help with LED current mirror, pulsed current source

[grizedale]

I have been very keen for you to proove me wrong, and am still trying to help you to do so. -but its not really me, because these are simply the views of the vast majority of the LED lighting industry.

From your last post, i am unsure where we are at now, the original point was about uncompensated parallel operation in LEDs.

If we take, say for example's sake, the XLamp range of Cree:-
**broken link removed**

...which of the XLamp range, (or any other range of your choice) are OK for uncompensated parallel operation.?

Am i right that in some cases, you appear to be saying that you have a special formula which allows you to add very small resistance values to each series string of a parallel'd bank of LEDs, and that resistance value equalises the current very well?


You have made statements about statistical spread of Vf values within a batch, but cannot corroborate this from LED foundry official documentation.

You say that you are selling LEDs to people, and some of them are putting them in parallel, without compensation.............i am wondering if these people are paralleling them because they know no other way.

We all know that LEDs in parallel without equalisation is ok if you pay the LED foundry a fortune to make up paralleled LED modules like that.....but few companies can stay afloat and pay that kind of outlay.

 

[godfreyl]

I have absolutely no experience with LED lighting but it seems to me that SunnySkyguy knows what he is talking about (and doing).

OTOH, I'm absolutely fascinated by Grizedale's about-turn on the subject. Just a few weeks ago, he wrote this:

Hello,My company has been putting 1A power LEDs from Cree and Osram in parallel for soome time now.

We only ever use a single current source, and NEVER use series current equalisation resistors or current mirrors.

We have encountered no problems......

On our production line, each finished product gets left ON for 20 minutes before being bagged and despatched..........if a failure was going to occur, it would occur there...but it never has, so we are thinking of doing away with this mini soak test.

After a couple of people on an internet forum suggested that that may be a bad idea, he seems to have completely reversed his opinion and has started writing things like this:

I think LEDs in parallel is out of the question.

I dont understand anyone who wants to take the risk of paralleling LEDs.......

Parallel LEDs are for Engineers who do not know how to design step-up type SMPS LED drivers.

By the way, i worked for Tridonic in Dornbirn as well as several other LED light companies...putting LEDs in parallel (with no form of current equalisation) is a TABOO subject !!

Wow! Was that really all written by the same person working for the same company?

 

[grizedale]

I wont deny for one minute that i haven't looked deeply into parallel LEDs........and yes my company has dabbled.........we wanted to do it because it reduced our wiring costs, could be done with simpler converters, and less converters for big arrays.

However, we could not get reliable Vf statistical assurances, and so were forced to look away, also there was the fact that thermal coupling had to be very good and this tended to be more expensive.

We thought about designing conservatively, but its more expensive.

So We have come back to LEDs in series.

-But im not going to be blinkered....if new ways and means come up and theyre cheaper, we're all in to it.


If you think Sunnyskyguy knows what he is talking about, then please could you state which part you were impressed with?

Sunnyskyguy would surely admit himself, that his views dont concur with the majority view in the LED lighting world, and so i am hesitating about re-going over his very long replies in order to find the best and most pertinent bits...i'd like to , but i dont have time right now.

In what paragraph of which post did Sunnyskyguy convince you that standard LEDs could be paralleled without any current equalisation?

 

[godfreyl]

In what paragraph of which post did Sunnyskyguy convince you that standard LEDs could be paralleled without any current equalisation?

You said it yourself right here:

We all know that LEDs in parallel without equalisation is ok if you pay the LED foundry a fortune to make up paralleled LED modules like that.....

So you agree that it is possible to operate LEDs in parallel successfully. The only remaining questions are "Who is able to do that?" and "How much does it cost?".

SunnySkyGuy claims he is able to do it reliably and economically, and has done so. You claim you are not able to do so. I don't have a problem with either of those claims, I believe both of you.

 

[grizedale]

I have asked Sunnyskyguy for pricing (he is now working as a LED salesman) but he will not tell me..........
-but i can assure you that this "LED module service" comes at a heck of a price, and is not financially viable.

Sunnyskyguy tells me that he doesnt think that i know what i am doing, so he doesnt want me to put LEDs in parallel, and he only sells to those who know what they are doing.

 

[BradtheRad]

I've seen advice not to parallel diodes/LEDs.

The problem is unequal current sharing (despite having equal supply voltage), sometimes to the point where one component hogs current and burns up.

With led's they might exhibit unequal lumen output, despite having equal supply voltage.

Seems to me it's the rogue outlying led that we have to worry about.

1.

I have thought parallel led's could work if:

(a) if the deviations can be confined to a sufficiently narrow range (more or less eliminating the rogue outliers),
and
(b) a sufficient amount of led's are assembled per string, to be pretty sure the individual V/I characteristics will average out.

I think then we would have uniform mA through each string.

2.

Or, a way to parallel individual led's successfully would be to include some kind of regulation, or equalization, on each led. Perhaps this would be in the form of a biased mosfet or two on the wafer, as I understand is used to fabricate power resistors in IC's.

I wouldn't be surprised if some such workaround has been on led's which Sunnyskyguy purchases.

3.

Sunnyskyguy claims a reason for his success is that manufacturers have made improvements in V/I deviations, over a single production run. This has been over the past 5 years. Barely enough time to alter our mindsets on the subject of paralleling led's.

Since he can also tell the manufacturer he'll guarantee regular volume orders, he's granted privileged service by receiving a run of led's fabricated at the same time.

Or for all we know they were examined, and the outliers were eliminated by a process none of us is aware of. It might be cheaper than testing each and every led to make sure it conforms to a narrow V/I range.

 

[grizedale]

Brad The Rad

Sunnyskyguy claims a reason for his success is that manufacturers have made improvements in V/I deviations, over a single production run

......You cannot get any statistical info on Vf variations in a batch from any LED foundry.

Osram have official docs (quoted amongst these posts) that clearly state that LED Vf is loose because LEDs are diodes that are optimimized for light output.

If a salesman, tells you that you can have something, but only he is able to confirm the spec of what he's wanting to sell you, how much do you believe any salesman?

 

[D.A.(Tony)Stewart]

I never indicated anyone could simply connect any LED's in parallel ad it works and getting precision voltage matching happens by accident in that they come from the same order of wafers as each customer is build to order. No disti stock, no factory stock.

You need to know the critical thermal and electrical resistance parameters for the heat source and the interconnects and make it the voltage increase with current, rather than decrease. I used distributed wiring resistance to accomplish this. Others over compensate, some do it just right, while the safe approach is CC mode albeit it is relatively inefficient for low voltage strings such as 12V but then car electrical efficiency is not as much an issue.

 

[godfreyl]

......You cannot get any statistical info on Vf variations in a batch from any LED foundry.

There are a couple of issues here.

Firstly:
"I can't find XYZ information on the internet" does not mean "XYZ information is not available from the manufacturers". If you can't find the information you want online, you need to make inquiries i.e. pick up the phone or send an email.

Bear in mind that if you want e.g. detailed information about Cree products, it is far better to direct your questions to Cree, rather than random strangers on an internet forum.

Secondly:
Imagine you are interested in a scheme which requires a particular LED parameter to be matched to within 5%. However the manufacturer tells you that that parameter is not tightly controlled, and is only guaranteed to match within 20%.

At first you may think your scheme is not workable.

But what if you buy a batch of 1000, measure them, and find they all match to within 2%? You may start getting cautiously optimistic. If the next batch and the one after that measure similarly, you may decide to go ahead with your scheme.

In this case you are not relying on information given out by the manufacturers, but rather on your own research.

I suspect that SunnySkyGuy may have invested a fair amount of time and effort in research, testing and experimentation. The reward for that kind of work is useful information that cannot be found in the manufacturer's literature.

 

[D.A.(Tony)Stewart]

It seems TI has figured out what I have in choosing a working matrix that leads towards electro-thermal stability, which is what I have been doing for years.

 

[BradtheRad]

It seems TI has figured out what I have in choosing a working matrix that leads towards electro-thermal stability, which is what I have been doing for years.

It appears to be an unembellished boost converter. It does substantiate what you've been saying, that paralleling led's is workable.

Along the lines of post #49 (Godfrey1) could there be a possibility that your vendor has an automatic testing apparatus in the factory, which tests a reel of led's, one at at time, and marks the ones that meet your V/I spec?

To test one might take two seconds.

They might leave the machine running unattended all day while it tests several thousand, and then they simply bag the led's that you contracted for.

The remainder might go to a customer who has non-precise specifications.

It might be a custom-built machine. Hence other manufacturers don't have it. The salesman can offer components that meet your specs and allow you to do something few manage to do successfully. It's done without alien technology. Current wisdom is not contradicted.

 

[FvM]

It appears to be an unembellished boost converter.

Yes, but I don't expect that TI claims a particular matching performance for the cicruit. It will be most likely sufficient for general lighting/signalling applications, but won't hardly achieve a matching in a 3-5% range, as required by the original poster, even with a forward voltage bin sorting offered by some manufacturers.

That's also the general problem of this thread, which moved from a rather clearly specified current matching problem to a broad discussion of LED parallel connection.

In my view, it's rather obvious, that a tight lighting uniformity specification, e.g. for the said machine vison application demands for a per LED-string current stabilization. LED grouping or individual intensity adjustment may be still necessary to take account for prodcuction variations.

 

[grizedale]

The TPS61165 circuit above with 9 parallel strings of 3-in-series LEDs is regulating the overall current to 182mA.
Since there are 9 parallel strings, thats 20mA in each LED.

-At such low current, you could buy those LEDs that have some "extra" resistance in them and get the current equalisation like that.

Alternatively, since its only 20mA, you could very easily add physically small resistors in series with each string and equalise the current like that.

Certainly, for that very low current, it probably wouldnt be worth having 9 current sources.

I am sure that some of the small 50mA rated LEDs often come with a decent bit of resistance built into them.

With Power LEDs operating on the steep part of the I-V characteristic its a different story.

BradtheRad:

Along the lines of post #49 (Godfrey1) could there be a possibility that your vendor has an automatic testing apparatus in the factory, which tests a reel of led's, one at at time, and marks the ones that meet your V/I spec?

Also, even LEDs with great Vf matching, still need to be well thermally coupled if mounted in parallel.

That is why LED foundrys' parallel LED offerings comprise tested and matched LEDs *pre-mounted* on special heatsinking so that they are well thermally coupled.
-as i mentioned , this service is very very expensive.
.....

As far as Vf binning is concerned, in case anyone is interested, this isnt so tight that you can parallel the LEDs....eg the Cree XPEHEW range where the binning range spans 250mV.
-This is nowhere near tight enough to mean paralleling is OK.

Vf binning for XPEWEW LEDs:- (page 8)
**broken link removed**

 

[grizedale]

godfreyl

But what if you buy a batch of 1000, measure them, and find they all match to within 2%? You may start getting cautiously optimistic.

Sorry but i wouldnt be too pleased with 2%....because take the I-V curve for the XPEHEW LED on page 5 of the datasheet....

XPEHEW LED DATASHEET:
**broken link removed**

V(f) = 3V at 350mA...So 2% is 60mV.....and this difference in voltage will mean a difference of about 100mA.

...so one of them will have 100mA more current in it, -so it will run hotter.....then its Vf will get lower than the other one, ...then it will hog even more current than 100mA......then it will get hotter......then its Vf will get less, then it will hog more current...................and we have Thermal Runaway......from a starting point of 2% difference in Vf.

This is the problem......and if they are not really very well thermally coupled, then the whole situation gets worse still.

 

[godfreyl]

Sorry but i wouldnt be too pleased with 2%....

It was a hypothetical example. I did not suggest you would be happy with Vf matched to 2%.

I suggested that if you wanted some parameter matched to 5%, then you would be happy if that parameter was matched to 2%.

 

[grizedale]

OK my apologies , i understand.

Though this has brought up the subject that the matching required is very very tight

 

[D.A.(Tony)Stewart]

My 5m LEDs are often matched within 1mV @ 20mA by quality of wafers per batch.
so 1/3200mV ~0.03%. But I do not expect or rely on that for power LEDs.


You guys still don't get it. Let me make it perfectly clear.

Let me use the XP-E as your example.

Re-examine page 3 specs.

Characteristics . . . . . . . . . . . . . . . . . Unit. . . . . Min. . . . . . . . Typ. . . . . Max.
Thermal resistance, jcn to solder pt,Tjsp - white [°C/W] . . . . . . . . . . 6
Viewing angle (FWHm) - white . . . . . . . . . . . [degrees] . . . . . . . . 120
Temperature coefficient of voltage - white . . . [mV/°C] . . . . . . . . -3
ESD classification (HBm per mil-Std-883D) . . . . . . . . . . . . . . . Class 2
DC forward current - white . . . . . . . . . . . . . . . . [mA] . . . . . . . . . . . . . . . 1000
Reverse voltage . . . . . . . . . . . . . . . . . . . . . . . . . [V]. . . . . . . . . . . . . . . . . . 5
Forward voltage (@ 350 mA) - white . . . . . . . . . [V]. . . . . . . . . . 3.0 . . . . . . 3.5
Forward voltage (@ 700 mA) - white . . . . . . . . . [V]. . . . . . . . . . 3.15
Forward voltage (@ 1000 mA) - white . . . . . . . . [V]. . . . . . . . . . 3.25
LED junction temperature . . . . . . . . . . . . . . . . . [°C] . . . . . . . . . . . . . . . . 150

Note above tests taken @ 25 °C as usual, by process control, so essentially test results use Tspa=0, while Tjsp=6 'C/W and Tspa is thermal resistance thru heatsink to ambient.

What is equivalent temp rise to offset voltage rise from ESR? ( and thus have zero net voltage change )

1.) from 350 to 700 mA? = 50 °C
Vf= 3.15 – 3.00 = 150 mV (Voltage rise from ESR )
from spec -3 [mV/°C] , thus 150 / 3 = 50 °C

2.) from 700 to 1000 mA? = 33 °C
Vf = 3.25 -3.15 = 100 mV
thus 100 / 3 = 33.3 °C
3.) from 350 to 1000 mA? = 40 °C
Vf = 3.25 – 3.00 = 250 mV
thus 250 / 3 = 83 °C , which agrees with adding results from above.
4.) Now what is the min. Thermal resistance of heatsink needed to meet above, and make an assumption for driver & cable ESR ?
∆Tj = (∆P) * (Tjsp + Tspa)

This may help you see my solution.
Where is your project designer?

- - - Updated - - -

you current mirrors ought to be in same package, so they are matched perfectly for hfe or Gm, Vt

 

[grizedale]

Thanks Sunnyskyguy...going through it now.

 

[D.A.(Tony)Stewart]

I guarantee with my method you can achieve better consistent regulation than with current mirrors as they are worse for consistency than LEDs. with a MOSFET driver you also have the added advantage of a PTC and ROHM makes triple output switches for many Amps that are low saturation voltage that are inexpensive and will outperform your current mirrors unless they are integrated into one package as they are in IC's.

The idea behind optimizing the Tja (Thermal resistance) and Rs (ESR) is to parallel at least 4 and preferably more for improved efficiency. If making say a 300W or 1KW wall wash light or organic growing lamp, I might choose high voltage strings in parallel or drive them with independent TRI output ROHM MOSFET drivers
such as these 10A drivers **broken link removed** 3x output with common gate mounted on copper clad thermally conductive ceramic PWB or metal clad PWB. Rectifying 120Vac with full wave mosfet bridge or Schottky bridge will yield 2√2 120= 336Vdc max so with suitable arrangement of matrix of LEDS one could drive 1kW of LEDs easily with only rectifier and driver losses of 3~4V or < 1% of the 1KW load and regulate the intensity with current control of the buck boost regulator using PWM from 0 to 100%.

One could drive direct from Line voltage rectified using only a line choke but at 120Hz, this tends to be a large choke rather than switching at HF. I would conformally coat the LED's in optically pure epoxy for safety insulation voltage and make it moisture proof for outdoor use. TVS diode may be needed to protect the VDS max of 20V is in series with the LED strings ( series / parallel matrix) ON to off threshold of +/- 15% or 3.2+/- 0.6V for 0 to 100% typ. So this also protects the MOSFET drivers in the off state. The line regulator can achieve PFC corrected and 95% efficient if done right with PWM intensity control.
No one is offering this configuration yet AFAIK.

Tony Stewart
EE since 1975.

 

[D.A.(Tony)Stewart]

BTW my avatar today shows my living room with white LEDS with that dim to only yellow LEDs then off using a surplus 65W Laptop supply with external pot to control the voltage on charger. It has 5 wire connector that terminates with remote sensing at the coaxial connector. Cost me a few bucks for wiring regulator and the LEDs were almost free. My wife wanted more sunlight in the bay window at max brightness it is about 3x as bright as direct sunlight with very little glare.