We had 3 of the original PS blown in just 2 weeks, all with a bulge appearing at the top of C10 like this:
IMG_8682.jpg
Therefore, I had to invest some time into this. The results are as follows.
The first decision was easy. Since SMPS of the required 69W (according to the specs posted at http://www.cloudcorerouter.com/) are not very expensive, we decided to replace them with a better model. The choice was the proven Traco Power TOP 100, although I’ll probably check out the newer and more compact TPI 100-124A-J later on. They both provide 100W.
One of the most sorely missed features in the CCR 1036 is dual power supplies. The problem is that space is tight inside the box, so a tiny PS was required. I settled for the Mean Well EPS-65S. While a little bit underpowered, it will allow for a peak power of 71.6W for 10 seconds which should be enough - none of the CCR 1036 in production take more than 55W continuously. Importantly, it has a low profile with a height of only 24mm PCB base to top.
To install this PS, I designed and 3D printed a base that attaches to the left of the mainboard using the original screws:
second ps support.png
I have attached the source for this base in OpenSCAD.
The PS is attached to the base with M3 screws, nuts and bolts.
Connectors for the PS are specified in their respective data sheets. Unfortunately, one uses Molex and the other JST; both are cheap and easy to work with.
Once I started testing, I found that when the Mean Well got their power input removed, it would for some reason cause the router to reset. I think it draws a peak of power and stalls the Traco Power, since that PS was not designed for redundancy. This was solved by inserting a 1N4004 diode between the +V pin of the Mean Well and the mainboard. In the next conversion, I’ll add the same diode for the other PS just in case.
With that out of the way, testing was smooth. Either PS can boot the router and keep it working for as long as I have tested so far. Any single removal of power goes unnoticed by the router.
The Traco Power get a bit warm, but not a lot. The Mean Well, probably because its inline diode keeps its voltage slightly under the other and therefor gets no load, keeps cool.
A bit more testing in the bench, then some testing in the datacenter, and if nothing pops up they will be signed off for production work.