I'm making a 2-3cell 35v booster/driver

[KreAture]

Hey, just wanted some input/feedback on my driver/booster.
I made it because I found some very lovely led flashlights at my local gizmo/EE shop and they are driving an array of 9 LED's directly from 3xAAA cells with just some dropdown resistors to prevent burnout on new cells.
I notice a very big difference bwteen light output the first minute vs later use on fresh cells so I thought It'd be fun to make a regulator for it.

Here's a picture of the torch. It's around 1" dia and 4" long using a 3xAAA cradle for batterys and only cost me $8!
http://www.clasohlson.no/Archive/Images/Products/Hi/362996_X_2007-08-08_101941_833.jpg

I made my own series-conencted LED-board for it reusing the stock LED's and made an additional driverboard that fits behind the LEDs. I used the only FET's I had available at the moment so it has issues at the low-voltage end. This I hope I can fix when I get my new FETs.
I designed the board so the springloaded battery-holder pushes against the board making it a drop-in replacement for the original connector-plate.
Here's the pcb so far:
http://kreature.dynu.com/ee/avr/avr_light2/avr_light2_LED_Driver_v1.2_pcb.png
The driver runns at a whopping 35kHz making the inductor required quite small. I'm playing with algorithms for the current-control and have come to one that seems to work well and can sustain 35v output at 40mA with little modulation.
Tests have shown around 80% efficiency at as low as 2.4v input. (0.8v/cell)

I know this is quite the overkill, but I will be adding different functions later excusing the use of a microcontroller.

So, waddya think?

Am I barking up the wrong tree in running series instead of paralell?
I came upon the impression that accurate current is important in driving LED-arrays and that process variations can cause the Vf of a LED to vary. This would lead to LEDs in a paralell array to draw different currents and possibly fail sooner causing the rest of the LEDs to fail almost immediately. When running much larger arrays (of which this is a prelude), I thought series would be the proper and safe way to do it.

(With the new SO08 FETs I have ordered I could make the driverboard even smaller. No need in my current favourite flashlight, but in case others would want to build their own I'd be willing to do it.
Hardest part is finding a low Vgs FET with high voltage handling. I currently design this with 50v max in mind just in case.=

 

[KreAture]

Here's an updated version of the PCB:
http://kreature.dynu.com/ee/avr/avr_light2/avr_light2_LED_Driver_so8_v1.2_pcb.png
It's really trivial but in case someone wants to understand what it does a little better, here's the schematic too:
http://kreature.dynu.com/ee/avr/avr_light2/avr_light2_LED_Driver_so8_v1.2_sch.png
I am thinking of reducing R1 to 1 Ohm as I will still have about 1mA accuracy. This would cut the wasted power sent to the LED's from 1% to 0.01% and that's almost 1% of the input power saved.
I think this is just as important as picking at the 80 or 81% efficiency. Isn't it?

Oh, and before you mention the lack of filtering on the LED-current and such, I know... I am filtering it in software reducing parts count.

 

[Doh!Nut]

Hi
You are certainly correct that running LED in series is the technically correct solution as it eliminates current variation between caused by varying Vfs.

If you wanted to do more of these torches you could look at Kaidomain or Dx - one of them has packs of seven torches very similar to yours that work out at a couple of buck each.

You could also look at BestHongKong which has LEDs that claim to be 50,000mcd, Then there are the ones that have 3 or 4 chips inside the LED (ie each LED takes 60-80ma instead of the standard 20ma). which could give you effectivly 36LED (4*9) if your circuit could produce sufficient current.

You would then have a really nice bright regulated LED light, but could you have got brighter for less money and effort by using a single high power LED?
I have got one of the torches I mentioned above and am looking at machining a little heatsink to hold a SSC P4, the trouble I think I am going to have is the increased length of the Heatsink/SSC/Optic combo.

The last thought is that a 18500 cell will fit in the space of the 3*AAA cells and has approx 60% more capacity (1600ma*3.6v)

Cheers
Nick

 

[KreAture]

Doh!Nut
Excellent idea about the 18500 cell!
Didn't even think of it! I have designed my circuit to cutoff at 2.7v and it will stay in discontinous boost mode all the way. Max input is 5.5v or 4x reechargeables.

I am now driving 40mA in my 30mA LED's just for the fun of it and my output voltage is 32.4v with a less than 400mV high frequency (55kHz) ripple and a secondary 800mV 1-2kHz ripple. Total variance this effects on current seems to be around 2-3mA.
Is this too much? Will LED's take any damage from this vs a linear regulation at same overcurrent? (Or from same current at all if I set it down to "safe" limits?)

My controller PCB builds 4mm now and is actually thinner than the original connector-disc. I will be increasing the frequency and reducing the size of the inductor in a new version to get the booster down to 2mm thickness and even smaller area.
I am running full pwm regulation btw, not just gated pwm with fixed duty. The goal is to get it up from 55kHz to 250kHz pwm in the next version.

This is way too much fun...

 

[georges80]

Doh!Nut
Excellent idea about the 18500 cell!
Didn't even think of it! I have designed my circuit to cutoff at 2.7v and it will stay in discontinous boost mode all the way. Max input is 5.5v or 4x reechargeables.

I am now driving 40mA in my 30mA LED's just for the fun of it and my output voltage is 32.4v with a less than 400mV high frequency (55kHz) ripple and a secondary 800mV 1-2kHz ripple. Total variance this effects on current seems to be around 2-3mA.
Is this too much? Will LED's take any damage from this vs a linear regulation at same overcurrent? (Or from same current at all if I set it down to "safe" limits?)

My controller PCB builds 4mm now and is actually thinner than the original connector-disc. I will be increasing the frequency and reducing the size of the inductor in a new version to get the booster down to 2mm thickness and even smaller area.
I am running full pwm regulation btw, not just gated pwm with fixed duty. The goal is to get it up from 55kHz to 250kHz pwm in the next version.

This is way too much fun...

Basically a similar concept to my early uFlex driver (see link below).

I used a tiny15 since it has a PLL on board for faster PWM and if I did it today I'd use a tiny45 since you have more code space to do it all in C and also has a 64MHz PLL (if run >2.7V). That would make your inductor way smaller. But uFlex was designed for 1A output as a buck converter.

Anyhow, you'll need to address the drive circuitry for the FET - as you go faster the uC will have trouble driving the gate capacitance - loss of efficiency since it will no longer be fast on/off cycles.

Current regulation and flicker control will also be 'fun'. Being digital the PWM will have 'steps' unless you do fancy software dithering and just one step is enough to make a visible change in intensity. "Real" switchers typically use a comparator and an analog ramp to vary the duty cycle - that gives them essentially infinitely small steps versus 8 bit PWM.

You will learn lots and have plenty of fun (and frustrations)

cheers,
george.

 

[KreAture]

Hi George! I was hoping you'd chime in!
I started with the tiny13 because I didn't have any tiny45's yet. Getting them here in Norway is tedious but they will come eventually...

Using a boost converter in discontinous mode allows me to have smoother control over the output than if it was in continous mode. I can keep the 31v on the output over 2.7-5.5v input-range with 27-75% duty at 60mA output and 25-54% for 30mA load. This keeps me in a fairly linear region and the full 8 bit pwm helps. (I try to avoid over 80% duty as it starts to get hairy very fast...)

I did a new layout yesterday just to see if it could be done...
I put the entire circuit onto the back of my LED-carrier board. The boards LED positions match a standard 9-led arrangement found in many 9-led torches and fits perfectly into the old reflector of my so far favourite torch.
http://kreature.dynu.com/ee/avr/avr_light2/avr_light2_singleboard_9_LEDs_v1.3_pcb_top.png
Before you mention it, yes it is cramped...
My biggest concern is finding a better FET than the FDV303. It has a Vds of 30v oops making it not suited for my needed 31.5v + diode forward + safety margin. It can however handle 680mA continous and 2A peak currents. Quite impressive for a SOT23 package. With it's fast switching and very low gate charge it's scary close to ideal for this small space. I am so mad at it for not handling more voltage. It's gate threshold isexcellent too btw... (Datasheet for this, one of my favourite small FETs: http://www.elfa.se/pdf/71/07109655.pdf)

Any similar FETs you know of with higher Vds would be very much apprechiated.

 

[KreAture]

Wohooo!
I finally got around to test the current prototypes actual efficiency.
Output is fixed at 40mA and 31.7v for now. (It ends up as 31.7v due to the current-control at 40mA) and to deliver this it needed 370mA at 3.75v! This works out to 1.268 watts output from 1.3875w input or around 91%. I am using a 10Ohm shunt so this eats 0.016w making actual output to LEDs only 1.252w. Overall system eff at 3.75v input is thus around 90%. Not shabby at all... With higher frequency and a bit poorer FET I expect more switching losses but I will also get less inductor losses with lower resistance in inductor. Will be interesting to see.

 

[PEU]

I think you will have more feedback about your circuit in the Electronics subforum, if you think the same, ask a moderator to move it there

cheers


Pablo

 

[KreAture]

Good idea!
I would have posted it there if I had seen that forum
Any mods around? Could you move it?