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

Re: [8023-POEP] Class signatures

Steve and all,

 

I review it and in general it looks reasonable.

I have few comments:

a) For future use we have to cover the range of 30 to 50W (max limit of 100W for 4P cable).

    We can take some margin on the upper limit say 10% so e may need to cover the range of 30 to 45W.

   

    In addition we have to allocate those future classes and defined them right now so vendors will not use them for proprietary implementation.

   

b) 2 fingers ping pong: We have to define the requirement for 3 finger ping – pong and use 2 now or what ever we decide so a future use of 3 fingers will not be blocked by timing issues in multi port system or other issues. Again for future use but defined properly.

 

c) In the classification table I see classes with less then 20% difference between them.  We can use larger space based on power measurement accuracy limitations which is 10% for current and 10% for voltage resulted by ~20-21 % and we may get few spare codes for the upper future range.

 

We can afford this due the use of layer 2.

 

Yair

 

From: owner-stds-802-3-poep@xxxxxxxx [mailto:owner-stds-802-3-poep@xxxxxxxx] On Behalf Of Steve Robbins
Sent: Monday, June 12, 2006 9:09 PM
To: STDS-802-3-POEP@xxxxxxxxxxxxxxxxx
Subject: [8023-POEP] Class signatures

 

Hi all,

 

This email is intended for the classification adhoc (is adhoc one word or two?) but comments are welcome from all.

 

The subject is to propose the new class levels for the extended L1 protocol, as part of the on-going discussions.  I’ve tried to take all the consensus we’ve reached so far, and put them together into some sort of organized synthesis.

 

GIVENS:

  1. The low end of the power scale shall be 2W.  (Per the vote at the Austin meeting.)
  2. The number of codes shall be 10.  (Also per a vote in Austin.)

 

ASSUMPTIONS:

  1. We use the ping pong method with 2 “fingers”.  (FYI, a time-based protocol hasn’t been ruled out yet.  I’m just using ping pong for this analysis.)
  2. The top of the power scale is 35W on 2P (70W on 4P which consists of 2 independent 2P systems).  I chose this level high enough to be somewhat “future proof” (I hope) but low enough to avoid excessive granularity.  (This max power level could easily change.  The group needs to discuss it.)

 

BRIEF REVIEW OF THE “PING PONG” PROTOCOL

The ping pong method is essentially a series of AF classification cycles, performed consecutively.  In this example, the series consists of only 2 cycles.  On each cycle an AT-PD may present a different class signature, while an AF-PD maintains a fixed class signature.  After sampling the current level of a class signature, the AT-PSE signals the AF-PD to change to the next class signature in the sequence by briefly moving the voltage outside the valid classification range (the voltage level and timing are still TBD).

 

ANALYSIS:

  1. Each class signature, or “finger”, may have 4 possible values: Class 1, 2, 3 or 4.  The current levels are unchanged from 802.3af.
  2. Two such fingers yield a total of 16 codes.
  3. For backward compatibility, the following code pairs are fixed because they are the codes presented by an AF-PD:
    1. <1,1> = 4W
    2. <2,2> = 7W
    3. <3,3> = 15.4W
  4. These fixed codes were each placed at the top of a power subrange, as shown in the table below.  (Power subranges are separated by the horizontal lines.)
    1. This maximizes the number of new codes in between the fixed codes.
    2. This also created some codes below 4W, to meet the goal of classification down to 2W.
    3. Another reason for organizing the codes in this order has to do with interoperability.  Suppose <2,2>=7W and <2,3>=10W.  Then, an AT-PD requiring 10W would be allocated only 7W when connected to an AF-PSE.  To avoid this problem, the <n,n> codes must be at the top of each subrange, at least for the bottom 3 subranges.
  5. The new codes were assigned power levels according to the following:
    1. The top power subrange uses a linear scale with codes at 5W intervals.
    2. The bottom three power subranges used exponential scales.  The fixed codes define the boundary conditions for each subrange.  Therefore each subrange uses a different exponential equation to fit the new codes.  (Overall the system could be called “piece-wise exponential”.)
  6. To reduce the number of implemented codes to 10, some codes were marked “reserved”.  My opinion is that these codes should NOT be marked “undefined” or “forbidden” because some vendors may try to use them to encode proprietary information.  In general I tried to reserve every other code, so that the remaining codes were still distributed fairly uniformly on a log scale.
  7. The codes for the second finger column may look somewhat random, but they’re not.  The lower three numbers for each subrange could be arranged arbitrarily, but I assigned 3 and 4 to the reserved codes, so that the implemented codes would hopefully produce less heating in the chips.  (But this is probably a minor consideration.)
  8. For the bottom subrange I set the lower boundary at 1.4W instead of 2W for three reasons:
    1. I felt there should be one reserved code in the bottom subrange, so that the top subrange could have 3 implemented codes.  (Only 10 codes are allowed, so they need to be used wisely.)
    2. It made sense to put the reserved code at the bottom of the subrange so that the three remaining codes could be equally spaced on the log scale.
    3. This preserves the option of going below 2W if we ever need to.
  9. In the top subrange the highest power level is reserved, because it’s the least likely to ever be used.

 

RESULTS

The 10 codes that are not marked “Reserved” are implemented.

 

First Finger             Second Finger              Power Level (W)                        Notes  

4                                  4                                  35                                 Reserved

4                                  3                                  30

4                                  2                                  25

            4                                  1                                  20                                            

3                                  3                                  15.4                              FIXED

3                                  4                                  12.6                              Reserved

3                                  1                                  10.4

            3                                  2                                  8.5                                Reserved

2                                  2                                  7.0                                FIXED

2                                  3                                  6.1                                Reserved

2                                  1                                  5.3

            2                                  4                                  4.6                                Reserved

1                                  1                                  4.0                                FIXED

1                                  2                                  2.8

1                                  3                                  2.0

            1                                  4                                  1.4                                Reserved

 

     NOTE: An AT-PD cannot present a Class 0 signature.  Only AF-PDs can be class 0, and it would be assigned 15.4W as per 802.3af.

 

I hope everyone finds this acceptable.  Again, it’s not set in stone, and comments are welcome.

 

Steve