[Martin Rowan] was lucky enough to get his hands on the revised Power Over Ethernet (PoE) hat for the Raspberry Pi. Lucky for us, he wrote it up for our benefit, including inspection of the new hat, it’s circuit, and electrical testing to compare to the original hardware.
You may remember the original release of the PoE hat for the Raspberry Pi, as well as the subsequent recall due to over-current issues. In testing the revised board, [Martin] powered a test load off the USB ports, and pulled over an amp — The first iteration of the PoE hat would often trip the over-current protection at 300 milliamps.
This afternoon, the redesigned PoE board was officially released, and the post mortem of the problem documented in a blog post. It’s a lesson in the hidden complexity of hardware design, as well as a cautionary tale about the importance of thorough testing, even when the product is late and the pressure is on.
The PoE hat converts 48 volt power down to a 5 volt supply for the Pi using a flyback transformer. The problem was that this transformer setup doesn’t deliver clean steady 5 volt power, but instead provides power as a series of spikes. While these spikes were theoretically in spec for powering the Pi and usb devices, some Raspberry Pis were detecting those spikes as too much current pushed through the USB ports. The official solution essentially consists of better power filtering between the hat and the Pi, flattening that power draw.
We’re looking forward to getting our hands on this new and improved PoE Hat, and using it in many project to come.
It’s a rather expensive and painful mistake which could have been prevented by having a handful of people test the product in the real world. The article linked tells us they omitted this and paid the price: “Our field testing programme was abbreviated because the product was late”
Yeah, but at least they shipped on time!
Same mantra triple A videogame publishers have nowadays since many of their customers has unmetered broadband.
“ship it and patch it later” just doesn’t work for hardware, and also rarely goes well with Operating System updates (right Microsoft?)
But they were out of beta and released it on time. The science got done, and they released a great hat, for the boards that were still alive…
Incredibly useful PoO “hat”. Obfuscating access to GPIO pins while generously giving a slot to access the oh-so-common CSI connector is genius ! Having the flyback transformer hovering over the board is noisily clever too.
What? And why do you thing, that there are holes on PoE HAT above GPIO? You just need to to use longer pass-through headers, which are mentioned even in official blog post about release of PoE HAT. Also, they’re cost like pennies, so price definitely isn’t the issue.
“Because the fan is controlled over I2C, none of the GPIO are used, so you can stack a second HAT on top of the connector. To do this you will need to buy some longer pass-through headers that expose the pins on the other side of the PoE HAT. You will need one for the 40-way and one for the 4-way connector that has the PoE splitters on it.”
https://www.raspberrypi.org/blog/introducing-power-over-ethernet-poe-hat/
So, when can I have your schematics, gerbers and a prototype board ready for production testing?
Armchair engineering at its finest.
The Raspberry PI PoE problem exists for one simple, very basic reason: the Raspberry Pi group doesn’t know how to design power supplies…or re-design poorly-designed power supplies.
Read Martin Rowan’s write-up for schematics, the “fix”, and insight as to how Eben Upton typically handles any RPi design problem.
After doing a ton of work, detailed testing and documentation, and even sending his hardware–along with detailed instructions as to how to duplicate the problem–to the RPi people, their response?– “…Within a few days the Foundation got back in touch to report they couldn’t reproduce the issue…”, and implied that Rowan’s Cisco hardware was the culprit.
The only thing which forced the RPi Group into serious action was that this story was picked up, and became a major news story on The Register (https://www DOT theregister DOT co DOT uk), a very powerful, very influential tech magazine.
Martin Rowan–“…but keeping the community in the dark about what was happening was a poor way of handling it…”
A very good job by Martin Rowan, and Jonathan Bennett
That thing is just poorly designed DC-DC converter. Check out Dave’s EEVBlog 1122 where he does analysis of official PoE hat pointing at obvious design flaws. You’ll probably get better PoE supply for your RPi on the eBay at a fraction of a price of official PoE hat. I use such supply for my Pi on the roof and it works flawlessly, I paid 5 bucks for it on eBay. I really love RPi and I think guys behind it did excellent job designing it, but PoE hat simply does not have such level of quality.
It was not “designed”; it was lifted right off the MP8007 data sheet, with some minor modifications made. Any power supply designer worth his salt would never have chosen this chip as the basis of a good design. This is simply a poor design with an incapable chip.
And the Raspberry Pi Organization claims they “designed” this? Do they claim this to enhance their credibility?
Worse yet, their “re-design” is not worthy of consideration or discussion; any third-rate electronics designer will ‘fix’ a poorly-performing power supply by adding more capacitors and an inductor to ‘smooth out’ a poor output.
One extremely interesting part of Dave’s EEVBlog (1122) is when he turns his thermal camera on the device, and finds temps on the input and output diodes, and the controller chip, of 110, 130, and 120 degrees C, respectively,
The other very interesting part is the variable-frequency, variable-amplitude noise of up to 500 mV (that’s 1/2 VOLTS, folks) on what is supposed to be THE 5-volt supply.
Just do a search on “eevblog 1122” for a real eye-opener on all the problems with this design; thanks, “Night’.
Dave’s EEVBlog 1122 does not do the Raspberry Pi Organization’s design credibility any favors.
Yabut: Dave was running it at over 12 watts, about an order of magnitude over normal operating conditions. Too bad he didn’t measure the temperatures with the thermal camera in the 1-5 watt range.
“EEVblog #1122 – Raspberry Pi 3 PoE Hat FAIL Investigation”; YouTube; time: 19:05.
I didn’t remember Dave saying anything about running the device at 12 watts, and decided not to sit through this video a third time; I did verify that this is the video I originally watched…twice, since the amount of detail and hard work invested in making it is so fascinating.
However…the Raspberry Pi PoE HAT IS spec’d at being able to deliver five volts at two and one-half amps (5 v. @ 2.5 A), with an input of 37 v.-57 v. ( I just checked Raspberry Pi’s web page for confirmation of this spec).
According to my calculations, 5 vdc X 2.5 A = 12.5 WATTS.
If Dave indeed was performing his test with the PoE HAT delivering 12 watts, he was running his test well within Raspberry Pi’s specs.
“…over 12 watts, about an order of magnitude over normal operating conditions…”.
Am I missing the point you’re trying to make?
Or are you not aware of the specs of this device?
Please clarify…
So I can get an inline PoE supply for a little under $4 shipped.
Or I can get a PoE hat style for the RPi range, not just the v3, for a little under $7.
And they work well in a production environment.
And the nay sayers will say it’ll kill my cat but he’s fine.
About a year ago, I bought one of the unofficial PoE hats for the 3b+. Everything went great, until I plugged in an HDMI cable. The Pi powered off unexpectedly. After a few minutes of troubleshooting, I figured out that the PoE port on the switch had died. Tried again, same result, second port dead.
Shouldn´t your switch have overcurrent / polarity protection ?
Have you tried to diagnose the issue ? Do you still have the link for the killer hat ?
It was this guy: https://www.amazon.com/NavoLabs-Raspberry-Pi-POE-Hat/dp/B07CLCFLH8/ref=sr_1_1?ie=UTF8&qid=1542136529&sr=8-1&keywords=pi+poe+hat
Ah I see it, indeed not isolated. Connecting the HDMI violated the PoE spec and is not intended to work (although the switch is also not allowed to break!)
Did you use an isolated PoE supply? If your product has no external IO (example: ip camera) you can use a cheaper unisolated converter, but then if you connect it to something external you can break things…