Domestic Led Lighting project.

[HackedFridgeMagnet]

My only query is that to produce a lot of light, you want good efficiency and so a higher voltage and longer led string would be better.

If you want to go lower light levels then go low voltage certainly.

For just a single channel then I think one of the six pin attinys might work.

I am not driving the leds that hard, about 50% of max current plus the instantaneous ripple. They are slightly more efficient at around 400mA than at 1-1.5 amps and more efficiency means less heat, which is very important. One of the downsides of this is a higher initial cost.

You might have trouble routing this circuit on a single sided board although you should manage it if you use zero ohm links to help.

You dont 100% need MCPCBs for mounting leds, you can also use FRP with vias and heat pads and other tricks. Worth considering.

The bjts I use I am very happy with. (Forgot the number but it's xx41 and xx40 I think). Though the ltspice models don't work.

As for fets you need a "logic level fet" for 5v, the fds5680 might actually be ok but you probably want a Vgs(th) a tiny bit lower. If you switch at 9v it would be fine.
Look at the Vgs vs Rds(on) graphs on the datasheet. Digikey lets you search on Logic Level fets.

[T4P]

There are two categories of high power LEDs :
Standard high power LEDs : usually around 900~1400mA random Fv
Flashlight type LEDs (higher efficiency) : random forward current 3.3V nominal Fv

So you see, one thing's for sure though, like all SMPSs, they work best when drawing lower current but high voltage and that a SMPS is best when used around the 50% load mark.

[sarfata]

Thanks again HackFridgeMagnet for your insights, and thanks T4P for jumping in!

To answer both your comments, I am using a Ledengin 10W RGB led that has R/G/B Fv of 2.0/6.4/3.2 and I am trying to find a good solution(here good also includes efficient although it's not my #1 priority) to control the three channels with one AtTiny. My goal is to build a current controller cheaper than "ready-to-go" package like the LT3518 I was using earlier or equivalents that cost $5 to $8. Turns out to be a very interesting challenge with lots of learning on the way

So I have built the design I sent two days ago. I used BC547/BC557 for the BJTs and a FQPF13N6L for the Mosfet. The inductor is 15uH, the LED capacitor 4.7uF and the zener is a 1N5361BG.

I have added two buttons that allow me to change the Ton period. The Ton+Toff is 8uS right now. All of this is breadboarded of course


This is a capture of the voltage at the gate (red) and at the drain (yellow) of the mosfet. You can see that it looks like the mosfet starts conducting current again before the beginning of the new period. I had never seen this in my simulation, any idea where that might come from?

Where do the huge oscillations at each state change come from? How could I reduce them?



This is what the PWM signal looks like out of the AtTiny (in red) and what the output of the RC filter looks like (in yellow). As you can see, it is extremely noisy but measuring this with my Fluke gives me very precisely what I expected (that is Idiode = Urcfilter * DutyCycle * 1/0.1ohm).



I have decoupling capacitors on VCC and on my 5V. I also added a 10nF capacitor on GND/VCC of the Attiny otherwise it kept resetting every time I increased the duty cycle. I am now fighting with the ADC converter, without gain and using VCC as the reference, it returns 2.5V for the rc output. This is not right... will look at this more tomorrow. I am hoping I will still be able to read something through the noise.

thomas

[sarfata]

Update: just found out that the bogus "mosfet conductions" only happens with a small duty cycle. If I increase it, it does not happen anymore. Also, if I remove the capacitor on the LED, they completely disappear at every duty cycle (but the oscillations are even worse).

[HackedFridgeMagnet]

I guess you are breadboarding this design?

If this is the case, it is very difficult to get rid of noise and inductive effects. In fact breadboarding gate drivers is hardly worth the effort, I have used vero board and it was ok. Dead bug might be worth a go.

So shorten all lead lengths especially high current ones.


Simplify your circuit first, even replace the load and cap with a 100 ohm resistor. and just get the gate voltage looking right.
Dont worry about the feedback voltage, it seems as messy as what it is measuring.
Can you lower the freq?

The lower freq oscillations are normal as you may have discovered. It is just when the Fet and the Diode are both effectively off. The will probably be minimal/no current flowing at this time.

As you know we need to get very close to a square wave on the gate. the top of the gate voltage square wave should be comfortably above the Vgs(th).

[sarfata]

I guess you are breadboarding this design?

If this is the case, it is very difficult to get rid of noise and inductive effects. In fact breadboarding gate drivers is hardly worth the effort, I have used vero board and it was ok. Dead bug might be worth a go.

Yes I am. I wanted to breadboard it to make sure I could understand every step along the way. I did get a brand new vero board to do something a little cleaner.

Simplify your circuit first, even replace the load and cap with a 100 ohm resistor. and just get the gate voltage looking right.
Dont worry about the feedback voltage, it seems as messy as what it is measuring.
Can you lower the freq?

I did try the mosfet with just a few resistors in parallel. i had also removed the inductor. It worked fine.

I could lower the freq. How do you think this could help?

The lower freq oscillations are normal as you may have discovered. It is just when the Fet and the Diode are both effectively off. The will probably be minimal/no current flowing at this time.

Well, I cant say I really understand it It seems to me that if everything is discharged and the mosfet is off, then the voltage at this point should be 9V, I dont get the spikes to 0V.

As you know we need to get very close to a square wave on the gate. the top of the gate voltage square wave should be comfortably above the Vgs(th).

I understand that well. The datasheet for this Mosfet (http://www.fairchildsemi.com/ds/FQ/FQPF13N06L.pdf) gives a Vgs of ~2V for 1A, I feel like I am safely above that.
What I am not sure about, is the positive spike when the mosfet turns on and the negative one when it is turned off. In my capture you can see a spike of +2V on the gate when the mosfet is turned on. Am I correct assuming that this is good and helps make sure that the switching time is as short as possible, and therefore the mosfet has the lowest possible resistance?

Thanks again for your insights!

thomas

[HackedFridgeMagnet]

I was just thinking that we should be putting this stuff onto your thread, instead of mine. No harm done but it would make it easier for other people and myself to refer to the one design in one thread. As much as possible anyway.
Hence I will give some opinion to your queries in your thread.
https://www.eevblog.com/forum/projects-designs-and-technical-stuff/looking-for-feedback-on-my-design-of-a-high-power-led-driver-9612/

[sarfata]

Hey,

Wanted to post on this thread this very interesting Application Note from ST Micro which basically describes what you have been doing:
http://www.st.com/internet/com/TECHNICAL_RESOURCES/TECHNICAL_LITERATURE/APPLICATION_NOTE/CD00263933.pdf

Two very interesting differences:
 - They dont use BJTs and drive the mosfet directly from the uC. Have you tried this? I thought the AVR would not be able to sink enough current but this got me thinking twice ...
 - They measure the current at Ton/2 to get the most accurate current reading. I think this is pretty smart (you probably have to read the article to understand this part

thomas

[HackedFridgeMagnet]

That's great Sarfata, thanks for posting.
I will try to see if I can find any improvements I can make.

[HackedFridgeMagnet]

I have still been working on this, when I get the chance, but I haven't updated for a while.

I got my 3 channel boards back and they work nicely, nothing gets hot on the driver at all, I only really made one error and that was one of my transistor footprints was wrong, so I had to flip the transistor over and do some ugly soldering.

The next trouble I found was the diffusing  of the light, I wanted a prismatic glass diffuser because the glass looks so nice when it is off, but it cannot quite diffuse the light enough, so you get this effect called dazzle. I think if I combine the diffuser with a very light pearl diffuser it may work but that can wait.

My next plan is indirect strip lighting, narrow strips hanging of the wall but giving room lighting, the indirect light is very pleasant though I guess the efficiency is not so good.

To do this I have made a few changes my 3 channel board to 2 channel, changed the format from 100mm x 100mm to 50mm x 100mm. to suit Seeed studio.
This thinner board will sit better on the strip and each channel will go off in opposite directions along the strip.

Other changes for version 3.
I added a thermistor,
fixed a problem with turn on where the leds flash momentarily at power on,
added a switch to completely turn off power when I want the light off but doesn't increase the length of the power cable run
made the supply 36 volt, up from 24v so I can run 10 leds in series per channel.

Anyway I sent Version 3 off to Seeed and I haven't received them back yet.
But since then I have realised how many mistakes I put onto Version 3 of the board while plotting. The board is a complete and irretrievable failure.
I didn't send the solder mask layer, I sent instead the solder paste layer.
All my 8 bjts are inverted.
My kelvin connection ground got flooded by the zone flooding.
this is what happens when you aren't careful and are rushing.


Last night I sent Version 4 off to Seeed, after fixing these mistakes.