And before you know it you have a pretty silly brute force unregulated supply that could deliver something like 15A. Yes, ideally you wouldn't have this iron in such close proximity, but this is all I could do in the case... these chokes are pretty large!
The supply has an off-load voltage of around 19VDC though it doesn't take much to drop it down to 17VDC. This is much higher than I need plus the off-load losses are significant - I was seeing 4W dissipation with nothing connected. Ouch... some of that will be in the 17V zeners, but this all seems like gross overkill.
It will make the beginnings of a good audio PSU but this perhaps isn't what I want for a network supply.
Now I had the NAS, I wanted to try running it with the existing 12V PSU I built early last year, but found that it just didn't have the grunt to cope with the spin-up of the HDs in the NAS. Sticking a big cap on the output only made it hiccup, so figured a better solution would simply be a beefier DC/DC conversion stage.
To keep efficiency, it had to be a synchronous buck topology, and a bit of research suggested that the LM3150 would be a good option. TI sell evaluation boards for these, so a bit of a play in their Web Bench suggested what parts I'd need to reconfigure the boards to do my bidding - they are made on beefy 2oz copper with wide tracks so should have no problem in delivering >10A of current.
Thankfully pretty much everything was available from Mouser. To handle the increased current, the N-channel MOSFETs needed to be changed. As it so happens, TI make some pretty good ones.. the catchily-named CSD16321Q5C is a 25V, 100A (okay, 31A in most real world cases) part which has more than enough current capability, and is also designed to be heatsinkable from the top if you need more current handling!
A literally big component change was replacing the inductor with a much beefier one - a SER2915L-682KL... it barely fitted on the board!
The aim was to try and keep copper losses as low as possible... this inductor is a >=20A part so should have no trouble at all delivering 5A, even in a box with very little ventilation.
To swap over the FETs, I used my new hot air gun, an Atten 858D. I saw the review on the EEVBlog a while back and it did seem to be the best tool for the job for a fairly small amount of money. When it comes to 2oz copper board with wide traces, even an 80W soldering iron will struggle to get enough heat in to remove parts - this is where a hot air gun really comes in handy. I don't think there's any way I would have gotten the solid caps off with my iron.
So I tested up the boards with crocodile clips - success - a clean 12V output! Although the regulation seemed a little poor... I later found out that the crocodile clips had a whopping 0.2 ohms resistance - horrendous for any kind of significant current testing! Need to make some fresh ones with thick copper cable...
The new DC/DC seemed to be working okay, but the existing 50VA transformer in the PSU box just wasn't going to have the oomph required. Even with a single HD, the cable modem, router and NAS were going to peak at 35W or so consumption when the drive spins up... the 50VA will be well out of its comfort zone.
Fortunately I had an 80VA handy... the catch being that it's a 2x9V rather than a 2x12V. Standard transformer voltages are a little bit annoying... the standard windings go along the lines of 6V, 9V, 12V, 15V... thing is that no transformer offers perfect regulation so there can be a significant disparity in off-load vs. full-load voltage. A 12V transformer after rectification and smoothing is likely to be something like 17VDC with only light loading, unless you put a big choke in before the cap.
2x9V would have been perfect for a 12V supply, but I need more than that to charge a battery... ideally between 13.5V and 13.8V for a 12V, depending on temperature.
To handle UPS functionality, I decided to go with a PicoUPS board.
This is quite a neat little board which is based around a couple of ORing controllers that let it use beefy NMOS as diodes, essentially... the controllers have a comparator which decides whether to switch battery or mains sourced DC to the output... when the mains voltage drops below the battery, it should seamlessly switch to the battery.
In order to charge the battery, the board uses a P-channel MOSFET with a little controller IC that appears to provide constant current. As far as I know, it doesn't apply any continuous voltage limiting, although the board itself should not be run above 17VDC input.
While I had a nice efficient UPS solution now, the transformer was a worry. 9VAC would barely rectify to above 14VDC, even off load... ideally this needs a 10VAC output transformer but such things are custom jobs and as such very expensive. The fact that the transformer is epoxy-sealed eliminates the possibility of adding extra turns... so let's try to make things as efficient as possible...
After some searching, I found what appear to be the very best diodes you can buy short of doing it all with MOSFETs and a controller.. the "Super Barrier" range from Diodes, Inc. The forward voltage drop is considerably less than any other diode I encountered, so combined with thick 30A copper cabling, should help to minimise losses. The particular model is the SBR10U40CT - a 2x5A rectifier in TO220 format.
Technically it is a dual diode, but as far as I can tell, they are fabricated on the same die and thermally connected so providing the resistances are well matched, should be okay to wire in parallel. I wouldn't recommend running them up to 10A in this configuration but it should be plenty good enough for 5A continuous, I'd imagine.
A simple means of mounting them was to use the mounting hole for the toroid and drill/cut out a piece of stripboard to mount them on. These things run very cool - not surprising at all given their very high efficiency. You do have to wonder why they even bothered putting them in a TO220 package as it's kind of self-defeating... the heatsinking is so good that they don't get warm enough to give anywhere near their best voltage drop performance.
The thing about diodes is that the voltage drop tends to decrease as the temperature goes up - so you'll get the worst performance Vf-wise at room temperature or less... at 85C, you'll get a considerably lower drop. On the minus side, the leakage is much higher and the devices lifespan will be impeded. Swings and roundabouts.
In any case, when I get around to buying one, cutting off the heatsink tabs with a high speed disc will probably help save a fraction of a volt - before doing this, will need to check the part temperatures in a closed box after sustained use as they may well be getting much warmer now!
Putting it all together... in goes the new DC/DC converter board, face up. Plenty of mylar tape to prevent shorts with the bottom of the case, lots of insulation on the mains wiring. There's a switch on the front to give the option of disconnecting the PicoUPS from the DC/DC converter, just for when you *really* want to cut the power.
So does it work? Indeed, yes! Voltage to the battery with the NAS powered down is about 13.9V, within acceptable limits. When the NAS is operational, voltage to the battery decreases to about 13.5V, which should just about enough to float charge it but probably marginal.
Switching off the mains seems to give a seamless switch to the battery, and it seems to be able to power the router, cable modem and NAS with no problems... I think it would have a tougher time starting them from battery, but definitely okay for switchover!
Efficiency seems to be about the same as before. The new gigabit router uses about a watt more than the old Linksys, so now up to about 8W with the NAS switched off. Not bad, though, given that now includes trickle charging the battery.
I should note that I haven't put in undervoltage protection - this was present on the old DC/DC converter board but didn't have a chance to figure out a good solution for the LM3150... this essentially means that in an extreme situation, the battery would discharge to the point of damage. However, this would be a very extended run on the battery (without NAS, I figure at least 4 hours) and the equipment attached is likely to stop drawing much current before things get really bad for the battery. Hopefully the desulphator won't get called into action too often!
