GXF22 - Very Low Resistance FET & Constant Current Driver with Fuzzy Logic + Microphone

Thanks everyone again for your kind words.

I don't claim to have the skills to be able to implement a neural network on a micro, but fortunately, lots of other people are. Here's an example: https://github.com/ARM-software/ML-KWS-for-MCU. Should be able to get this to run on a Cortex M4 core. But let's not get ahead of ourselves; maybe a future project, but definitely not for this one right now!

Unfortunately it would be hard to add phone connection to the driver (e.g. via bluetooth), mostly because of how the driver is completely surrounded by metal, and a bit too much time needed to write the firmware for both the driver and the phone~

Anyway I hope this driver will be useful in the end since the main useful features are the constant current drive mode, good temperature regulation, good firefly mode, aux led control, and a little bit of fun with the microphone. The DD FET mode is impractical in my opinion, just a for-fun thing, and optional to even install that half of the driver.

The maximum sustained thermal load the D4 can sink will lead to only a few hundred lumens output.

@gchart, this means thermal sinking of the CC FET is not as crucial as one may think. That said, I found a perfect silicone pad to interface between the fet body and the metal D4 shelf, so thermal sinking is decent.

@i42dk, glad this feature is something you like.

@WTF, yes my main hobby is in big power electronics. For my coils, I'm just standing back to enjoy the show, and to keep safe. They do light up fluorescent bulbs and other low-pressure tubes from a distance though!

Have a great day everyone!

Very interesting and inovative linear driver. But I have one question whats mean no-nonsense linear drive. You just no use a sense resistor for current measurements? What method do you use for current measurements. Maybe you just measure the Rdson of the MOSFET via voltage drop. Also do you use dedicated Charge pump IC for mosfet driver, I assume you also use some sort of transistors driver to turn on power MOSFET. Most of charge pump IC only have 2X of input voltage so it is interesting how do you boost voltage to 10V from single battery.

ha i think loneoceans mean ‘no nonsense’ meaning it is a linear constant current driver that just works without trouble. I think i see sense resistor in the photograph which is R1.

also my guess is loneoceans using some charge pump gate drive ic like this one: https://www.analog.com/media/en/technical-documentation/data-sheets/19812fs.pdf. for example it has charge pump with “internal voltage tripler allows gates to be driven without the use of any external components.” This explain 10v from one cell, but maybe not this ic since it has 7.5V max, but still good enough!

Thanks.
For lower cost of driver to make a cheap charge pump we can use something like these EEVBlog #473 - Microcontroller Voltage Doubler - YouTube
But integrated method with mosfet driver IC is better but there is no so many chips which to operate at lower power supply voltages like these LTC driver which also have a integrated charge pump.

IMO, I think the main limiting factor of such low resistance drivers are the contacts.

It’s not too much of a problem with FET drivers(unless you want absolute maximum output), but with boost drivers like the other lonoceans made, contacts with removable cells are a large bottleneck.

Hmmm. Alexa and Cortana are taken... How about

Emisar Nightlight

or

Emisar Ignite

or....

As Lexel write there are another limiting current resistances, like PCB copper thickness, pcb width traces must be huge for lower resistance and so on. It will be interesting to view how much power PCB board can dissipate from MOSFET to host which is main limiting factor for that driver. For best results the board must be 4 layer design and small thickness. So that driver is very similar as Texas Commander approach from 2 years ago Texas Commander "TC" Constant current Opamp driver without PWM . Aslo very interesting project which has been abandoned for thermal issues with heat dissipation.

One way to do this would be to use thermal adhesive/a PCM material to transfer the heat from the chip to the light.

looking forward to see how this driver will do, it seem like loneoceans has thought about thermal since it seem like he chose fet for lowest junction to case thermal, and he mentioned some silicone thermal pad to interface fet body and metal shelf like you suggest blueswordM!

It sounds to me like you are underestimating loneoceans. The gxb172 driver he created makes a lot of heat and power and works fine. There has been some pretty sharp people here try to make a driver like that and gave up after multiple tries. I don’t know of any commercial 17mm single board driver that makes that much power and they are designed by experienced professionals. He’s big into Tesla coils, designing and making the power electronics. Tesla coils make thousands times more voltage and power than a car ignition, everything needs to be considered. Flashlight drivers are easy in comparison. He’s very humble and polite just like Del was, and like Del incredibly good at this stuff.

I know a lot of people are focusing on the low resistance aspect of this driver, but it also looks like a simple, straightforward constant current driver that does away with 7135’s.
I’m hoping it can be adapted to use multiple cells. 2S 21700 or 4S 18650 driving an xhp70 looks like fun.

I don’t underestimate the loneoceans skills. His switching drivers are great and very efficient. In best cases we will have with them very low heat losses around 5% and here we talk for linear driver which need to dissipate all losses as heat power from MOSFET. Little example if we whant to drive something like new XHP50.2 3V version which consume about 6A at 3V and if we have fresh battery around 4.1V so we need dissipate 6.6W from MOSFET through PCB without ovverburn it to host. So thermal design of PCB is most important in that project because is act as heatsink. Just read Texas Ace consideration in Texas Commander thread.
If someone interested I find these great Study from NXP about PCB heat dissipation and device enclosure https://assets.nexperia.com/documents/application-note/AN11113.pdf
From that AN we can see that most important is thermal transfer under PCB for best coolling. But in case of most LED drivers the thermal contact of PCB is only around of the edges which is not the best. Also colling from components packages to surrounding environment like air is not also very efficient. With thermal compound we have improvements. Also cooling botom side of PCB under components is best aprroach at all with thermal encapsulation.

Note that top cooling with partial encapsulation scores great in that study as well.

Put a dab of a weak thermal glue on the driver to mount it in place like Emisar does and call it success. Simple, cheap. Inconvenient but possible to replace.
If the glue is hard (as in “incompressible”, not necessarily “tough”) it will also transfer any impact energy to the body, greatly improving drop toughness.

In addition with like these we will have very good improvement over thermal resistance between PCB components and flashlight host.

This is what I like about it most. What interests me in a linear driver right now isn't FET performance, because I can always solder wires directly to the MCPCB and battery and beat it. What I'm interested in is reducing the voltage drop across the driver to maintain regulation longer, having true constant current regulation (rather than a PWM'd FET that still loses output as the battery voltage decreases), and then the normal flashlight features like low voltage protection, reverse polarity protection, support/adapt for 2S/etc., working with tail-switch vs. e-switch, supporting lighted tails.

I think the real limit we are always going to be up against is heat. I have to hope for LED efficiency to continue to increase, but it's possible the next big BLF thing would be proper active cooling. Some real heatsink fins (not the "fins" on the outside of tacticool lights that are over 1mm thick, but stuff resembling CPU heatsink fins) and small fans to pump air through them. I can already see the first scandal when muggles hold the flashlight wrong and cover the air intake. Water resistance will be an exciting challenge... Anyway, we can start using thermal pads or adhesives for drivers if we need to, we can move FETs to MCPCBs as Led4Power has already begun to do, we could even start mounting drivers in the tail so the flashlight heats from both ends (and really easy tail E-switches then). That kills the aux LEDs, FET on MCPCB, and temperature sensing, though.

A ribbon connector could probably solve all of those if you really wanted to.

There are also other ways.

Using a finned battery tube can help a lot in terms of heat dissipation.

Using PCM materials with enough finning can help extend high power use.

I think the Nitecore unibody design of EA45S is very good example for efficient heat dissipation through the flashlight body.

EA45S indeed looks ok, especially for a cold rod. But I believe it could be improved further. :wink:

Yes, also Zebralights which don't come apart at the head either.

Thanks everyone for the comments and feedback.

Yes system thermal capacity is a known issue right from the beginning - driving LEDs at such high power levels necessarily generates a lot of heat, and no small, compact flashlight will be able to support that kind of power handling capability for more than its initial thermal capacity.

I designed the GX series of drivers with this in mind - to create the highest power handling capability, but not for practicality. Like I said, I recommend looking at the GX drivers as a for-fun driver, capable of producing a lot of light as a bonus feature, and then throttling down sensibly to reasonable power levels to maintain operating temperature.

For this driver, removing heat from the pass FET is important, and in fact, is a fundamental part of this design. Components were specifically chosen, to interface with specifically chosen thermal materials and thicknesses to ensure optimal thermal performance, and optimized to be not ridiculously expensive and still easy to construct. Essentially, the driver is heat-sunk to the thick shelf of the D4 as much as possible via a thermal pad and thicknesses match for a good fit. Fuzzylogic software then takes care of the thermal throttling thereafter.

I'll find some time to post more photos and videos and more details of my completed build soon. Meanwhile, keep those suggestions and feedback coming since this is still afterall a version 1 build which I whipped up real quick together and I'm sure has lots of room for improvement.