Cdycam 3pcs/1bag PoE Splitter Power Over Ethernet 48V to 12V IEEE802.3af Standard Type 15W Power Supply Module For NonPoe Camera (WHL #119)
Here’s something that lasted just shy of five years – a PoE to 12V converter.
I just checked, I never wrote about these things…so here we go. Last week, the ESP32 in my living room stopped working. It does serve as a sensor station (temperature, humidity) for the DECT heater valve in the living room, and it also runs some LED stripes around the house, some of which are automated via iobroker, and some are activated by infrared sensors. They all can be changed over a simple web interface as well. Turns out: The ESP is fine, but the power supply, a PoE to 12V converter, is dead. And now that I think about it: That was no sudden event, it had it coming for quite some time.
First things first: I bought this piece of equipment back in July 2020 on Aliexpress, for a whopping 8,56€ including shipping…for a pack of three. I didn’t need three at the time, but a pack of two would have probably cost me about the same. I also bought a gigabit unit at the same time, which cost a flat 10,00€ and is therefore triple, almost quadruple the cost – but for these, I didn’t need the gigabit passthrough, so that’s fine. My PoE switch generally has enough ports to supply everything directly.
The unit itself looks like this, aside from a bit of marking there really isn’t any “front side” or “back side”, it’s a small, black case, put together with some glue on undersized clips, and that’s about it:
It does have a standard 5,5/2,1mm barrel jack and a 4P8C 100Mbit RJ45 connector cable sticking out on one side, and the other side has a full 8P8C RJ45 jack. It’s probably more expensive to source a 8P4C that would be sufficient for the job, so they just use the standard full pin part instead.
Once cracked open (very crunchy – the replacement unit did not make these sounds!), this is the top with nice alignment pins molded into the case:
And this is the bottom of the PCB once taken out:
Not a large component count at all, is it..
(Solder joints on the cables touched up with some leaded solder, as they broke off when I removed the case. Not great soldering from the beginning, plus tiny wiring diameter, plus all stuffed into the small case)
Failure mode of my unit was obviously the capacitors that are already missing on those two photos. Here are they, desoldered for your enjoyment:
EC1 is the 47µF 63V cap (6.3mm diameter x 13mm height), and EC2 is the 100µF 16V cap (5mm diameter x 11.5mm height). Both 105°C (not sure about that…) electrolytics, both from a “CIA” (?!) brand (yeah!), both “low ESR” types. EC2 was folded over the wiring, it’s pretty crammed at the end.
PoE power is strictly taken from the PCB connector, it doesn’t work the other way round. For my Christmas tree, I have even removed the wired RJ45 plug because connecting the wrong end was so annoying.
So after desoldering the capacitors (and looking at some traces of heat on the top half of the case), I put them to the test.
The 47µF only buffers the PoE DC voltage, so this shouldn’t be critical. The cheapest cap in that spec and dimension with explicit ESR figures in the datasheet would be the Rubycon 63YXJ47M6.3X11 at 27 cents a piece (MOQ1), which has 350mR at 20°C and 100kHz. That’s still a general purpose capacitor, not a low ESR one.
Our specimen has… 11 ohms. Well, that’s clearly not low ESR. Not anymore, at least, maybe it once was. But the 46µF of remaining capacity sound alright.
But that wasn’t the culprit. That would be the 100µF cap after the switching regulator, the one that needs to smooth the converted voltage. Low ESR would be great here since this cap will deal with some ripple. Same search on Mouser, this time I got a Würth Elektronik WCAP-ATLL 860160372011 at 13 cents per unit. Again, not a low ESR type, but GP. This one has 1.1R at 20°C/100kHz, so absolutely in GP territory.
Aaand our trusty CIA branded one?
Yeah. 0.15k or 150 Ohms ESR. At 15µF or 15% remaining capacity. Whoops 
It also has a Vloss of 32%, which, roughly speaking, indicates that one third of the stored energy evaporates into heat. Which leads us back to my initial observation that this was no sudden death of a device, this was a slow yet steadily accelerating process of drying out this electrolytic cap (more heat -> faster drying -> worse ESR -> more heat). It’s a classic Rumsfeld “unknown known” – for quite a while I noticed that with the LED strip on, the temperature reading on this ESP goes up by quite a bit. That’s no issue for the dew point calculation (as humidity falls accordingly), but it messes with the room temperature reading, which in turn does weird things with the central heating commands.
Here’s an example from February: I guess the patio door was briefly open at like 10:40 PM, and starting at exactly midnight, iobroker sets the LED brightness to full and then slowly dims to 0% at 1 AM (“time to go to bed!”). All other rooms are unaffected, but the living room temperature shown in red rises by almost one degree Celsius. That ESP is pretty close to the PoE adapter – of course the MOSFETs that run the show are even closer to the temperature sensor, so this was my original explanation.
Now I don’t have a proper comparison dataset because with the 20°C+ degree weather outside the same door is typically open until late at night, so this might be the best I can show you:
Dataset from yesterday – door closed at 23:41-ish, and all rooms with open doors (affected by the nice breeze) show a sharp rise in temperature. I don’t see an additional hard bump in temperature when the lights turned on at midnight, and I also don’t see any temperature offset of the living room compared to the others that slowly dies down at/after 1 AM.
…and this was taken with a new PoE adapter, because as I told earlier, this was a pack of three, and I didn’t use all of them right away. It was stored away and probably still has fresh capacitors that don’t burn through another Watt or two during that heavy load period. It was the fucking dried up capacitor that messed with my ESP temperature monitoring!
Safe to say that after one failed attempt to replace both caps, I will not spend any more time and money to fix a 3€ converter, and use the surplus unit instead. And if in, let’s say, late 2030 this PoE adapter dies as well from the same causes – I will know what to check, and I will have a replacement unit ready in case it’s also unfixable with reasonable efforts.
So there’s that, bloody capacitors strike again 
