Should Long Series-Parallel Strings of LEDs have Current Equalization?

Thread Starter

Infinion

Joined Apr 2, 2016
9
Hey all,

My question is focused on the design of long series-parallel strings of LEDs driven by a constant current, and whether active current equalization with current sources like the NSVC2050 or current mirrors are necessary when all LEDs are thermally coupled, and the strings are very long such that the statistical variation of forward voltage between each string of LEDs is about the same.

I'm developing a round light engine PCB using 4 parallel strings of 28 LEDs as seen below for a total of 112 LEDs

Series-Parallel Current Matched.png
figure 1: Four series-parallel strings of LEDs, each drawing 60mA of current which are each regulated by the AL3066



My first design had each of the parallel strings of LEDs being current-matched and regulated through an IC called the AL3066 https://www.diodes.com/assets/Datasheets/AL3066.pdf, an amazing PWM controllable constant current step-up LED driver. An outright beast of its kind.
Al3066.PNG
figure 2: Application circuit for the AL3066 to show the full schematic

In my first version, I ordered a 2-sided FR4 PCB (about 5.5" in diameter) with LEDs on one side and the LED Driver circuit on the other, and hit temperatures around 70-80°C with LEDs facing downwards (measured using a non-contact thermometer 6" from the surface) and at full brightness (about 22 watts and nearly 3000 lumens). In either case I decided to run the lights at a lower duty cycle/power so the light engine hit a more reasonable temperature of around 65°C

In any case I decided to work on a V2 to improve the thermals. The LED Driver is now on a separate PCB and wires will be connected to the light engine, which will be a single-sided MCPCB instead of an FR4 PCB (aluminum instead of fiberglass). Rather than running 5 wires between the boards (Vout and 4 channels), I want to simplify the connections so that only 2 wires are required. This means I have to tie the 4 channels together and can't use the AL3066's current matching feature to balance current....But is it still necessary?
Series-Parallel.png
figure 3: Four series-parallel strings of LEDs, driven by a constant current of 240mA

I'm trying to determine if I need to design a current mirror or use CCRs on the light engine to balance currents. The LEDs are thermally coupled on the FR4 board from my V1 and will be even more coupled with the very thermally conductive MCPCB, and like mentioned above, 28 LEDs brings the Vf variation between LEDs to a fairly consistent level across the strings.

An example of a product that uses the same approach would be Forge's light engines https://www.forge.co.uk/products/light-engines/200mm-round-light-engine-select where in their datasheet (page 4) they use fewer series-parallel LEDs but probably much stricter binning.

I've been trying to think of the best way to do this for a while now and I've also tested running a smaller set of 21 LEDs/string in series-parallel without equalization, resulting in no thermal runaway, no nonuniform lighting, or any LED failures or color shifts. So I feel fairly confident that I have a good design here but could use a sanity check from someone.

All that being said, I appreciate all for reading this post and giving it some thought in an effort to help and/or learn.
 

oz93666

Joined Sep 7, 2010
739
With the design you've specified you definitely won't need current equalization ... with 28 in each line ... all leds from the same bin , and selected randomly when making up the lines , the laws of statistics say each line will draw the same current (within about 1%)...

Always good to under-drive leds for a long life , below about 80% max recommended current , even less if the heat sink is not the best possible.
 

ArakelTheDragon

Joined Nov 18, 2016
1,362
Yes! It will need current limitation! Despite that in general the 1% accuracy is right for the "equalization", I see no "limitation" what so ever, as a whole the LEDs will drain too much current and burn. It is not recommended to use a LED without a current limiting resistor.
 

WBahn

Joined Mar 31, 2012
29,976
The matching really doesn't matter a whole lot with regards to whether you need to some kind of channel equalization since thermal runaway doesn't give much of a hoot about how well matched the LEDs are. What will determine whether you need to equalize the channels is how effective your thermal coupling is.
 

dendad

Joined Feb 20, 2016
4,451
You may get away with it but it is better parctice to have a current limiting resistor in each leg. Some cheap products come out with LEDs in paralell but I would advise against it.
 

Thread Starter

Infinion

Joined Apr 2, 2016
9
Thanks, everyone, for the great responses! There's some excellent insight to digest here.


With the design you've specified you definitely won't need current equalization ... with 28 in each line ... all leds from the same bin , and selected randomly when making up the lines , the laws of statistics say each line will draw the same current (within about 1%)...

Always good to under-drive leds for a long life , below about 80% max recommended current , even less if the heat sink is not the best possible.
This is good to hear! Yes, in this case, my LEDs are binned. It is as simple as asking Digikey sales support for binned LEDs from a specific manufacturer's part number. In the case of series strings of LEDs that are not binned, it is likely the shape of the statistical bell curve / normal distribution is wider, leading to a larger standard deviation, and needs more LEDs to bring the variance closer together. I'm curious if a mathematical expression has been used that defined the standard deviation between parallel LED strings with respect to the number of LEDs in those strings so a designer could determine how many LEDs to put in series to achieve a target variance in string voltage. Depending on the result, they could use that information to select the best control scheme that minimizes system losses, or cost, or maximizes the luminous efficacy of their LEDs. A voltage regulated string with parallel resistors, a current regulated string with nothing, or CV or CC regulated string with constant current regulators in series with each string.

Actually, in writing this, I found a curious paper discussing this probabilistic approach here. Fantastic! http://www.ece.neu.edu/groups/power/lehman/Publications/Pub2007/2007_6_Bhattacharya.pdf


Yes! It will need current limitation! Despite that in general the 1% accuracy is right for the "equalization", I see no "limitation" what so ever, as a whole the LEDs will drain too much current and burn. It is not recommended to use a LED without a current limiting resistor.
Normally I would agree with you if a string became disconnected due to the eventual mechanical fatigue from thermal cycling, or if a constant voltage were instead applied to the series-parallel strings for some desired power output, and the strings were physically mounted to different surfaces with no thermal coupling. However, in my case, this circuit is driven by a constant current of 240mA as mentioned in the body of my original post's opening paragraph and suggested in figure 3. The Al3066 is the DC-DC converter that regulates this constant current. So the LEDs cannot burn out as long as all strings have continuity.

The issue I had with using current limiting resistors was the fact that a current-limiting resistor would do little to protect the LEDs in the failure case of a constant current driven setup. They may even fail before the LEDs do if their power rating was exceeded after a failure because typically they are sized and rated for the worst-case current of a string.

That's actually an interesting thought though, in a constant current setup, resistors are no more than power-dissipating slow fuses, but if you anticipated an entire string failing, or being hot swapped out of some product while the circuit was running, you could place resettable PTCs in series with the strings to trigger the OVP of the regulator. That being said, I would much prefer to throw in a constant current regulating diode instead of a PTC, like the NSIC2050.

I suppose rather than a current-limiting resistor, I could employ balancing resistors for passive current equalization, something like 0.5 - 10 ohm resistors depending on how badly the strings were balanced, but something with a very low resistance value.


The matching really doesn't matter a whole lot with regards to whether you need to some kind of channel equalization since thermal runaway doesn't give much of a hoot about how well matched the LEDs are. What will determine whether you need to equalize the channels is how effective your thermal coupling is.
This is a good point. Well, in my first version I had the LEDs spaced radially on a 1 oz copper FR4 PCB. These are all on the same board with 16 mil traces and a full copper pour around them to distribute heat. I imagine the LEDs on the outside circumference would be slightly cooler than the LEDs placed concentrically inside them, as there are a lack of heat sources fully surrounding them. However, each of the four series channels exist in a 1/4 quadrant of the circle, so each channel would have equal numbers of LEDs along their portions of the outer circumference. Here's my FR4 version for anyone's reference.

IMG_20180418_193104smaller.jpg

In the version I'm working on, I'll be using a metal-core PCB (aluminum with copper traces on top) instead of a typical fiberglass one, so the thermal coupling should be on the level of COBs, a bit worse due to the discrete LED packages having more things in-between them and the board.

You may get away with it but it is better parctice to have a current limiting resistor in each leg. Some cheap products come out with LEDs in paralell but I would advise against it.
ArakelTheDragon said something similar regarding current limiting resistors, but were you aware that my circuit was driven by a constant current? Also yeah for sure, there are a lot of cheap products that come up with just absolutely terrible designs that are, from a reliability standpoint, made to fail very prematurely because of either lack of due diligence or a strong focus on minimum viable product in favor of getting to market as quickly as possible.

It's funny, but I actually feel the contrary when I see products with parallel resistors -- that the products with passive current limiting are the "cheap" solutions. LED strips are notorious for this. As is usually the case in 12V strips, for every three series LEDs, there is one current-limit resistor, and just like that, 25% of the supplied power is wasted as heat ... and yet they are purchased and used widely everywhere. Think about that for a second, for every 100 Watts of power supplied to LED strips everywhere, 25 watts of it is dissipated just to limit current and balance the LEDs for convenience.

That's really what pushes me to embrace active constant current regulation with the use of DC-DC converters to power larger strings of LEDs. If current limiting resistors are used, it just negates the benefits of the great luminous efficacy of LEDs, and drags the efficiency of the system through the mud.
 

dendad

Joined Feb 20, 2016
4,451
ArakelTheDragon said something similar regarding current limiting resistors, but were you aware that my circuit was driven by a constant current?
Constant current drive is the best. But have one per string.
Just s'pose, in your circuit with the 4 strings, you have 240mA constant current, = 60mA per string. What if there is a crook joint in one string?
Now you still have 240mA constant current, but in only 3 strings, so it is 80mA per string.
It can go down hill from there. But even if you run a number of strings on 1 constant current supply, you do need resistors to help match the current in each string else a small imbalance can cascade.
That is one reason why a constant voltage and current limiting resistor for each string works pretty well.
But I do agree, constant current is best. One per string!
 

ArakelTheDragon

Joined Nov 18, 2016
1,362
Constant current drive is the best. But have one per string.
Just s'pose, in your circuit with the 4 strings, you have 240mA constant current, = 60mA per string. What if there is a crook joint in one string?
Now you still have 240mA constant current, but in only 3 strings, so it is 80mA per string.
It can go down hill from there. But even if you run a number of strings on 1 constant current supply, you do need resistors to help match the current in each string else a small imbalance can cascade.
That is one reason why a constant voltage and current limiting resistor for each string works pretty well.
But I do agree, constant current is best. One per string!
I agree. I have used LED strips combined by 3 and I had the problem or them being connected directly to the regulator. 1 current limiting resistor should work(something like 1 Ohm) but I do not like that design.
 

Thread Starter

Infinion

Joined Apr 2, 2016
9
Constant current drive is the best. But have one per string.
Just s'pose, in your circuit with the 4 strings, you have 240mA constant current, = 60mA per string. What if there is a crook joint in one string?
Now you still have 240mA constant current, but in only 3 strings, so it is 80mA per string.
It can go down hill from there. But even if you run a number of strings on 1 constant current supply, you do need resistors to help match the current in each string else a small imbalance can cascade.
That is one reason why a constant voltage and current limiting resistor for each string works pretty well.
But I do agree, constant current is best. One per string!
Absolutely, in the situation where an entire string fails, the other LEDs will run above their recommended rating and operate hotter. But if the bad joint (severed trace or loose bonding wire) were the culprit, I would argue a resistor wouldn't have had a role to play, it was the joint that caused the failure. If the situation played itself out and we had 240mA through 3 strings, the resistor added for the other strings would be dissipating I^2*R power, so it would dissipate double the power at 80mA that it did at 60mA. I agree that the resistors would at least reduce the difference in current between the other 2 strings, but even without them, the situation won't really change because there is still thermal coupling that shares temperature with the other LEDs so they vary together against a cascade. Ironically, the resistors would introduce a secondary failure mode, where every series string failure increases the resistor's dissipated power exponentially, until we are operating beyond its recommended limit, and it fails open before the LEDs do, leading to a failure cascade of all resistors.

So in a constant current setup where the output voltage can vary to maintain a constant current, the V=IR relationship that resistors use to limit current is no longer useful because the output voltage simply sits at a higher voltage to overcome it.

But yeah, I agree if I was dealing with a CV supply, a resistor would be perfectly appropriate part for current limiting and balance. In this case, we would be designing around expected failure rather than reducing the likelihood of it, but a combination of constant current regulating diodes (instead of resistors) in each string like you say seems to be the most robust solution. They have parts like the NSIC2050 where I hear they act like the analog to zener diodes, where they maintain a constant current across their pins instead of a constant voltage https://en.wikipedia.org/wiki/Constant-current_diode . If their internal structure consists of a JFET with its gate shorted to source, then it acts like an active variable resistor and raises its resistance in response to an increasing voltage. So a combination of these with the voltage divider to the OVP of the AL3066 should safely disable the system in the case of an anomalous failure. A linear regulator and sense resistor combo might work similarly too.


Mmm maybe for that particular manufacturer, but most will have bin codes for luminous intensity as well as forward voltage or even dominant wavelength as mentioned in this blog here http://lednique.com/current-voltage-relationships/variations-vf-binning/

With the LEDs I use, the manufacturer had forward voltage bins that I asked the distributor for and they gave them to me as a value-added service (cost of binned LEDs = cost of unbinned LEDs)

upload_2018-10-21_15-46-18.png
 
Top