Re: [8023-POEP] 2P v 4P and safety
Hi All,
Just a couple of comments on the imbalance issue. There's 2 cases to be
concerned about:
1) The current balance within a pair, which affects transformer
performance. This case also relates to the issue of fault conditions
causing high currents in a single wire.
2) The current balance between pairs in a 4-pair system.
In case 1, sensing and compensation must occur at the transformer
end-taps. High impedance sensing could be done fairly easily, and could
perhaps be integrated as a low-cost solution. Compensation on the other
hand would be expensive and complex, in part because filtering is
required to maintain data integrity.
In case 2, sensing and compensation can be performed at the center tap.
Compensation for this would be much simpler because the filtering
requirement is removed, however the increase in cost versus 2-pair still
must be considered, and there are control issues for fault conditions,
etc. Another consideration is that the balance requirement in case 2
can be much more relaxed than in case 1. If there's a concern about
heating in the compensation circuit or in the overall cost of it, a
second alternative might be to simply shut down the port once a balance
limit is exceeded.
Regards,
Ken
Steve Robbins wrote:
> Wael and all,
>
> In regard to the question, should a 4P PSE = 2 independent 2P PSEs, I
> got an interesting question from someone a couple of days
> ago, suggesting an application I hadn't thought of.
>
> Basically, the guy was asking the following: Suppose you had a 4P
> PSE. Could he make some sort of passive splitter that would allow him
> to power two independent 2P PDs? It sounded like the splitter he was
> thinking of was just some RJ45 connectors and some wire. He told me
> there is already some similar gadget on the market (from Siemon?) that
> allows data to be split somehow, and he was wondering if power could
> also be split.
>
> Of course I told him I didn't know the answer to that, because a 4P
> PSE hasn't been defined yet. But obviously, if you combine an endspan
> 802.3af PSE with a midspan 802.3af PSE, then you have a system you
> could split easily.
>
> I don't know what the specific application is, or if anyone else would
> be interested in this. But it might be something to think about in
> the 2P vs 4P debate.
>
>
>
> The main problem with defining 4P=2x(2P) is that it basically forces
> the PD to have two separate PD controller circuits, and two separate
> DC/DC converters. This increases the cost of the PD, which
> understandably, the group is loathe to do.
>
> But the alternative (using diodes in the PD to passively combine the
> power from all 4 pairs, into a single DC/DC converter) is messy too.
> As Clay Stanford from Linear Tech showed in his presentation at the
> Austin meeting, you'd have to actively balance the currents at the PSE
> end. Otherwise, small differences in wire resistances and/or diode
> drops could lead to big pair-to-pair current imbalances. So it
> appears this approach makes the PSE significantly more complex and
> costly, which the group is also loathe to do.
>
> The lack of an attractive 4P solution is the main incentive for
> considering high-current on 2P. But then of course you get into other
> issues.
>
> That's basically where the debate stands right now, with no clear
> answers yet, although most people have an opinion.
>
> Steve Robbins
>
> -----Original Message-----
> *From:* owner-stds-802-3-poep@xxxxxxxx
> [mailto:owner-stds-802-3-poep@xxxxxxxx]*On Behalf Of *Wael Diab
> (wdiab)
> *Sent:* Friday, June 03, 2005 5:03 AM
> *To:* STDS-802-3-POEP@xxxxxxxxxxxxxxxxx
> *Subject:* Re: [8023-POEP] 2P v 4P and safety
>
> Hi Steve, Yair, Chad and all,
>
> Ive been trying to catch up on this thread, which brings up many
> good points. One thing that I am not sure we have exhausted yet is
> Steve's original point #6 (repeated below for convinience so that
> you dont have to scroll through the entire thread).
>
> I dont understand why that would be more complex than just picking
> one. Today, in .af, the PD has to accept power on either pair.
> Increasing power via a 2P scheme would be similar, with hopefully
> enhancements to the classification to differentiate it from leacy
> 2P power. Additionally, the PD may have the option to be in yet a
> higher power class and draw power over all 4P. In that case we
> would have to address the balancing issues that were touched on
> (but we would have to do that anyway if we went with 4P regardless).
>
> So we have PSEs that can support a higher power mode over 2P and
> others that can go further and support even more power over 4P. In
> my mind these could address different needs in the market and I
> would hate to have to just pick between one power point and the
> other unless I really have to.
>
> As an aside, I think the questions of whether or not 4P = 2x(2P)
> and how much power can be drawn over 2P max and 4P are both good
> ones, but I would be interested in anwering these first before
> pre-empting which scheme we go with and/or whether we allow both
> options in the standard.
>
> here is the original point so that you dont have to scroll all the
> way down
>
>
> 6. Suppose PoE Plus allows both 2P and 4P. How do we define this
> without getting into several
>
> different types of PSE and PD, with all the complexity of
> hooking them together? I think the
>
> answer is simple: We define a 2P system, and then say that
> it's legal to have one or two of these
>
> on the same cable. So, two independent 2P systems make a 4P
> system. I think this is really the
>
> only way to go, but of course I'm open to other ideas.
>
>
>
> wael
> *From:* owner-stds-802-3-poep@xxxxxxxx on behalf of Yair Darshan
> *Sent:* Fri 6/3/2005 1:22 AM
> *To:* STDS-802-3-POEP@xxxxxxxxxxxxxxxxx
> *Subject:* Re: [8023-POEP] 2P v 4P and safety
>
> Hi Steve and all,
>
>
>
> I absolutely agree with you that the complexity difference between
> 2P and 4P are narrow due too the other problems that we have with
> 2P higher current.
>
>
>
> It is worth to check the advantages of detecting broken wires
> compared to other alternatives in order to decide what alternative
> is good enough and keep
>
> reasonable cost and reliability for the most popular installations.
>
>
>
> Yair
>
>
>
> * From: * owner-stds-802-3-poep@xxxxxxxx
> [mailto:owner-stds-802-3-poep@xxxxxxxx] *On Behalf Of *Steve Robbins
> *Sent:* Thursday, June 02, 2005 9:23 PM
> *To:* STDS-802-3-POEP@xxxxxxxxxxxxxxxxx
> *Subject:* Re: [8023-POEP] 2P v 4P and safety
>
>
>
>
>
> I totally agree that we need help from the cable group to
> determine more realistic current limits.
>
>
>
> When I started this email chain, I was looking for some way to
> decide between 2P and 4P on the basis of complexity and cost.
> Everyone knows that 4P is more expensive, because it requires more
> circuitry to switch and current-limit the extra pairs. But I
> realized that a 2P system might need extra circuitry too, if it
> was pushing the current closer to safe limits. (That's how the
> safety issue got included.) So the complexity/cost gap between 2P
> and 4P is narrowed by this argument. That's the main point I was
> trying to make.
>
>
>
> But now I see all this discussion has led to another big
> point: Worst-case analysis is greatly simplified if we have
> circuitry that detects broken wires, because then we only have
> to consider one case, where all wires are carrying nearly equal
> currents. Without detection, broken wires just makes a mess of
> any worst-case analysis. Basically, where do you stop? How many
> broken wires are reasonable? Unfortunately, I don't think "one"
> is a reasonable answer because if a cable suffers damage
> sufficient to cut one wire, there could very well be two cut.
> There appears to be a large number of possibilities and I don't
> see any good way to trim it down to a manageable few, unless we
> rule them all out by making the silicon a little smarter.
>
>
>
> -----Original Message-----
> *From:* owner-stds-802-3-poep@xxxxxxxx
> [mailto:owner-stds-802-3-poep@xxxxxxxx]*On Behalf Of *Chad
> Jones (cmjones)
> *Sent:* Wednesday, June 01, 2005 8:16 PM
> *To:* STDS-802-3-POEP@xxxxxxxxxxxxxxxxx
> *Subject:* Re: [8023-POEP] 2P v 4P and safety
>
> Time to throw into this conversation:
>
>
>
> What I read in this email chain is that we need to limit the
> current on the PAIR to the limit of one conductor, presently
> quoted as 420mA. I'm going to guess that only incrementally
> raising the current from 350mA to 420mA and raising the min
> voltage to something like 51V is going to cause a decent
> number of present participants to drop out. This is only
> 21.4W at the PSE for two pair (and 42.8W for 4P). It will
> certainly cool my interest in the project. I will also
> reiterate stated goal of pushing the physical limits of the
> cable.
>
>
>
> We definitely need help from the cable group with this
> project. One of the questions I have is how they derived the
> 420mA number. Does it change if we only use 2P in a 4P
> cable? Also, does it change if only 1 wire in a pair is
> carrying current (like when one conductor breaks)? How much
> margin is built in (in other words do we have use their number
> as the max or do we also add margin on top of it)?
>
>
>
> Last comment, I would not be opposed to mandating current
> balance monitoring within a pair and between the pairs for PDs
> that need the high power modes that could cause this 'safety
> hazard'.
>
>
>
> - Chad Jones
>
>
>
> * From: * owner-stds-802-3-poep@xxxxxxxx
> [mailto:owner-stds-802-3-poep@xxxxxxxx] *On Behalf Of *Yair
> Darshan
> *Sent:* Wednesday, June 01, 2005 4:04 AM
> *To:* STDS-802-3-POEP@xxxxxxxxxxxxxxxxx
> *Subject:* Re: [8023-POEP] 2P v 4P and safety
>
>
>
> Hi Steve and all,
>
>
>
> I basically agree with you regarding the situation you have
> described.
>
> (One wire broken==>Current is doubled ==> temperature rise
> increased ==> cable and/or connector parameters may be
> impaired==>
>
> we need to sense this situation and protect the infrastructure).
>
>
>
> Actually we are protecting the infrastructure in the
> IEEE802.3af by limiting the current, time and duty cycle.
> However, we are not sensing a broken wire in order to protect
> from doubling the current from the simple reason that it is
> not needed IF the time duration peak current and duty cycle of
> the event kept low enough so increase in temperature rise will
> not happen. The numbers of the IEEE802.3af were chosen in
> order to ensure this behavior (there are additional reasons
> for the numbers that we have decided in the IEEE802.3af and
> the end result was integration of all argument to get reliable
> operation while keeping the cost of silicon down)
>
>
>
> I guess that we should aim to the same concept in our new
> project i.e. set the numbers that in case of failure keep the
> infrastructure parameters in their operation range without
> additional sensing and features in order to keep the system
> simple and low cost.
>
> Probably we will need to revisit the protection mechanism by
> optionally adding a sense function that will alert of broken
> wire due higher current in 2P concept compared to the
> IEEE802.3; however we need test results in order to decide the
> way to go and some of us working on getting lab data.
>
>
>
>
>
> Other comments:
>
>
>
> 1. PD has to have IEEE defined components (PHY or others)
> otherwise it is not a compliant PD. PD is a powered DTE. DTE
> is a device defined by IEEE802.3. Hence PD must meet
> IEEE802.3. We don't care about other implementations that are
> using PoE technology and are not IEEE802.3 compatible.
>
>
>
> 2. You are right that a link is not required at the first
> place; actually according to the spec PSE operations are
> independent of data link status (33.2.4) however the spec
> allows wide flexibility for specify implementation specific
> errors under the variable "error_condition" which will results
> by power shut off.
>
> In other words if A PSE vendor/ System designer decides that
> the PSE sees something that is not normal (he decides what is
> normal or not; he is the vendor and he should know.) and it is
> not specifically covered by the spec, (there are many
> scenarios that meets that definition) it can be considered
> "error_condition" as specified by the state diagram.
>
>
>
> According to the above, if during operation system reports to
> PSE that "something bad has happen" (by getting information
> that data was good and now is dead and current is higher than
> normal for example) PSE is encouraged to do something about it.
>
>
>
> 3. I agree with the statement that true worst-case conditions
> in a 2P system must assume one wire is broken. In 4P, if we
> stay around the AF currents that we don't have to take in
> consideration that we have problem if one wire was broken due
> to the fact that this condition was tested and we don't have
> problems with it. If in 4P we will have higher current than
> 350mA than we have to test again.
>
>
>
> Yair
>
>
>
> * From: * owner-stds-802-3-poep@xxxxxxxx
> [mailto:owner-stds-802-3-poep@xxxxxxxx] *On Behalf Of *Steve
> Robbins
> *Sent:* Wednesday, June 01, 2005 4:12 AM
> *To:* STDS-802-3-POEP@xxxxxxxxxxxxxxxxx
> *Subject:* Re: [8023-POEP] 2P v 4P and safety
>
>
>
>
>
> Yair and all,
>
>
>
> Perhaps I could make this a bit clearer. Here is a
> hypothetical situation: Suppose we determine that CAT5 cable
> can carry 800mA per twisted-pair (400mA per conductor), and
> under worst-case conditions (max amb temp, big bundle, etc.)
> the wire temp stays 10C below the point where the cheapest
> insulation starts to soften. So we set I_CUT to 810mA, and it
> seems safe.
>
>
>
> Now suppose that we have actual worst-case conditions
> somewhere, but then one wire breaks. The whole 800mA is
> flowing through one wire, not two. The wire resistance is
> doubled, so power dissipation in the twisted-pair doubles, and
> the temp rise increases. (I'm not sure that the temp rise
> would double, since the power dissipation in the other pairs
> hasn't increased. But obviously the wire carrying 800mA will
> get hotter, probably significantly hotter.) If it goes up
> 11C, it exceeds the softening point.
>
>
>
> True, the Ethernet data would be messed up, but who cares? A
> PD doesn't even need to have a phy chip. Loss of link is not
> a reason for the PSE to shut off power, at least under
> 802.3af, because link is not required in the first place. In
> the above example the load current stays below I_CUT, so power
> stays on and the wire temp stays above the softening
> point until eventually something bad happens.
>
>
>
> The point is, true worst-case conditions in a 2P system must
> assume one wire is broken. In a 4P system, I think we'd have
> to assume one wire was broken in each of the pairs. (Don't
> assume broken wires are statistically independant events. If
> a wire breaks there is a good reason, and other adjacent wires
> could very well be broken too. Once, when I was at Boeing I
> had a whole bundle of wires come loose in my hand when I gave
> it a small tug. They all broke off right at the connector
> pins. Yet every individual connector contact in that bundle
> had survived a much stronger pull test right after
> crimping. A lengthy failure investigation determined someone
> had used an improperly calibrated crimp tool, which made the
> wire brittle and eventually led to fatigue. After two years
> in the field, some were actually broken but still just barely
> touching, which led to intermittent equipment failures.)
>
>
>
>
>
> Steve
>
> -----Original Message-----
> *From:* Yair Darshan [mailto:YairD@xxxxxxxxxxxxxx]
> *Sent:* Tuesday, May 31, 2005 1:03 AM
> *To:* Steve Robbins; STDS-802-3-POEP@xxxxxxxxxxxxxxxxx
> *Subject:* RE: [8023-POEP] 2P v 4P and safety
>
> Hi Steve and all,
>
>
>
> Please find attached below few comments on the thoughts
> presented below:
>
>
>
> 1. There are basically two roadblocks to higher power:
>
> a. The limited amount of current that can be carried
> on a single wire (not a pair of wires).
>
> We must assume that sometimes a wire will break,
> or a connector pin will go bad,
>
> so all the current will be on one wire instead of
> a pair. This fault condition must define
>
> the limit on how high I_CUT can go, for safety.
>
> b. The affects of heating and current imbalance on
> the magnetics.
>
> ----------------
>
>
>
> 1a. If a wire or a pin will break the data path will not
> work due to the fact that the magnetic will immediately
> saturate.(Lets assume that at the worst case all pairs are
> data pairs such in 1G)
>
> There fore assuming the above condition may not be
> considered as a worst case that we have to handle since it
> is a case that the system is not working and we don't care
> what happens as long as we are not causing damage.
>
>
>
> Regarding causing damage:
>
> If one conductor is break, we will have twice the current
> going through the other conductor and if we keep the
> TLIM/TOVLD and duty cycle than even for 2-4 times the
> current overload/Ilim numbers we probably will not have
> damage issue due to the averaging factor over time.
>
> Connector pin according to its spec need to meet 750mA at
> 60C. With cables it is a temperature rise issue which is
> not an issue for the overload/ILIM timings and duty cycle.
>
>
>
> Regarding safety issue: There is no safety issue here in
> terms of human safety. There is an issue of
> "Infrastructure safety or equipment safety" in terms of
> damage however as mentioned above, with TLIM/TOVLD and
> duty cycle limitation the equipment is kept un damaged and
> "safe".
>
>
>
> 1b. Yes, this is one of the major issues in 2P.
>
>
>
> --------------
>
>
>
> 2. To make sure we get it right in terms of safety, I
> propose:
>
> a. The objective regarding the goal of higher power
> should be amended to include
>
> a few words about safety.
>
> b. Let's look for a worst-case scenario to keep in
> mind while we work on the standard.
>
> The best one I've thought of so far is the
> airline industry. Assume PoE is used onboard
>
> a jet so that passengers can surf the net or
> watch DVDs in flight, without the batteries
>
> in their laptops dying. How hot do you think the
> FAA will allow the wires to get? How
>
> much current can a single conductor (in a
> bundle) carry before it reaches that temp?
>
>
>
> -------
>
>
>
> 2a - We don't have a mandate to define safety (human
> safety) in our standard. We need to address
>
> the reader to current safety standard.
>
> We should address issues that have impact on
> keeping the equipment and infrastructure undamaged.
>
> 2b - I agree that we should look for the worst case
> HOWEVER the worst case scenarios should be real cases and
> not corner or theoretical cases in order not to increase
> the cost of solution or eliminating some implementation or
> support some applications.
>
> Regarding the example presented: I guess that it is not
> the worst case since in aircraft designs there are a lot
> of margin in the design (Aircraft...) so in my opinion it
> is not representing the worst case.
>
> --------
>
>
>
> 3. I thought about Clays presentation a lot over the
> weekend. It seems to me that reaching
>
> higher power (>30W) will require two things:
>
> a. Current imbalance sensing to detect broken wires,
> so we can increase I_CUT beyond
>
> what a single conductor can handle, without
> compromising safety.
>
> b. Active current balancing to keep the magnetics
> happy. Although this won't solve the
>
> heating problem.
>
>
>
> ------
>
>
>
> 3a. The intent is not clear. Do you mean current balance
> in 4P? If this is the case than Yes, we need to address
> this issue and your suggestions sounds reasonable.
>
> 3b- Active and passive current balancing to the magnetic
> was discussed in the AF and we didn't want to force this
> requirements in the standard due to the following reasons:
>
> - The vendor should keep his system working
> under his equipment spec. How he does it is implementation
> specific.
>
> - Forcing a requirement for active/passive
> current balancing will increase the port cost while in
> most cases it is not required (above some cabling length
> there is no issue)
>
> Actually we address this issue in the informative part of
> the standard and let the designer to decide how to handle
> the worst case situation.
>
>
>
> In our new project doing active current balancing will
> dramatically increase the port cost so the the 2P concept
> will cost much higher that the 4P... In addition as you
> have mentioned, the heat issue still exist.
>
>
>
> Bottom line: In my opinion: Current balancing is not cost
> effective
>
> --------
>
>
>
> 5. Compare the complexity of a 2P and a 4P system that
> carry equal power. The 2P system
>
> would need about twice the current per conductor as
> the 4P system. So, while the current on
>
> 4P system might just be low enough to avoid the need
> for ACB/CIS, the 2P system would almost
>
> certainly need these features. So the complexity
> difference between 2P and 4P may not be as
>
> big as previously thought, at least in some medium
> power range. For much higher power, 4P
>
> would require ACB/CIS as well, but 2P becomes
> infeasible because of safety.
>
>
>
> ---
>
>
>
> 5. Agree. From complexity wise, the is no difference
> between 2P and 4P high power and in my opinion 4P from
> system point of view will cost less than 2P and will be
> more reliable by far.
>
> (Regarding safety: again there is no safety issue here
> (the system may not work... but no safety issue))
>
>
>
> ----
>
>
>
> 6. Suppose PoE Plus allows both 2P and 4P. How do we
> define this without getting into several
>
> different types of PSE and PD, with all the
> complexity of hooking them together? I think the
>
> answer is simple: We define a 2P system, and then
> say that it's legal to have one or two of these
>
> on the same cable. So, two independent 2P systems
> make a 4P system. I think this is really the
>
> only way to go, but of course I'm open to other ideas.
>
> -----
>
>
>
> 6. Basically I agree with this concept which is what the
> 4P concept (part of it) represents: Let's take two
> improved AF power channels, combined them together to have
> twice or more the power by having the 4P..
>
> The problem that I have with part of your suggestion is
> that it will allow at the PSE side, PSE that supports only
> 2P high power and PSE's that support only 4P which leads
> to the undesired environments were we have many different
> PSE's and PD's that will cause us complex system.
>
>
>
> I suggest looking at the 4P as the only HP concept that
> will address all PD types.
>
> By using this approach, the 4P design will be efficient,
> and simple.
>
> And when it will be integrated, the difference in chip
> level between 4P and 2P will be small enough to stay cost
> effective.
>
> From system point of view (cost, reliability, flexibility
> etc) 4P will represents better cost numbers than 2P.
>
>
>
>
>
> Yair
>
> * From: * owner-stds-802-3-poep@xxxxxxxx
> [mailto:owner-stds-802-3-poep@xxxxxxxx] *On Behalf Of
> *Steve Robbins
> *Sent:* Monday, May 23, 2005 7:50 PM
> *To:* STDS-802-3-POEP@xxxxxxxxxxxxxxxxx
> *Subject:* [8023-POEP] 2P v 4P and safety
>
>
>
>
>
> PoE Plus Team,
>
>
>
> Here are some thought I had over the weekend regarding 2P
> vs. 4P, and safety. I checked with Mike, and he wrote
> that it's okay to post this on the reflector and get a
> converstation going.
>
>
>
> 1. There are basically two roadblocks to higher power:
>
> a. The limited amount of current that can be carried
> on a single wire (not a pair of wires).
>
> We must assume that sometimes a wire will break,
> or a connector pin will go bad,
>
> so all the current will be on one wire instead of
> a pair. This fault condition must define
>
> the limit on how high I_CUT can go, for safety.
>
> b. The affects of heating and current imbalance on
> the magnetics.
>
>
>
> 2. To make sure we get it right in terms of safety, I
> propose:
>
> a. The objective regarding the goal of higher power
> should be amended to include
>
> a few words about safety.
>
> b. Let's look for a worst-case scenario to keep in
> mind while we work on the standard.
>
> The best one I've thought of so far is the
> airline industry. Assume PoE is used onboard
>
> a jet so that passengers can surf the net or
> watch DVDs in flight, without the batteries
>
> in their laptops dying. How hot do you think the
> FAA will allow the wires to get? How
>
> much current can a single conductor (in a
> bundle) carry before it reaches that temp?
>
>
>
> 3. I thought about Clays presentation a lot over the
> weekend. It seems to me that reaching
>
> higher power (>30W) will require two things:
>
> a. Current imbalance sensing to detect broken wires,
> so we can increase I_CUT beyond
>
> what a single conductor can handle, without
> compromising safety.
>
> b. Active current balancing to keep the magnetics
> happy. Although this won't solve the
>
> heating problem.
>
>
>
> 4. Maybe it's time for an acronym here. How about Active
> Current Balance (ACB) and Current Imbalance
>
> Sensing (CIS). They seem to go together, so "ACB/CIS"?
>
>
>
> 5. Compare the complexity of a 2P and a 4P system that
> carry equal power. The 2P system
>
> would need about twice the current per conductor as
> the 4P system. So, while the current on
>
> 4P system might just be low enough to avoid the need
> for ACB/CIS, the 2P system would almost
>
> certainly need these features. So the complexity
> difference between 2P and 4P may not be as
>
> big as previously thought, at least in some medium
> power range. For much higher power, 4P
>
> would require ACB/CIS as well, but 2P becomes
> infeasable because of safety.
>
>
>
> 6. Suppose PoE Plus allows both 2P and 4P. How do we
> define this without getting into several
>
> different types of PSE and PD, with all the
> complexity of hooking them together? I think the
>
> answer is simple: We define a 2P system, and then
> say that it's legal to have one or two of these
>
> on the same cable. So, two independant 2P systems
> make a 4P system. I think this is really the
>
> only way to go, but of course I'm open to other ideas.
>
>
>
>
>
> Steve Robbins
>