EEVblog #1122 - Raspberry Pi 3 PoE Hat FAIL Investigation

[EEVblog]

The new official Power Over Ethernet (PoE) hat for the Raspberry Pi 3 B+ has been reported to fail when powering USB devices.
Dave investigates, until, well, something goes horribly wrong...

[Zbig]

The Register has interviewed the Raspberry Pi guy who has elaborated somewhat on the cause of the issue: https://www.theregister.co.uk/2018/09/11/raspberry_pi_poe_hat_issue/

[Berni]

This must have been designed by a intern or something because pretty much everythyng is wrong with this design!


So the reason that i think this is causing a problem is the sharp rise time in the ripple on the 5V output. That fast risetime will try to quickly charge any capacitance on the other side of the USB hub protection circuitry, once the load is big enough the spikes in current will exceed the trip point and make it go into overcurrent mode. The reason it happens on all ports is that the Raspberry Pi 3 uses a single 5V high side switch with built in protection to switch all 4 ports (Cost saving most likley). The pins on the USB hub are wired together so that a overcurrent signal is sent for all 4 ports, that's why it will report an error for all ports. I know about this because i had to reverse engineer that part of the circuit some time ago.

Lets start listing whats wrong with it:
- The HV clearance is very important for PoE compliance as you can get massive spikes on long ethernet cables. That slot is doing absolutely nothing due to the copper fill mistake and im sure the clearances are not to spec.
- The fact that plugging it in gets in the way of the GPIO header is a major fail. You could have used one of those long pinned troughhole headders that arudinos and so many other Pi Hats use.
- That tiny 4 pin connector should have at least had some large copper pads with vias to hold it down (Or even better just use troughhole)
- 250mV of ripple is pretty bad, but its made even worse by that sharp rising edge in it
- This can NOT run reliably at 130°C on major semiconductor components.
- They did not use the opportunity of a copper fill around the diode to cool it (Not even giving it vias to help drain the heat away)
- The layout is a bit meh around the secondary side as it has a fair bit of current loop by the time it gets rectified to DC (These are very fast swings in this topology)
- The common mode cap over the isolation barier has a really long way around. This will likely make the board radiate radio interference like crazy
- That common mode current also has two paths, each one around the fan, not a good design with fast transient currents
- Output voltage goes out of control with no load calls for at least some sort of zenner diode to keep it within sane limits (This topology can really run away badly with no load)
- It should survive such an output short circuit just fine as im sure a lot of people will eventually short out the 5V rail trough some stupid mistake (Most regulator designs are short circuit tolerant these days)

[Monkeh]

The Register has interviewed the Raspberry Pi guy who has elaborated somewhat on the cause of the issue: https://www.theregister.co.uk/2018/09/11/raspberry_pi_poe_hat_issue/

It was, Upton observed, "dumb luck" that heavy load testing was done with one brand of switch while lighter testing occurred with the other. Thus the PoE HAT passed product validation.

I'm not quite sure 'dumb luck' is the term for not performing full tests against all variations of your product.

[Habropoda]

The slot is for the camera cable.  It has nothing to do with clearance.

[free_electron]

without even looking at anything else : who the hell suspends a transformer in a hole like that ?

[JustMeHere]

without even looking at anything else : who the hell suspends a transformer in a hole like that ?

Watch the video.

[thm_w]

I don't understand why there was so much hype for POE hat.

You can spend less money, and get a properly enclosed product, that can be reused for another project down the road if needed:
https://www.aliexpress.com/store/product/Cdycam-IEEE-802-3af-Micro-USB-Active-PoE-Splitter-Power-Over-Ethernet-48V-to-5V-2/1920360_32841790680.html
https://www.aliexpress.com/store/product/Xinfi-DC-12V-POE-Adapter-Injector-Splitter-Connector-IEEE802-3af-Active-10-100Mbps-For-IP-Cameras/1543317_32620368747.html

The only advantage I see is its slightly more compact.

[dardosordi]

Hey Dave, check this one http://www.ivmech.com/magaza/en/development-modules-c-4/poe-hat-for-raspberry-pi-3-model-b-p-111

[JustMeHere]

I don't understand why there was so much hype for POE hat.

You can spend less money, and get a properly enclosed product, that can be reused for another project down the road if needed:
https://www.aliexpress.com/store/product/Cdycam-IEEE-802-3af-Micro-USB-Active-PoE-Splitter-Power-Over-Ethernet-48V-to-5V-2/1920360_32841790680.html
https://www.aliexpress.com/store/product/Xinfi-DC-12V-POE-Adapter-Injector-Splitter-Connector-IEEE802-3af-Active-10-100Mbps-For-IP-Cameras/1543317_32620368747.html

The only advantage I see is its slightly more compact.

POE makes it very easy to get power to a distant location.  POE cameras are a popular example.  You don't have to pay for an electrician.  That is big.  Cat5 is very cheap compared to power wire.

Also POE gives you the ability to remotely control the power.

[EEVblog]

I don't understand why there was so much hype for POE hat.

You can spend less money, and get a properly enclosed product, that can be reused for another project down the road if needed:
https://www.aliexpress.com/store/product/Cdycam-IEEE-802-3af-Micro-USB-Active-PoE-Splitter-Power-Over-Ethernet-48V-to-5V-2/1920360_32841790680.html
https://www.aliexpress.com/store/product/Xinfi-DC-12V-POE-Adapter-Injector-Splitter-Connector-IEEE802-3af-Active-10-100Mbps-For-IP-Cameras/1543317_32620368747.html

The only advantage I see is its slightly more compact.

Yeah, plenty of existing robust solutions exist to give you 5V @ 2.5A, but just that it's a separate box.

[EEVblog]

without even looking at anything else : who the hell suspends a transformer in a hole like that ?

Very comment technique to get the height profile down. You see it on countless DC-DC converter designs.
Bonus is it's also a lower creepage path

[ogden]

- The HV clearance is very important for PoE compliance as you can get massive spikes on long ethernet cables. That slot is doing absolutely nothing due to the copper fill mistake and im sure the clearances are not to spec.

As we all saw in the video, Raspberry Pi 3 board fails miserably in this regard.

[TurboTom]

Shame the POE board got damaged. It shouldn't, just from a short at its output rail. What I would have liked to see is an attempt to operate the RPI with the POE unit located separately. The "antenna" effect (induced voltage in the probe ground lead, especially by that magnitude) in the vicinity of the transformer should have made you suspicious. The RPI board is just millimeters away when the POE board is installed and the stray field of the transformer will induce the same kind of voltages there as well.  I can very well imagine that this is actually causing all the trouble observed. It might be worth trying to put a short winding (copper foil) externally around the transformer as to be found frequently on bigger switchmode transformers. Anyway, since it's broken now, testing is over...

Thanks anyway for demonstrating the problem.

Cheers,
Thomas

[Peabody]

So if instead of using the POE hat you provided a nice clean 5V supply into that four-pin connector on the PI that the hat normally uses, there would be no problems with USB?  Just a test to eliminate the possiblility that getting 5V from that connector doesn't contribute to the problem in some way.

[free_electron]

without even looking at anything else : who the hell suspends a transformer in a hole like that ?

Very comment technique to get the height profile down. You see it on countless DC-DC converter designs.
Bonus is it's also a lower creepage path

Provided the transformer is mechanically supported .... This one is wafting in the wind. It is pretty big and will cause plenty of stress on the solderjoints and the copper adhesion. i hope they have shot some via's in the pads...

[free_electron]

a couple of design remarks
- why does that output diode run so hot ? That is not normal.
- Are we sure that transformer is properly made for that switching frequency ? It wouldn't be the first time they are running at the wrong frequency. some of these chips come in different frequency settings and the transformer needs to be adapted.
- primary side regulation is crap. that only works if you use a post-regulator.

[ogden]

Light (1W) load frequency of that converter was audible frequency, 9.47 KHz. Audio jack output could become unusable when powered from PoE hat. In next video (if any) would be nice to check that as well.

[Fungus]

That thing needs diagnosing/fixing. It might just be a diode.

[ricktendo]

I don't understand why there was so much hype for POE hat.

You can spend less money, and get a properly enclosed product, that can be reused for another project down the road if needed:
https://www.aliexpress.com/store/product/Cdycam-IEEE-802-3af-Micro-USB-Active-PoE-Splitter-Power-Over-Ethernet-48V-to-5V-2/1920360_32841790680.html
https://www.aliexpress.com/store/product/Xinfi-DC-12V-POE-Adapter-Injector-Splitter-Connector-IEEE802-3af-Active-10-100Mbps-For-IP-Cameras/1543317_32620368747.html

The only advantage I see is its slightly more compact.

I have one of these and it works fine with my Pi 2B+, but I can report the PoE part works but the LAN cable does not work with the Pi 3B+.

Note: the LAN cable on the PoE injector is only 10/100 (Fast Ethernet,) if you can find a 10/100/1000 (Gigabit) version of the same injector it may work.

P.S. I have this 3rd party PoE adter from Navo Labs, I will test it see if this too has the same issue

https://navolabs.com/product/raspberry-pi-3-micro-poe-hat/

Edit: I do not have a load tester so I do not know for sure, but as a test I plugged in an external HDD and my Pi did not power down like in Dave's video

[ogden]

I think there's nothing wrong with diode from design point of view. Diodes usually have to dissipate some power in switching converters (which are not synchronous thus w/o diodes). Most likely they use diode used in MP8007 reference design, SBR8U60P5 or similar. Such diode at 2.5A forward current have ~0.3V drop meaning dissipated power is 0.75W which is more or less what we see in thermal imager. IMHO problem is not diode but specification promising PoE Class0 12.5W with quite lossy dc-dc converter design. They are just trying to jump over their head/hat. Users would be fine with PoE Class2 6.5W (5V @ 1.3A) resulting in much lower heat dissipation.

[thm_w]

POE makes it very easy to get power to a distant location.  POE cameras are a popular example.  You don't have to pay for an electrician.  That is big.  Cat5 is very cheap compared to power wire.

Also POE gives you the ability to remotely control the power.

I mean the fact that there was a "POE hat" added, was the hype. POE itself is great, didn't think about the remote power control, would be good to remotely reset the unit.

I have one of these and it works fine with my Pi 2B+, but I can report the PoE part works but the LAN cable does not work with the Pi 3B+.

Note: the LAN cable on the PoE injector is only 10/100 (Fast Ethernet,) if you can find a 10/100/1000 (Gigabit) version of the same injector it may work.

Odd that it wouldn't work on the 3 but does on 2, you'd think the pi would just auto-negotiate to 10 or 100mbps. Maybe forcing 100mbps somehow would help.
The second link lets you choose a gigabit option, presumably adds the missing pairs.

[ogden]

IMHO rPI is screaming for low cost non-isolated "passive" 12V PoE hat which comes with 12V wal-wart supply and "pigtail injector". Usual rPI buyers rarely have 802.3AF-compatible switches around, PoE brick supply Dave used in the video also is not that wildly popular or cheap compared to 12V 1A "plugs with tail".

[wraper]

without even looking at anything else : who the hell suspends a transformer in a hole like that ?

Most of the thin TV PSUs.

[ricktendo]

I have one of these and it works fine with my Pi 2B+, but I can report the PoE part works but the LAN cable does not work with the Pi 3B+.

Note: the LAN cable on the PoE injector is only 10/100 (Fast Ethernet,) if you can find a 10/100/1000 (Gigabit) version of the same injector it may work.

Odd that it wouldn't work on the 3 but does on 2, you'd think the pi would just auto-negotiate to 10 or 100mbps. Maybe forcing 100mbps somehow would help.
The second link lets you choose a gigabit option, presumably adds the missing pairs.

Here are detailed PCB pictures of a 12v model, maybe this will provide a clue

http://www.knutsel.org/2016/05/31/poe-pi-powering-a-raspberry-pi-with-power-over-ethernet/

BTW it does appear that they have an updated Gigabit model for the Pi 3B+

eBay auction: #123357699802