Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

# Re: [802.3_4PPOE] Regarding adding perfectly balanced resistance to a cable measurement

Jeff,

Let's discuss it at San Diego, and I explain what I did and how to interpret the findings.

Soon you will see final work at the reflector that will explain all issues in clearer way .

The 16 use cases includes 4 of Wayne use cases (considered as realistic use case) and the rest are use cases around this point to perform sensitivity analysis. This points were selected with reason. They represent roots of polynomial _expression_ so I can check around the derivatives near the peaks. In this way I can find faster were to look for problems. I can do it with spice simulations too (We used it in most of our adhoc work me and Christian were it was most relevant to do) however when I need better visibility to specific issues I used different tools so I can activate different mathematical operators that in typical spice software are not available. In this work I had to write the software to do exactly what I want and present it in different angles.

Yair

From: Jeff Heath [mailto:jheath@xxxxxxxxxx]
Sent: Tuesday, July 08, 2014 1:02 AM
To: Darshan, Yair; STDS-802-3-4PPOE@xxxxxxxxxxxxxxxxx
Subject: RE: [802.3_4PPOE] Regarding adding perfectly balanced resistance to a cable measurement

Yair,

Simulations are math based on circuit theory.  As we are analyzing circuits they pspice is a great choice to understand current imbalance. J  We are all doing analytical analysis. I am using circuit analysis to analyze a circuit.

You were indeed adding a perfectly balanced 0.2 ohms to the resistance to each pair of the channel to reduce the percentage for most of the use cases to 7% or below (to fit your model which you have chosen 7% as a maximum.)  It is very clear that this was the point of what you presented. It is very explicit in slides 24 and 25. J

Note that on slide 22, of your 16 proposed cases for study, only 2 of 16 were close to 7%.

Don’t worry, I am not ‘bothering’ to do ‘long simulations’.  They go very quickly and are much more accurate than the math method you favor.  They are able to show the current imbalance, not the percentage resistance imbalance.

My proposal for resistance differences will indeed work for the PI definition and I have listed my simulation values for each of the components including a base resistance value used for simulation.  I was trying to avoid specifying a minimum PSE resistance as this is implementation.  Since this time, you have suggested that the PSE should have a minimum resistance of 0.1 ohms.

I have no problem with a minimum resistance if it is based on traceable assumptions but I suggest that this be brought up to the larger task force.  My lowest simulation value was 0.12 ohms on the high side so we are not far off there.  On the low side I have minimums of 0.24 ohms with DCMPS and 0.29 ohmes for ACMPS so we differ considerably there.  See PSE PI Values and Simulation Support slides where I show my assumptions for all of my PSE PI Values.  I would love to see how you arrive at your numbers.

Lastly, I have include all three components in my presentation in the ad hoc, the PSE PI, the PD PI, and the channel to obtain a maximum pair current which is my goal.  We are talking about the channel and my assumptions on the were included using the then current model of 7% or 0.2 ohms whichever is greater.  I am including Wayne’s A,B,C, and D use cases in my next presentation.

BTW, right now, it appears that Wayne A is the defining case for maximum pair current using the existing diode models.

Regards,

 Jeff Heath Design Center Manager paper: 402 East Carrillo Street, Suite D Santa Barbara, California 93101 voice: 805.965.6400 fax: 805.965.1701 computer:

From: Darshan, Yair [mailto:YDarshan@xxxxxxxxxxxxx]
Sent: Monday, July 07, 2014 1:59 PM
To: Jeff Heath; STDS-802-3-4PPOE@xxxxxxxxxxxxxxxxx
Subject: RE: [802.3_4PPOE] Regarding adding perfectly balanced resistance to a cable measurement

Hi Jeff,

Since our last discussion I have solved all my worries. I will send later presentation that doesn’t require test setup and doing the work with only two parameters (CP2PRUNB and resistance difference as in my motion from May 2014. We just need to update the numbers to the correct numbers that I have found in my work with my math.. J.

But first, Ill address your comments:

1.       We are discussing the channel and not something else. I do need math and not simulation, otherwise we will miss deep understanding of the issues.

2.       My math and data are accurate.

3.       I am not trying to fit the use cases to target unbalance limit by adding perfect resistance. I was trying to filter out the P2PRUNB peaks that was resulted from unrealistic use cases as irrelevant points, during channel compliance tests, to prevent false interpretation of 18% unbalance while it is when cable resistance is close to nothing which is actually the connectors unbalance and connectors are treated in the spec by other parameters. Meanwhile I have found a solution for that issue by simply finding the mathematical rule (J ) to verify that typical use cases with short and long channels and non-typical  typical use cases with short and long use cases are covered completely by our spec and will not be subjected to interpretations.

4.       Your proposal to use only resistance difference will not work. It will not work for the channel and not to PSE PI and PD PI as you presented in the adhoc. See my previous presentation (and also the new one which is shorter and clearer) for the reasons. If something mathematically can't be no reason to bother and doing simulations. This  is an example of what we can miss if not doing the analytical work first and doing simulation later. I had some quality time (1.5h) with Sterling after the adhoc meeting in which I show him the math and he agree with it.

We can discuss more at July meeting face to face.

5.       You said below that you have presented a way to deal with the entire imbalance current. Are you talking about your presentation at the adhoc? . Remember that our discussion was only about the channel not regarding the entire system. See also (4).

Best Regards,

Yair

From: Jeff Heath [mailto:jheath@xxxxxxxxxx]
Sent: Monday, July 07, 2014 10:57 PM
To: Darshan, Yair; STDS-802-3-4PPOE@xxxxxxxxxxxxxxxxx
Subject: RE: [802.3_4PPOE] Regarding adding perfectly balanced resistance to a cable measurement

Yair,

Your math does not represent the data.  Your solution is to try and make the real use cases under study for the channel fit a percentage model by adding a perfectly balanced resistance.

Math is great but when the math does not match the data the wrong math model was chosen.  The channel data use cases is not represented by your choice of math.

I have presented a way to deal with the entire imbalance current which is to represent the uses cases we agree on and do simulations which will give us the accurate answers based on our agreed upon use cases.

Regards,

 Jeff Heath Design Center Manager paper: 402 East Carrillo Street, Suite D Santa Barbara, California 93101 voice: 805.965.6400 fax: 805.965.1701 computer:

From: Darshan, Yair [mailto:YDarshan@xxxxxxxxxxxxx]
Sent: Wednesday, July 02, 2014 5:06 PM
To: STDS-802-3-4PPOE@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_4PPOE] Regarding adding perfectly balanced resistance to a cable measurement

Jeff,

You missing the point since you though I use Vdiff for my channel analysis which is not relevant.

We are talking about the channel.

We agree on previous adhoc to use one number for the channel, PSE PI and PD PI. This was based on long discussion and detailed work.

The channel always had one number. See IEEE spec and other standards. It true for pair unbalance  and it also true for P2P unbalance no difference in the concept.

Yes the channel is a part that its Runb percentage is changes at short cable/low resistance as function of connectors. This is known fact, we are prepared for it, no need to re-invent the wheel, we know what to do.

Yes I am insisting on one percentage number. This is the best way and nobody show better way for the last 20 years or more and I guess it will not be changed since it is follows the traditional definition that has also physical meaning to us. See again the ad hoc material al l>130 pages of it and presentation, read the standard and we can discuss more on it.

I have shown you the data how the percentage is changes with number of connectors and cable length, so you don’t need to explain it to me since I have also the solution for it.

If you have better way please suggest it.

Yair

From: Jeff Heath [mailto:jheath@xxxxxxxxxx]
Sent: Wednesday, July 02, 2014 7:21 PM
To: Darshan, Yair; STDS-802-3-4PPOE@xxxxxxxxxxxxxxxxx
Subject: RE: [802.3_4PPOE] Regarding adding perfectly balanced resistance to a cable measurement

Yair,

I am not missing the point, I am disagreeing with you point.  Perhaps it is you that are missing the point?  Let me expain it so you understand the point.

You are still insisting that one percentage number be used for channel resistance imbalance, yet the channel is a system where the percentage changes drastically depending on the length of the cable and the number of connectors.

Why do you insist on using a linear equation for a completely nonlinear data set?

Regards,

 Jeff Heath Design Center Manager paper: 402 East Carrillo Street, Suite D Santa Barbara, California 93101 voice: 805.965.6400 fax: 805.965.1701 computer:

From: Darshan, Yair [mailto:YDarshan@xxxxxxxxxxxxx]
Sent: Wednesday, July 02, 2014 7:26 AM
To: Jeff Heath; STDS-802-3-4PPOE@xxxxxxxxxxxxxxxxx
Subject: RE: [802.3_4PPOE] Regarding adding perfectly balanced resistance to a cable measurement

Hi Jeff,

1.      Yes I have invested effort in analyzing the channel to close the Channel P2PRUNB which is specified in the standard in [%]

in similar way that the pair unbalance is specified in IEEE802.3-2012 and per others standards.

2.      This is per our motion from May 2014.

3.      YOU Miss the point (again ..:J )  . This is channel analysis, the formal channel with up to 4 connectors SO NO VOLTAGE DIFFERENCE IS INVOLVED.

It is just computation of (Rmax-Rmin/(Rmax+Rmin).

3.1  I didn’t define the system components in term of percentage. They are all in ohms. I only calculate the effect of it which is our target objective C_P2PRUNB

4.      I didn't run to a problem due to specific technique. The only way to define channel is by P2PRUNB. Any technique you will suggest will have the same issues to solve.

Try it, and I will comment on it.

Moreover you explain of what causes the peaks (connector, discontinuities etc.) is repeating my explanation and not suggesting a solution. Try suggesting a solution in addition to my 4 proposals so we can have some common base line to discuss.

5.      Using Resistance Unbalance Tec technique is not new. Please read previous adhoc material and presentations. It is the technique we always used also in 802.3AT for pair unbalance. There is no better technique then using channel components parameter and specifying a single number. All technique will lead to the same results. Try to suggest different technique and we can discuss it.

I have shown it to Sterling after we has yesterday additional 1.5 hour of quality time.

6.      we are discussing the relevant use cases for 3 weeks now between the experts so it is not a new topic. The main issue is:

-There are 4 use cases that we agree that they are valid use cases (A,B,CD).

-IN this use case we saw peaks above what we want.

-Question 1: what to do with it? Accept it as it is and use higher limit than 7% or find justification to ignore it without causing other issues.

-There are other use cases that show much higher peaks. They are less or nonrealistic. The problem is how to justify ignoring it.

The high level problem as I see it as system vendor is: If we set the limit to 7% and somebody that is not expert and don’t understand what he is doing decided to use cable 1m+connector+1m because he wanted 2m and he has only two 1m cables. He do it and when testing the channel he gets 10%. What he will do? He will complain to me and I will wasting my time on it. Hope it is clear.

There are solutions for it being discussed now, but we need to close it.

7.      You ask " As an example of this, I believe you  have a 0.25m cable with 1 connector which was thought to be very unlikely.  If you really insist that anything should be considered, why not include a 0.25m cable with 4 connectors? "

My answer is simply:

(a)   It is practically may cause issues with channel performance so it will fail there.

(b)   It is not practical to insert 4 connectors in 25cm. Try it and you will see.

But to answer this without pre-assumptions the answer is:

(c)    And theoretically if I will check this use case I will get 25% peak. This is the worst case possible (mathematically the highest limit per the numbers I have used).

If I solve for >7% I solve it for all use case.

8.      You said "The channel is there and we should define and agree on the use cases we are to consider in this ad-hoc and accept the results in the current imbalance they create when simulated with the PSE and PD PIs and report the maximum numbers.

Yair: So you agree with me finally. This is what I am trying to do. Please read my presentation again and tell me which slides cause you to understand it differently than what I am trying to convey. By the way Wayne, Sterling and George knows and understand that this is what I am trying to do. They also understand what I am trying to solve here. And we still talking about it until we all happy.

Please see more below

Yair

From: Jeff Heath [mailto:jheath@xxxxxxxxxx]
Sent: Tuesday, July 01, 2014 11:27 PM
To: STDS-802-3-4PPOE@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_4PPOE] Regarding adding perfectly balanced resistance to a cable measurement

From: Jeff Heath [mailto:jheath@xxxxxxxxxx]
Sent: Tuesday, July 01, 2014 12:54 PM
To: STDS-802-3-4PPOE@xxxxxxxxxxxxxxxxx
Subject: Regarding adding perfectly balanced resistance to a cable measurement

Hi Yair,

You have invested a lot of effort in using percentage resistance imbalance.  I believe your explanation today was:

1.    Use simulation of resistance and voltage offsets to create imbalance current numbers

2.    Convert this data to a resistance imbalance percentage for the worst case of the simulation(s)

3.    Define the system components in terms of percentage

Yair: No this is not what I did. See above.

----

Using this technique you have run into a problem.  The cable resistance is not well suited to using a percentage because it depends on a function that has discontinuities in it.  These jumps in percentage are due mainly to the addition of connectors.  Since this ‘breaks’ your desire to have a system components that are defined by percentages, you have decided to add some resistance, perfectly balanced resistance in sufficient quantity to a cable measurement so that the percentage falls below a number you have chosen (I believe that number for the cable is 7% right now in your presentation).

Yair: Not correct. See above.

This new technique has several problems:

Yair: This is not new technique. It is the technique we always used also in 802.3AT for pair unbalance. There is no better technique. Al ltechnique will lead to the same results.

I have shown it to Sterling after we has yesterday additional 1.5 hour of quality time.

---------------------

1.    It does not represent the actual cable

Yair: It represent the actual Channel. It my not a realistic use case which is different question.

2.    You suggest that the resistance you are adding is already present in the PSE and PD but

Yair: This is one of the proposals and it is one of the best.  There are other proposals to filter the unrealistic behavior of a standalone channel that I am working on.

a.    These resistances are not in the channel

Yair: Correct, They emulate the correct behavior of the channel in "minimum resistance" end to end channel while the peaks on the channel will be replaced by the peaks of the PSE and PD PI which will be higher. So we can justify ignoring the channel peaks when the channel is tested as standard alone.

b.    These resistances are not balanced

Yair: This is correct, but not important due to the fact that it is channel test and not end to end tests. It just to filter channel peaks NOT END TO END PEAKS.

Also, during the meeting it was brought up that some of the use cases you created for the channel should be discussed because they are extremely unlikely.

Yair. Yes we are discussing it for 3 weeks now between the experts so it is not a new topic. The main issue is:

-There are 4 use cases that we agree that they are valid use cases (A,B,CD).

-IN this use case we saw peaks above what we want.

-Question 1: what to do with it.

-There are other use cases that show much higher peaks. They are less or nonrealistic. The problem is how to justify ignore it.

The high level problem is: If we set the limit to 7% and somebody that is not expert and don’t understand what he is doing decided use cable 1m+connector+1m because he wanted 2m and he has only two 1m cables. He do it and when testing the channelhe gets 10%. What he will do?

There are solutions for it being discussed now, but we need to close it.

To that end they should be discussed and a consensus reached.  One of your arguments was that anybody could do anything which is not a practical approach to solving imbalance if it is not at all likely.

Yair: I disagree. The difference in opinion between us is due to the fact that I am system vendor and you as a chip will have less concerns.

As a system vendor if I know of potential issues even it has low probability ,I can't ignore it. The cost of ignoring may be very high for me.

You/we need to do a lot of analysis and verifications in areas  that we are not experts to be sure that we are not causing future issues.

Cabling experts are familiar with the "normal" use cases, there are other use cases in the field in strange place.

------

As an example of this, I believe you  have a 0.25m cable with 1 connector which was thought to be very unlikely.  If you really insist that anything should be considered, why not include a 0.25m cable with 4 connectors?

Yair:

(a)    It is practically may cause issues with channel performance

(b)   It is not practical to insert 4 connectors in 25cm.

(c)    And theoretically if I will check this use case I will get 25% peak. This is the worst case possible (mathematically the highest limit). If I solve for >7% I solve it all.

The channel is there and we should define and agree on the use cases we are to consider in this ad-hoc and accept the results in the current imbalance they create when simulated with the PSE and PD PIs and report the maximum numbers.

Yair: So you agree with me finally. This is what I am trying to do. Please read my presentation again and tell me which slides cause you to understand it differently that what I am trying to convey. By the way Wayne, Sterling and George knows and understand that this is what I am trying to do. They also understand what I am trying to solve here.

Regards

Yair

Regards,

 Jeff Heath Design Center Manager paper: 402 East Carrillo Street, Suite D Santa Barbara, California 93101 voice: 805.965.6400 fax: 805.965.1701 computer:

• References: