To me this all seems MUCH easier than assembling a driver (which you still have to do but...) and it's quite a bit cheaper and faster than ordering new stars. If a few cuts in traces, and existing jumper points (the extra led pads, or just add proper jumper pads) can make this reconfigurable is it really too late to push it through?
(are they heat sink pads case grounded? do they need to be avoided electrically?)
Yes, the MCPCB is DTP. Like I said it could be converted but I personally hate cutting traces, I have always had massive issues getting a clean cut that doesn’t short out.
The $4 for new stars would be well worth the cost for me but to each their own.
Solved all that:
http://imgur.com/gallery/ebS7c (well solved in theory, I'd have to learn dip trace to actually solve it)
4p as shown, positive is always to the left (or negative if you prefer).
Can do 4p, 2s, or 4s. No cutting, only jumpers (4s requires jumpering two of the wire pads, although there is room to route a trace for a zero-ohm resistor instead for that too, but jumping the wire pads seems ok). LED's do not need to be reversed. This should work as a stock board. (I deteted the thermistors because they were in my way at the time, but this is just a concept version obviously)
I had considered this, actually had a design half made up for something like this but decided against it as people would complain that the traces were not big enough.
If Miller says he is ok with trying to change it at this point I can finish my first design and see what the Q8 guys think of it.
Either way it doesn’t effect this driver much. The goal is to design it for overkill and then work backwards for smaller versions. I figure I would like to work down to at least a 26mm and maybe get a 20mm if possible since the mtnmax is not available in that size. Although that would be limited to the tiny13 which is not a great option but thats for a later design.
What's wrong with the one above? The places where traces are narrow, it's obvious how to improve it, just didn't for the concept drawing. Random addressing is fine, but that plus laying it out for short jumpers for the most used paths without ever needing LED reversal is a bit more of a trick. I did try to keep the central pad/hole design the way it was because I didn't know the constraints. If I could put pads anywhere I want, traces get a little better yet, but I'd still do it like this basically.
Anyway, if I inspired movement on it, that's great to me.
There is nothing wrong with it, but if going for a fully customizable board, might as well go all out IMHO. I tossed together another version and posted it in the Q8 thread that allows any setup you could want using jumpers.
The issue is practicality of thorfire actually using it or any of our designs.
Get it to production. There are a number of us that wish it………………when the Q8 is ready. Hope we get an update from TH soon!
Edit.. moved my comments there. This is was buck driver thread.
Ok, I felt if the board had a chance it had to be now. Done all we can on that, so back to the buck.
So I'm mostly finding power diodes with about 500mV Vf. I'll get a summary of something soon. That produces 10% resistive power loss in the diode alone for all 1S configurations, regardless of current level. I don't like that.
But I did find a couple of gems like this:
Which is really slow (and can't find just how slow) and just not meant for this. It's half tempting to put one in parallel with a faster diode though, so it can kick in when it catches up, if it ever even does though.
I think without finding a current regulated synchrounous IC though, that's just reality. We could try to parallel two diodes (or more), but it's not guaraunteed to help much since whichever takes more power, gets hotter.. lowers Vf, and takes even more power. So I'll try to round up the best of the bunch, mostly already done probably.
I was also thinking about parallel diodes.
Keep in mind that the primary usage of this driver will be for 4S setups or 2s setups. For example if you were running 1s 3V LED’s then you would run the driver as 2s voltage. If using 2s LED’s then at 4s voltage and obviously 4s voltage for 4s LED’s.
I don’t see a reason to use 4S input to 1S output in any case. Although having the option is great naturally.
I tend to agree. There's not much reason for 4:1. 2:1 is showing about 3.4% loss for the diode, and 1:1 (close to it anyway) well I haven't run the numbers yet but it will be next to nothing, so it's not so bad.
This is neat:
http://www.linear.com/product/LTC4412
It can turn an asynchrounous buck into a syncronous one as a drop in replacement for the diode, without support from the buck IC.
It uses a MOSFET (of our choosing) to do switching in place of the diode, but instead of being controlled by the main IC, it has a voltage sense. When there's voltage on the cathode, it closes the switch. The body diode of the MOSFET is used to cover the slight delay from sensing. This won't work for a 17mm version I guess. No room for the extra IC.
Anyway. I said before, and it's written many places, that max ripple current is at 50% duty cycle. As with many things technical that statement has to be taken in (a fairly unfair in this case) context. It turns out that's true for equal current. For equal power, the ripple as a percentage of average current is a bit worse or anyway pretty similar at high duty cycle/high output voltage compared to 2:1 operation. Most things though are easiest at high voltage output and get harder/worse at lower voltages.
I've been thinking about total voltage drop issues a little for 1:1 (now included in my tables), and adding some tweaks to the math. It seems to me on first scratch like it won't be an issue.
The input capacitor however is showing up as a very the large majority of the remaining 5% innefficiency in for near 1:1 operation. I'll double check there, but this can be useful info.
Edited commentary. I first increased the capacitance and it got better, but then I realized that's because I calculate ESR from the capacitance, frequency and dissipation factor, so maybe that's not right. Then I realized many of the bigger 0805 caps are just stacks of 2 or three smaller ones. And it seems to me two parallel caps have the same phase angle as one, so the original conclusion is right. More capacitance, less ESR. Also checking some spec sheets dissipation factor doesn't seem to vary so much across capacitance values.
Anyway, I'll see if I can find some cap or combined stack that works a little better. 5% isn't a big deal, but it's probably easy to make it 1.5% instead, which would just be a nice boast.
That is very interesting indeed.
If we kept to a 17mm inductor then we should be able to fit that without much trouble, including the mosfet. With a 22mm it would be real tight but just might fit, although not sure if it would compromise the overall EMF interference due to cramming things closer together.
Although looking at existing buck driver designs, they cram components right next to each other with no apparent side effects, although they also put out a fraction of the current we plan to.
We could try to use the LTC4412 if nothing better presents itself in the Q8 driver setup. For the smaller setups I doubt we will need more then 10 amps of current ability so it should not be as big of a deal anyways.
I kinda figured that things would get harder as we decreased the output voltage. Circuits pretty much always handle voltage better then amps, Ideally it would be something like a 4S input to a 3S output, just enough to keep it in regulation till the batteries are depleted. but thats just not practical in the real world. That is unless you swapped the Q8 out for 4x triples and built a DIY M43 (or M34?).
Well here are a ton of numbers:
Caps are 10 uF 7.5% dissipation factor.
Inductor is the 15uf one previously (lower resistance than the 22)
Diode is 0.5V Vf Mosfet numbers are a bit vague, but I think at least Rdson is about right, wouldn’t trust the switching losses much at this point though. Input voltage is 16.8, output is in multiples of 3.5 for now (I may add a correction for current, especially since it matters in 1:1). The output power at 4:1 is less because I limited it to 15A for now.
I just noticed, the 5% mode on this light is going to brighter than most pocket lights (not BLF ones). I’m pretty sure we’re going to need that PWM for moonlight and even just an indoor low.
So for a true synchronous buck the freewheel mosfet is opened late and closed early to prevent reverse current, using the body diode during the transition. But I realized for this thing that works during opening, but not closing. It's already open and it closes too late.. and I guess there should be reverse current briefly then. I don't know how ideally that thing really works. A very smart IC could predict closing based on the last cycle and only have issues then during load changes.
Good data, the specs look good. Ripple is actually better then I expected. I was thinking we would have to settle for 2+ due to space constants. Less then 2 will be a big step up from what is currently available at 10% ripple.
What is the input cap ripple exactly?
3W of loss through the diode while obviously not good could be dealt with I think. If nothing else you could stack silicon cubes on it to contact the upper shelf. The synchronous IC would be a great option if it works but we might want to stick with a simple setup to start out with and make sure it works as we desire first.
Overall an efficiency of ~90% is what I was hoping for, supposedly it would beat that for 1:1 and match that for 2:1, which is all we really care about. 4:1 is simply for outlying cases that are good to consider in the early stages.
It took me awhile to come to terms with this input cap thing. It seems to assume that 100% of the AC/ripple current component is provided by an input capacitor not the battery. The size of the cap then determines the translation of that current ripple into a voltage ripple and, as there's ripple current through the cap, of course there are losses there too. Supposedly this cap should be connected as close to the high side mosfet as possible, and I guess should also be grounded pretty well to avoid inductive pickup on the ground side of the cap.
Yeah, I agree the synchronous stuff is a bit elaborate. However it takes time to get boards. If we really think this diode replacement circuit could work, could layout pads for both in a prototype board. Might need to get some breadboard and scopes involved for that kind of development though. I do have access to good scopes but not sure I would find time.
It only needs a 10uf cap to handle that kind of power? Seems kinda small?
Far as the synchronous side of things, I am not sure it is worth it, I think we would come out ahead with a larger inductor over that personally and smaller drivers sure won’t have room for it. So it kinda leaves us at square one and doing all that work for a single driver / setup. It can be done, just not sure it is worth the time and effort.
So now that we know the specs we just need to settle on exact components and finalize a design so we can try it out to see how it works.
Just discussion for fun. I think we'll have a traditional diode this round and 4% diode loss in 2:1 is ok with me, but the larger inductor doesn't actually help efficiency until ripple current saturates, so below 3.5W as 200% ripple means you hit zero current. Just means you use PWM from a little higher mark.
As for the cap, it's only handling the ripple current, and at very high frequency, so low charge (I*t) per cycle. This isn't 60hz AC. I'll check the numbers again by hand though. I did just kind of bash in the equation from that article for that one without much thought, other than following the idea about at the detail I just stated. Still I think there's a tendancy to think like the output cap is supporting power during off time, but the inductor is doing the bulk of that work, at least under optimal conditions.