2

What are the minimum handshake elements required to get 12W out of a PoE switch (with 8P8C modular connector) in a compliant way? After some basic but time-consuming research, I found the following:

  • 25 kohm detection resistor
  • do not connect load until line reaches 21V (maybe even twice that for true compliance)
  • keep load above 10mA after that ("maintain power signature")

So far as I can tell from experiments people have posted, skipping any of these will cause the handshake to fail with any major brand of active PoE switch. For example this guy only connects the 25 kohm resistor, and gets unstable voltages and blinking status lights. Presumably because of missing "maintain power signature"... which isn't even discussed in the PoE wikipedia article.

Besides those listed above, are there any other elements required to make a PD compliant, operational, and/or reliable? And for whatever list is determined, could one estimate or bound the number of pages of IEEE documents one would have to read in order to confirm its completeness?

personal_cloud
  • 407
  • 5
  • 11
  • 2
    An industry/ manufacturer minded person would use dedicated poe PD ics to handle this in a standards compliant way. Trying to homebrew it is... not something that lends itself to wide compatibility or scaling. – Passerby Sep 13 '23 at 04:22
  • 2
    @Passerby I understand that some folks prefer to think in terms of dedicated ICs. That is fine, but the question still applies as phrased. (Suppose we wanted to design a minimal dedicated IC. What would it need to contain?) – personal_cloud Sep 13 '23 at 04:25
  • "which isn't even discussed in the PoE wikipedia article." - Yes, it is. 5th paragraph under "Powering devices". – Finbarr Sep 13 '23 at 10:41
  • @Finbarr Ah yes, there it is, buried under piles of irrelevant stuff. I wonder if the Wikipedia editors would consider making it a more prominent section labeled "maintain power signature", given that it's 1/3 of the whole protocol. – personal_cloud Sep 13 '23 at 19:49
  • @personal_cloud Anybody can edit Wikipedia. If you think it's better organised differently, just fix it yourself. – jonathanjo Sep 14 '23 at 09:43

2 Answers2

1

POE has a lot of subtleties to it and special cases. The reason it's tricky is because many of the cases are tricky and/or potential dangerous. If you're just playing about on the bench, play away. If you're thinking of designing a powered-device controller, read the standards documents (and not Wikipedia, excellent though it is.)

As a comment on your question: asking "how much do I have to read to do task X" is an excellent engineering question: it makes a numerical estimate of the complexity of a task, for a person with the education to read those documents.

Could one estimate the number of pages of IEEE documents one would have to read?

The standard 802.3-2022 is about 7,000 pages long. The parts you need are perhaps 500 pages.

You might care to start with these introductions, which are about 50 pages each. The second one probably covers your exact material the best.

IEEE Standards: The documents you want are currently the 2022 version of 802.3, which is about 7,000 pages, and the corrections-and-clarifications amendment about POE, about 30 pages. You can read all except the most recent standards for free at IEEE's Get 802 program by registering at its site.

  • 802.3-2022 - IEEE Standard for Ethernet
  • 802.3dd-2022 - IEEE Standard for Ethernet Amendment 1: Power over Data Lines of Single Pair Ethernet

PS. It's remarkable to make the comparison with the Internet protocol documents which are extremely short in comparison: IPv4 is defined by RFC 791 at 44 pages, UDP by RFC 768 at 3 pages, and TCP by RFC 791 at 84 pages, ancillary documents about corrections, checksums, special cases, DNS and whatever, perhaps 300 pages.

jonathanjo
  • 14,034
  • 3
  • 34
  • 66
  • Can you explain how 802.3bt is useful for a 12W PD? Good point about 802.3dd - I hadn't specified the medium. The question now specifies 8P8C modular connector. – personal_cloud Sep 13 '23 at 19:56
  • The Ethernet Allication document I listed for 802.3bt gives a very good overview about the range of POE behaviours and interoperabilities, with better diagrams. You just skip the parts which don't apply to whatever you're designing. It's very useful to keep your eye on later standards to make sure you're not doing something which will be a problem with later devices. Especilly if you're in any way trying to skip parts of a standard. – jonathanjo Sep 14 '23 at 09:41
  • That's a good point, that later devices might introduce incompatibilities. However, the Ethernet standards are pretty good about backwards compatibility. I use plenty of devices that were designed decades ago. Also I agree that the 802.3bt doc is better written than the 802.3af standard. However the doc you linked is a bit high-level; I'm not sure one could actually build a working device from it (e.g., it mentions several options but doesn't describe how to know which options are enabled). It does however give a blessing for a simple 12W PD similar to as described in the question. – personal_cloud Sep 17 '23 at 03:56
  • Agreed: certainly couldn't implement against those docs. But overviews like those are very helpful when reading the actual standards. – jonathanjo Sep 17 '23 at 13:21
1

A 12W PD only needs to support 802.3af. (Sadly the document is not written in a focused way nor does it provide simple examples. It is loaded with extra text that neither clarifies nor improves readability nor adds information. Probably because it was written by a committee.)

Sections relevant to the question are about 50 pages, as follows. The question asks what requirements are missing from the list, so I've put those items in bold.

  • 33.3.2. Pinout and 25 k ohm signature defined. And state diagram for the PD. With the required steps being present signature, draw power. Gotta love IEEE for making a whole diagram out of this. OK nothing that's not already noted in the question here.
  • Annex 33A warns not to connect anything other than 15 k ohm while detection is occurring. With lots of examples of things you can connect to a circuit that aren't a 15 k ohm resistor. Already noted in the question.
  • 33.3.6 Specifies a 10mA minimum load for maintain power signature.
  • 33.3.2 limits PD input inductance to 0-100uH. Presumably for safety reasons. Common sense.
  • 33.3.5 Defines the 36-57V operating range, cable losses, etc. And warns not to backfeed by 2.8V more than what you receive -- but only if you are receiving 57V. In other words, don't apply more than 59.8V to the line.
  • 33.3.5 defines powered-state hysteresis: turn on at 42V, turn off at 30V. However the device is also allowed to assume it gets a minimum of 36V while powered, so the most reasonable interpretation is turn on at 42V, turn off at 36V. The point being to avoid startup oscillation.
    1. Adds a full electrical isolation requirement. And 33.4 specifies 1500V hi-pot tests.

Sections that are not relevant to the question:

  • sections 22, 30, 33.6, etc. define management software. Not a requirement for the PD.
  • 33.1 - 33.2 apply only to the PSE side.
  • 3.3.4 - classifications... optional if the goal is 12W.
  • 33.3.2 - Limits input capacitance to 50-120nF. And 33.3.5 Gives some peak current values that are higher than the averages, in case you need inrush or ripple current. But then both of these are effectively walked back in 33.3.5.5: "there is no max capacitance".
  • 33.4 short circuit tests - adds nothing to 802.3, at least on the PD side.
  • 33.4.3. Gives a minimum common-mode impedance formula that doesn't seem right (try plugging any reasonable values into it...) and therefore is likely not implemented by anybody.
  • 33.4.4-7. Gives some common-mode output specs for 1MHz+. Doesn't add anything to original 802.3.
  • All the midspan PSE stuff. I'm not a PSE.
  • 33.3.6 Specifies a 50nF minimum capacitance for maintain power signature. However, the standard doesn't describe how this capacitance is actually measured, what frequency to measure at, etc., so it's likely not implemented. (The standard says make two current measurements at voltages 1V apart, suggesting that the PSE only needs to measure resistance.) The 50nF capacitor is not present in many PoE circuit diagrams available online. (Probably not just because it's not really required, but also because your PSU will probably already have one, and will engage to the lines when current is present)
  • 33.5 - general safety. Yes, be safe and label the port. Doesn't add anything to 802.3 apart from labeling required as discussed above.
  • 33D. Specifies a 30 ohm minimum load impedance for the PD, so the PD-cable-PSE loop doesn't ring. This applies "up to the PD power supply crossover frequency". (Somehow I doubt this was meant to apply to DC, as that would give up 2.7W.) In any case this section is labeled as just "recommendation". I think the intent is (1) don't introduce ringing and (2) test with a short cable and make sure there's no ringing. If providing the recommended 50nF input capacitor, then you'll need only, say, 1uH of input inductance, to keep the resonant frequency 3X below any of the operating frequencies (2MHz+). Unfortunately the IEEE document doesn't really explain this.
  • 33.5.9 Label the device as "44-57V 100BASE-TX PD". This is labeled as "recommendation".
personal_cloud
  • 407
  • 5
  • 11
  • I see you've using the clause numbering of 802.3af-2003. Can I suggest you renumber for 802.3-2022 for several reasons: 1/ it's the current standard; 2. it's easily available for free at IEEE; 3. Very little equipment currently in the field was designed against this 20 year old superseded standard. – jonathanjo Sep 14 '23 at 10:40
  • May I suggest you remove your sarcasm from this answer? It is extremely easy to throw scorn on a standard, written by a committee as you note. Yes: maintained over 35 years by hundreds of experts in all the areas required, giving us one of the most important public standards in engineering; this work is much harder than you'd believe. For example, if you don't understand why the standard uses 2.7W for something, say so. Suggesting they are stupid for saying it reflects on you, not them. – jonathanjo Sep 14 '23 at 10:48
  • @jonathanjo. I've reworded the offending sentence. However, I don't see why we should give the committee the benefit of the doubt on the "high impedance below frequency X". It really just seems like a typo. The way to fix ringing is keep the SMPS frequencies above the resonances. The text in the standard doesn't show a clear understanding of this. So let's just uplevel this, take their point to deal with ringing, and call it a day. – personal_cloud Sep 17 '23 at 03:41
  • Good notes, upvoted. Tag your opinions as such and it would be even better. Happy to call it a day. What does "uplevel" mean? Kind regards. – jonathanjo Sep 17 '23 at 13:18