The Texas Buck driver series, Q8 / Skyray King 2S/4S buck driver RELEASED!

You could post it up though as highly experimental anytime. Some wildcatter might take an early stab at frying some leds with it.

I guess the right thing to test it on at first is 4S xp-l. I don't know if that's a reasonable use case, but they should handle full voltage without dying... maybe.

Yeah, once we are happy with the design and we do not think there is anything else we can do without actually trying it out I will release everything like normal.

The cost is a turn off for me, I just don’t see me building these for lights that cost as much as the driver itself. I like modding but I am also practical.

That said I want to see this come full circle, I hate unfinished projects.

More exploratory stuff:

http://www.linear.com/product/LT1910

Can drive the n-channel current sensing mosfet, but needs 8 V. The sense resistor there seems uneeded. It's just an overcurrent protection, redundant.

A bit complicated, but those three sense resistors take a bunch of real estate. I'll keep looking for better ways.

http://datasheets.maximintegrated.com/en/ds/MAX1614.pdf

takes 5V, need to look at the details more though. There seem to be several of these available.

Another option is just to turn the whole buck loop upside down and low side switch it. Hang the diode and the cap off Vin, put the inductor and the fet on the ground return, (some Ti drivers work this way by default), have to think about how to charge coff.. still have to flip the gate output though I think, so still need a small fet for that, but no charge-pump controller, big re-arrangement.

I'm kind of liking this though. Why wouldn't this work? In terms of components, just replacing three giant resistors with a level-reversing transistor on the gate drive, and using the current-sensing n-channel fet. Seems like a win. What am I missing? Maybe the level reversing isn't so easy. It's not like an attiny pwm, we don't get to drop the whole controller reference voltage.

I have been known to fry a thing or two. I’m game when ready.

How do you want to run it? Can you deliver 4S battery to it and wire 4S xp-l/xm-l led's? I'd run it 4S battery and 2S led personally for xp-l/xm-l, but it's more dangerous while the bugs are being worked out (because if it opens up too much you're putting up to 16.8V on 7V worth of leds). 1S to 1S might be reasonable too. Some of the components should probably be tuned to your application, especially the sense resistance and maybe RoffCoff.

I just used my quad xpl hi but will order another one. Will PM TA for host advice.

Just saw that current sensing MOSFET is 16mm x 10mm. Still, it would replace the existing fet and the three big resistors, but needs a couple of small components to go with it too. They're also $3 !

Well I guess an SRK with the SRK tailcap (more money) is the best. I don't know what TA has done with the battery contacts here but I guess it should work with the tailcap board to be configurable in 1S 2S or 4S battery input. Of course the BLF Q8 will be the best.

lol………yes. It has begun

I’m not going to pretend that I understand all of this thread, although I did read it all. But if you aim for 3A to 4 XHP-35’s you’re likely to be popping emitters. It’d be best to aim for 2.8A. 3 is right on the edge with virtually no tolerance consideration.

Ya’ll were busy last night.

I don’t have time to see what those high side drivers actually do, mind giving a quick readers digest version? That said we need to keep in mine that if these work out it would be nice to shrink them down to smaller drivers, so less components = good.

I thought about flipping the circuit around but I think at least for the prototype we should try to stick as close to the baseline diagram as possible the more we dig into this, that way we can see what works and what doesn’t and work out the issues from there.

The current sensing mofset being that large (and expensive) puts it into the keep it in mind if the sense resistors don’t work catogy IMHO. Fitting that on the board would be quite difficult and we would have to eliminate it for any smaller drivers anyways, might as well stick with things that are more easily scaled down IMHO.

Well this is being designed for the Q8 but since that will be awhile the next best option is a 4x SRK. Then obviously a custom shelf to handle the heat. A bit of work for sure until the Q8 comes out.

I have not even worked on the bottom side of the driver yet, but it will be like the other multicell TA drivers I have.

We are just designing it for max power right now, the power can be stepped down easily but going up is another story.

Although this driver should have MUCH less ripple then existing buck drivers (largely because of the new IC and the larger size) so it will not be as hard as some. Either way I figure finding the max amps for the XHP35 is something to be figured out later. I actually plan to order an XHP35 with intentions of finding it’s limits when I have some extra funds.

I've been thinking in some detail about flipping it around. There are three interface points:

a) coff.. This is trivial to deal with. Really I don't like how it is now anyway. You can just charge it straight off Vin, it won't be a problem.

b) the gate driver, that's your first question.. cliffs in a minute

c) The voltage sense. Will the LM3409 mind if the sense resistor is biased very near ground? It can certainly handle it at near 4 volts, within spec.

There's no other problem. Solve these and it all just works.

I've been working on b but first the reason.

That mosfet I showed you has an adjustable low-loss current sense, but it's an n-channel mosfet (better anyway).

So, b then. Need to convert off at 16.8V and on at 10 to off at zero and on at well 5 or more.

That's what those drivers can do (maybe not in the end), but actually all you need is a very small pfet. That's it. You drive the nfet from the drain of the pfet. You'll be driving the gate at 16.8V which isn't ideal (more switching loss) but WAAY better than doubling the heat through those fat sense resistors. Done. solved.

c then is my only problem. Will there be any issue sensing across a resistor with near zero offset voltage. I'm not sure yet. One way is to try it.

I don't think this anything to sneeze at. Going from 1 caps and pfet to 3 or 4 caps and nfet reduces heat from 9 watts to 1.5. That all adds up in a big way especially for performance. Sure there are some details, but I'm working through them. Nothing sinister or hidden. It's not complicated really.

It sounds reasonable, assuming the LM3409 does have any catches that would cause issues with this internally. Although how big is that gate driver? If it is too large to fit on anything smaller then the 46mm SRK then we would be doing all this work for this one single driver and have to start over for smaller versions. Seems like it would be wiser to find a scalable design that works for all of them (be it with different current capability ect).

Regardig the cin thing (an issue by itself anyway). Now coff is charged off vout. Vout is just a constant voltage though, or close enough for a give led configuration that changes don't matter at that level in our application. The coff time is just the cap charging time from that voltage. You can as well use the cin voltage and a different RC which we have to pick anyway.

The reason they use Vout is because they want the same ripple current at every Vout. But this doesn't actually help us. In fact it kind of hurts. All this and the silly comparator flipping (that actually significantly increases ripple current and does cause problems), is not designed for better efficiency or such. It seems designed to give very predictable power output in different applications, so a 36W ceiling light is not 40W I suppose. The reason they want constant ripple is that average current is peak minus half of ripple, and so that makes calculating average power easier and more linear, especially for high inductor ripple, which is popular because it's cheaper. But for us, if output voltage is changing it's because we've drifted to 100% duty cycle are no longer able to support the same Vf, and power is going to fall off regardless. Otherwise, it just makes the power settings slightly not linear. (level 8 might be 2.1 times higher than level 4) So what. Heat and part variation also make them slightly off. For us it will be pretty close at high power where it matters. (That's not even true, this is the same for either, it doesn't impact dependence on load only dependance on output voltage, and the output voltage change vs load is relatively small).

This does NOT make performance better or more uniform across different LED configurations. It actually makes it significantly worse (I have numbers) and I would absolutely target a build to a particular led configuration anyway and just calculate COff to get the frequency I want at that voltage. You can do that if anything easier and more consistently if Coff is related always to the same charging voltage, Vin. A fixed frequency at all output voltages for a generic setup is better than their frequency curve for our purposes anyway.

Those are the only three interfaces and once dealing with those three things, the internals don't know any different. Forget the gate driver. That was an early idea. They won't work well, or not those anyway. You just need a small pfet to drive the nfet. But yeah the one "internal" thing is if it handles voltage sensing at low bias offsets well. That's not entirely clear. I might buy one and try to test. It's not simple. Need a scope, and well, maybe need to build a buck, probably not though, just a scope and some RC. I'll stare at their pseudo circuit more.

Ok, so the other thing is switching delay, but the IC needs some accomodation for this anyway, and as I've read in detail how this works, I'm very convinced that's fine. As switching delay gets too long you'll get a minimum duty cycle at high frequency since it will take time for the on cycle to shut off, which raises the voltage minimum. 10% would be doing really bad (100ns delay at 1Mhz) and that's a 1.7V increase, and even then you can just cut back the frequency to reduce that for 1S drivers.

The thing that's big is that current sensing n-fet, but it's probably not too big.

EDIT: now I almost forgot what one of those drivers does. One of them actually can drive an n-channel mosfet on the high side, generating the extra voltage needed to drive it. That actually is cool because it doesn't require flipping the buck and might even get around worries about the sense resistor, but I think I decided that one was too large and a bit slow. I'll have to look again. I kind of like flipping the buck better if it will work.

Anyway, I wouldn't abandon this simple version. It's almost done. But if you want we could maybe work out the other one too. I might though argue to run coff from vin even on the simple one. I'll post numbers later. Oh, and if you did iadj RC from the manual, I'm not sure how great that is. They aren't expecting a PWM input. Probablu multiplying C by 10 would about cover it, just as a guess, but we should run numbers. Probably 0603 covers any of it though.