This is a continuation of part 1 which can be found here:
http://www.fieldlines.com/story/2007/4/20/95351/1614
Let's look at some Vf figures for some real leds, and explore the consequences.
I bought a bag of 100 5mm White SuperBright leds off an australian seller on ebay.com.au.
The published specs are;
20,000 mcd at 20mA (which is also the max continuous forward current)
Vf = 3.0 typical
viewing angle 15 degrees
max power dissipation 100mW
I took a random sample of 20 units, and tested them for Vf at 2 different currents, 11.00mA and 18.00mA, at a room temperature of about 20 degrees C or just slightly above.
Without going into all the boring details, the summary is;
at 11.00mA
average Vf = 2.961
standard deviation(sd) = 0.049
min = 2.845
max = 3.024
at 18.00mA
average Vf = 3.068
standard deviation(sd) = 0.055
min = 2.939
max = 3.132
Average and sd calculated by a spreadsheet.
If you want to understand the significance of sd further, remember "Google is your friend".
This is actually really difficult to do, because as soon as you apply current, the leds start to warm up, and Vf starts to decrease (at 2mV per degC as I explained last week). So what I did was wait about a minute for each one to let the Vf stabilise at some voltage.
Now, whats the significance of all this?
Consider last weeks circuits, in particular 3 & 4, where we have multiple parallel strings powered by one constant current source.
Let's say in the first string of 3 leds, we just happen to randomly pick 3 leds that have a very low Vf, and conversely for the second string, pick 3 leds which have a very high Vf. If we set the value of Req to zero, the string with 3 very low Vf's are going to hog all the current. So, to make the 2(or more) strings of leds play nicely together, we need to set a value of Req greater than zero, but by how much?
At this point, I set up an experiment and did some testing. I picked 3 leds which had a Vf of Avg - 1sd (< 2.912 at 11.00mA) and 3 leds with a Vf of Avg + 1sd (> 3.010 at 11.00mA).
Req = zero, currents were 6.91 mA and 3.28 mA
Req = 22 ohms 6.66 and 3.68
Req = 47 ohms 6.46 and 3.98
Req = 100 ohms 6.20 and 4.37
I know the 2 currents in each case don't exactly add to 11.00 mA, this is due to the internal resistance of the meter and moving it from one string to the other. No prizes for guessing what would happen if we had assumed perfect leds, and set it up to run 20mA in each string of leds; after all, the spec is so many milliCandela at 20mA.
You can see that in the worst case scenario, it takes a large value of Req to swamp the differences in Vf of the 2 strings.
Knowing this, it should be obvious that a little effort in matching the 2 strings is probably worth it.
Now that I have these 2 strings, I want to go back to square one, and show what happens when you use the simple, single resistor per string, just for comparison. Using a certain online calculator, I input my specs. 12 volt supply (because it is a 12 volt battery, isn't it?), Vf = 3.0 volts, I = 20mA. It gave me a value of 150 ohms for the resistor.
at 11.0 volts, currents were 14.06 and 11.86
at 12.0 volts, 19.09 and 16.62
at 13.0 volts, 24.07 and 21.40
at 14.0 volts, 29.56 and 26.69
at 15.0 volts, 34.7 and 32.3
You can see that with a real lead-acid battery, during charge and discharge cycles, this simple approach leaves us pushing way too much current through the leds most of the time. This will result in very short lifetimes for the leds.
So, lets use the maximum battery voltage of 15 volts, and select a resistor based on that. The previously mentioned online calculator gives us a value of 330 ohms. Re-doing the previus test gives;
at 11.0 volts, currents were 7.38 and 6.45
at 12.0 volts, 9.99 and 8.94
at 13.0 volts, 12.63 and 11.50
at 14.0 volts, 15.31 and 14.12
at 15.0 volts, 17.96 and 16.71
Now we keep the current below the leds rated maximum continuous current; but the light output varies very noticeably as a function of the battery voltage.
This simple single resistor approach can be made better, as far as the led is concerned, by using less leds in the string and increasing the value of the resistor. Taking it to the extreme, we end up with one led per resistor, which is not very efficient. I suppose I could have run a test with 2 leds per string and one led per string, for the sake of completeness. But, even though the variability of current vs battery voltage would be decreased, it still would not be completely eliminated.
Summary of parts one and two.
The stated forward voltage of an led is not a cast in stone constant. It will vary from led to led, even from the same batch. It will vary with the amount of current you pass through it. It will vary as an inverse function of temperature; approx 2mV per degree C. Remember that when you plan on stuffing a whole heap of leds in a tiny, un-vented space.
The light output is a function of current.
Take careful heed of the rated maximum continuous current. Exceed this and your leds will die an early death.
There is no perfect led driver circuit. Cheap, efficient, simple. Pick one.
The true art of design is balancing all the compromises to achieve the best result in a particular situation. If you have grasped the concepts presented, you will be in a better situation to maximise the compromises for your next led lighting project.
In part 3, I will build a prototype of circuit 4, show how we work out a few values, and at least one method of matching the leds for the parallel strings.
Amanda