Looking for a specific DC CC driver for LEDs

[karimwassef]

Here's what putting a lens does:

<a href="http://s1062.photobucket.com/user/karimwassef/media/LED%20experiments/4_zpsvw5rsonh.png.html" target="_blank"><img src="http://i1062.photobucket.com/albums/t496/karimwassef/LED%20experiments/4_zpsvw5rsonh.png" border="0" alt=" photo 4_zpsvw5rsonh.png"/></a>

Going from 80 degrees without to 55 degrees with changes the covered area. When I compared the light power L (W) using the same formula, the radiated power with and without the lens were the same! This, at least, validates the calculation.

There is some variability in the I x V when there really shouldn't be, but I think that's part of my measurement error. I've tabulated the results - maybe there's something there, or it could just be error.

 

[karimwassef]

I also found that temperature is probably an important variable to capture. At higher currents, the I x V relationship changed to a different linear fit. I attribute this to a higher resistance due to higher local temperatures...

Here's what it looks on one of my LEDs and on theatrus's data

<a href="http://s1062.photobucket.com/user/karimwassef/media/LED%20experiments/8_zpsimqe5klg.png.html" target="_blank"><img src="http://i1062.photobucket.com/albums/t496/karimwassef/LED%20experiments/8_zpsimqe5klg.png" border="0" alt=" photo 8_zpsimqe5klg.png"/></a>

I captured the calculations for hot and cold, but I'm not convinced that the demarcation is very clear... so I also ran the calculations for all data together. It's all in the tables.

 

[karimwassef]

So, with power P(W) now separable into light power L(W) and heat power Q(W), we can calculate efficiency

<a href="http://s1062.photobucket.com/user/karimwassef/media/LED%20experiments/5_zps8j3dgzar.png.html" target="_blank"><img src="http://i1062.photobucket.com/albums/t496/karimwassef/LED%20experiments/5_zps8j3dgzar.png" border="0" alt=" photo 5_zps8j3dgzar.png"/></a>

 

[karimwassef]

Here is the table of results so far. This doesn't separate the data into cold and hot.

<a href="http://s1062.photobucket.com/user/karimwassef/media/LED%20experiments/1_zpsjffjxus1.png.html" target="_blank"><img src="http://i1062.photobucket.com/albums/t496/karimwassef/LED%20experiments/1_zpsjffjxus1.png" border="0" alt=" photo 1_zpsjffjxus1.png"/></a>

Here, I attempted to fit the hot and cold using power loss and the quality of the fit for guidance. I'm not very happy with it, but I'll share it in case others can make use of it also - or see a pattern in the data

<a href="http://s1062.photobucket.com/user/karimwassef/media/LED%20experiments/0_zpslyqt4qm5.png.html" target="_blank"><img src="http://i1062.photobucket.com/albums/t496/karimwassef/LED%20experiments/0_zpslyqt4qm5.png" border="0" alt=" photo 0_zpslyqt4qm5.png"/></a>

 

[karimwassef]

I'll go into what I think the table says, but here's my main conclusion:

<a href="http://s1062.photobucket.com/user/karimwassef/media/LED%20experiments/7_zpszl5xbkcf.png.html" target="_blank"><img src="http://i1062.photobucket.com/albums/t496/karimwassef/LED%20experiments/7_zpszl5xbkcf.png" border="0" alt=" photo 7_zpszl5xbkcf.png"/></a>

At least for the LEDs I have tested so far, it's possible to determine a good chip from a deteriorated one by looking at the resistance obtained through regression of I vs. V. LED # 4 so example has 5x the normal resistance of a comparable 100W blue chip and the PAR is significantly lower.

So I can envision data collection every morning when the lights come on. As the current is slowly ramped, each LED is interrogated to capture the I vs. V. The slope determines the resistance and that resistance is captured over time and temperature. Over time (months), the resistance of a poor LED would increase much faster resulting in a higher voltage needed to drive the same current. I would put a 20% increase in resistance as an arbitrary threshold for the end of life of the LED. The rate of degradation can also be captured - warning of the need to prematurely replace a chip.

 

[karimwassef]

Ok. So here's my interpretation of the data:

<a href="http://s1062.photobucket.com/user/karimwassef/media/LED%20experiments/1_zpsjffjxus1.png.html" target="_blank"><img src="http://i1062.photobucket.com/albums/t496/karimwassef/LED%20experiments/1_zpsjffjxus1.png" border="0" alt=" photo 1_zpsjffjxus1.png"/></a>

First, the efficiency of most good chips is 15%-20%. This means that a 100W chip only generates about 15W of PAR power. The resistance is around 1-2 Ohms.

Paving the power breakout means that we can also break out the current, resistance and voltage resulting in heat vs. light. The question is what model to use.. resistance and voltage in series,.. or parallel... or both.

The tables assumes parallel, so the voltage across both is the same, and the current splits.

An alternate assumption is to put them in series, so the current across both is the same and the voltage is different. Here's what that looks like:

<a href="http://s1062.photobucket.com/user/karimwassef/media/LED%20experiments/2_zpswjsuwsad.png.html" target="_blank"><img src="http://i1062.photobucket.com/albums/t496/karimwassef/LED%20experiments/2_zpswjsuwsad.png" border="0" alt=" photo 2_zpswjsuwsad.png"/></a>

So here, the fit results a negative resistance for light power... which isn't physically possible.

I guess that's expected since the closest model should have resistance in parallel and in series.

I haven't quite finished a model that includes both, but I thought I'd share what I have so far. The idea is to force the resistance for the light power to zero (voltage only to drive light output) and then fit the model.

 

[karimwassef]

This thread started looking for a DC-DC converter and theatrus has actually developed one that works quite well. It's evolved into a "smart" power supply that interrogates the LED lights and determines their health.

I'll keep working on a model, but I would like some feedback

 

[theatrus]

I went back to my current shunt and I found that it wandered by 0.5A. I'm not sure why... I decided to bypass it and use my handheld multimeter for the current measurement. I have two so I checked both and got the same current (to two digits), so I'm using that.

Under load? Do you know the tempco of the shunt?

 

[theatrus]

This thread started looking for a DC-DC converter and theatrus has actually developed one that works quite well. It's evolved into a "smart" power supply that interrogates the LED lights and determines their health.

I'll keep working on a model, but I would like some feedback

Cool!

I still owe you some alternate test data. I got the lab power supply stack assembled again today, and will setup for a current/Vf/PAR/Temp measurement on the RB chips I have.

 

[karimwassef]

Thanks theatrus! Nice equipment... let me guess - auction sale during an asset reduction event?

 

[karimwassef]

So.. I've figured out why I'm getting a negative resistance for the radiated power and why the I x V curve looks like a downward quadratic fit.

It's been a while but it's coming back to me. The formula is an exponential

So.. a little more complex than I was looking for, but we just need a simplified enough form to extract key coefficients, even if they are not representative of real elements like resistance.

 

[theatrus]

Thanks theatrus! Nice equipment... let me guess - auction sale during an asset reduction event?

Yup! There are enough large companies in the area disposing of stuff

 

[karimwassef]

What do you think of the results?

I didn't expect the efficiency to be only 15%. I wish I could do the same for my halides to compare. If I put the probe 12" away from the bottom of my 400W MH bulb and use the radiation area from the reflector to run the math, it may provide some usable comparison?

With the LED lens, the pattern is a simple cone and one angle measurement at a known height is sufficient to describe the radiated area very well.

 

[theatrus]

15% is shockingly low. I'd expect closer to 30% for a decent setup (with 40%+ possible).

I don't have access to an integrating sphere for proper power measurements though.

After some rummaging, I got a better multichip cooling solution in place: a socket 2011 heat pipe tower cooler. The case heatsink interface is a mere 45C when running at 4A.

The Vf at 4.000A is 37.96V for another quick data point.

 

[karimwassef]

My fit of your last dataset would put it at 36V for 4A. Higher voltage would indicate more loss. It could be more light output, but with my LED, the higher voltage was more heat.

How's the PAR set up?

 

[karimwassef]

Well, the exponential fit didn't work very well. It would predict reduced radiated power at higher currents, etc...

I got a hyperbolic fit that was best. Better even than exponentials, logs, etc...

<a href="http://s1062.photobucket.com/user/karimwassef/media/LED%20experiments/0_zpsxq2nvzlw.png.html" target="_blank"><img src="http://i1062.photobucket.com/albums/t496/karimwassef/LED%20experiments/0_zpsxq2nvzlw.png" border="0" alt=" photo 0_zpsxq2nvzlw.png"/></a>

This basically predicts that the LED saturates at a given power level and a radiation impedance with a minimum voltage drop to initiate radiation. I don't know if there's a physical basis, but for the purposes of this - it's good enough (I think).

This was accurate to 0.99999 with the coefficients I got. Vo and Isat are the real coefficients, but I converted them into impedance Z and saturation power to make it convenient to put into the tables

Now, the lower the radiation impedance and the higher the initiation voltage, the better = more radiation power as a function of current. It'll be interesting to track these over time too.

 

[karimwassef]

Ok. Lots of stuff coming. I hope I can post it all this weekend.

Theatrus - I do need more in the supply though.

A normal LED power supply uses PWM modulation to target a set current (constant current) and the voltage is what it needs to be to drive the LED string.

To interrogate the LED, we need a different kind of behavior. The power supply needs to start up in voltage mode, increasing voltage from zero volts until the setpoint current is achieved (linear mode). This is really to capture the IV characteristic.

Then, at the setpoint current, the power supply transitions to PWM in constant current mode. Of course, if we get to the forward voltage before the setpoint current is achieved, then the power supply naturally converts from linear voltage increase and voltage mode into PWM mode and constant current.

It may be as simple as changing the voltage rise characteristic of the IC? Alternatively, a current limiter on the input with a large bank of capacitors could work too.

Ideas?

 

[theatrus]

Ok. Lots of stuff coming. I hope I can post it all this weekend.

Theatrus - I do need more in the supply though.

A normal LED power supply uses PWM modulation to target a set current (constant current) and the voltage is what it needs to be to drive the LED string.

To interrogate the LED, we need a different kind of behavior. The power supply needs to start up in voltage mode, increasing voltage from zero volts until the setpoint current is achieved (linear mode). This is really to capture the IV characteristic.

Then, at the setpoint current, the power supply transitions to PWM in constant current mode.

This is how pretty much all switch mode supplies work - in this case the ramp is controlled by the primary inductor. The current limit is controlled by the sense element.

Of course, if we get to the forward voltage before the setpoint current is achieved, then the power supply naturally converts from linear voltage increase and voltage mode into PWM mode and constant current.

This complicates things a lot... you would be running in an CV or CC mode with limits, which is usually done with a microcontroller subsuming all control functions.

It may be as simple as changing the voltage rise characteristic of the IC? Alternatively, a current limiter on the input with a large bank of capacitors could work too.

Ideas?

Yeah - many switchers have a soft-start capability, the budget LM3409 does not

 

[karimwassef]

So. How would you do it?

 

[theatrus]

I assume the Vf setting is dynamic? Where is it derived from?


Since most controllers have a current adjustment input, you can use the micro to adjust Vf down via the current limit. This could also be used for ramping, with everything in switch mode (but that's a very solved DSP problem, need some current ramp)