I'm a software engineer, so please excuse my gross ignorance of this topic. We are trying to figure out why our DC-DC converters sometimes fail and stop outputting power after a month or so of runtime.

Please refer to the drawing. Below is also a quick textual description of the drawing.

power component layout

We use a 48v/1A passive PoE power injector to introduce 48v into the system. As far as we can tell (measured by multimeter), it seems to provide stable output at different loads. We then split that power two-ways: one to a 5v/3A step-down and another to a 12v/2A step-down. The 5v converter powers a tablet PC (approx. 10W-nominal / 15W-max load) and the 12v converter powers a battery (approx. 15W-nominal / 22W max peak). Note, the battery has a custom designed charging controller to self-limit its power draw to < 2A. Also, I believe the converters are 95%+ efficient (if the supplier is telling the truth).

We've used DC-DC converters from different suppliers, some with even higher output load capacities, and we still experience intermittent failures. We seem to have about 1 failure out of 10 per month.

We've also tried using a diode between the battery and the 12v converter (to ensure voltage is not fed back into the 12v converter's input side from the battery). This did not seem to make any difference. I've also verified that the tablet PC is NOT feeding back-voltage to its converter.

So, my starting question: Is there anything obviously/inherently wrong with the way we're doing this, or the components we're using? All I really know is the power-budget and that we need to stay well under it (like 80% or less of max).

If any of this is unclear, is there any other information/testing I can provide?

UPDATE #1: Power cable length can be up to 100m, but we typically use less than 5m (copper Cat-5e or better).

UPDATE #2: Specifications are hard to get about the actual components in datasheet format, but below's images of specifications is the best I can do for now.

UPDATE #3: I've included photos of oscilloscope measurements on the 48v side, further below. I really have no clue how to properly use a this device, so I just connected the ground to the PoE/48v negative and the probe to the PoE/48v positive. I then just pushed the "auto" button on the scope.

First image is for the 12v converter... 12v specs

The next images are for the 5v converter... 5v specs 1 of 3 5v specs 2 of 3 5v specs 3 of 3

This is the oscilloscope reading on the 48v side, with no DC converters connected (no loads whatsoever), as a baseline: 48v bus reading with nothing connected

This is the oscilloscope reading on the 48v side, with only the 5v DC converter connected (and its load, the tablet PC, connected): 48v bus reading with only loaded 5v converter connected

And just for fun, this is the oscilloscope reading on the 48v side, with ANOTHER brand of 5v DC converter connected (with tablet load connected): 48v bus reading with another loaded brand connected

  • 3
    \$\begingroup\$ You should include links to the (datasheets of) DCDC converters so that we can properly understand what they are. Also the photo shows a short ethernet cable. In your real setup I expect that cable to be much longer. Is that so? How long is that cable? \$\endgroup\$ Commented Feb 21, 2019 at 15:17
  • \$\begingroup\$ Is that a class D amplifier? How much power does that use? \$\endgroup\$ Commented Feb 21, 2019 at 15:20
  • 2
    \$\begingroup\$ Which DC-DC is failing? Have you opened any of the DC-DC to see what went wrong? \$\endgroup\$
    – smajli
    Commented Feb 21, 2019 at 15:27
  • \$\begingroup\$ @smajli OP states: I'm a software engineer, so please excuse my gross ignorance of this topic. so I can fully understand why that investigation was not done and the DCDC converters were simply replaced. And personally I respect that even more than trying to find the cause without having the faintest clue how a DCDC converter works. \$\endgroup\$ Commented Feb 21, 2019 at 15:30
  • 1
    \$\begingroup\$ You say the battery "has a custom designed charging controller to self-limit its power draw". I'm starting to suspect that this controller doesn't work as well as you think it does. Show us the schematics. \$\endgroup\$ Commented Feb 21, 2019 at 15:36

3 Answers 3


One possibility is that your DC/DC converters may be overheating or that their lifetime is shortened by operating at a high temperature. I've seen capacitors rated for 3,000 hours at 80C so it's not crazy to think that this might be an expected failure mode for an input or output capacitor. Can you check the worst case operating temperature? If it's close to 80C you might want to cool it better, even a few degrees can have a big impact on component life.

A second possibility I see looking at your scope plot is that the supply line's inductance is causing weird transients on the inputs. You might want to try introducing the largest decoupling capacitor you can find (rated for 100V) to your system at the DC-DC converter inputs so that it can supply power as the converters switch open and shut (before current is able to start flowing on the long wire). This decoupling capacitor will also store energy as current continues to flow after it's no longer needed.

  • \$\begingroup\$ When adding capacitors to lines with impulses/spikes, go for caps marked '105 deg C' \$\endgroup\$
    – smajli
    Commented Feb 21, 2019 at 20:04

Operating on the apparent notion that the 5v converter was introducing some noise/spikes/transients/etc. back onto the 48v side of things... and that was contributing to failure of other components on that bus (like the 12v converter).....

I bridged a .1uf capacitor ("104M" marking) across the 48v input on the 5v converter. Upon doing so, the scope plot almost completely cleared up. I found similar behavior at the fault of some 12v converters, as well.

I'll be informing the supplier of their shoddy quality, and maybe commissioning a quantity of "snubbers" for current inventory.

I thank everyone for their help. This StackExchange site is much more friendly than StackOverflow (where I frequent software related discussions), so thanks for everyone's participation. I'm not quite sure who to singly give credit to!


Yup, it's the components

You didn't link the exact data sheets, but others did, and what they pointed to was cheap Cheese junk. https://yucoo.en.alibaba.com/productgrouplist-807018122/48v_to_12v.html?spm=a2700.icbuShop.98.17.122636feingm2j. You really don't even need to click, just spot the "alibaba" in the URL.

This is $5 for a DC/DC that should cost $40.

They tend to be optimistically specced, so loading it at near its max (80%) is just asking for trouble. I would use 80% on a General Electric unit. For the rubbish from eBay/Ali/Amazon Marketplace, I figure more like 33%.

Part of the problem is these products have no accountability as to quality. The "RoHS" and "CE" marks are universally forged by non-European makers, because it is a voluntary "honor system" mark; Enforcement is by catching a manufacturer making substandard products and essentially suing them. But this is toothless outside the EU. In that case importers are responsible, and in direct mail, the importer is you!

enter image description here

Now contrast with this product here, containing this mark. As discussed, RoHS and CE are rubbish. But look at the cЯUus" mark. That is these guys, which are the premier US testing lab. What that mark means is if you go to get the whole machine UL listed, those components are pre-approved, and UL won't bother torture testing the components.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.