea of going to a 20dB channel. In my opinion that’s the lever big enough to get a real system working. Unfortunately several (8 Molex, 4 Tyco, 3 Intel) channels violate that SDD21 up to 5GHz. See attached.
-Shannon
From: Charles Moore
[mailto:charles.moore@avagotech.com]
guys, -- |--------------------------------------------------------------------| | Charles Moore | Avago Technologies | Image Solutions Division | charles.moore@avagotech.com | (970) 288-4561 |--------------------------------------------------------------------| |