help with LED current mirror, pulsed current source

OK thanks, i will look into it.

Regarding ESR , i am not familiar with the term for LEDs.......for LEDs i am used to looking into "dynamic resistance"

I dont see how you can get ESR because it would change depending where you are on the I-V curve.

I dont have access to enough LEDs to be able to do representative testing on what you mention.
-The LED companies are the only ones who can do this, and they tell people not to parallel them.

I dont understand anyone who wants to take the risk of paralleling LEDs.......you have to design conservatively which is much more expensive.

If you parallel LEDs, you can often use a buck converter to drive them, but if you use series chains , you can use a buckboost or boost or sepic (or flyback etc etc)

.....these are all just as simple as bucks, so i cant see why any desire to parallel LEDs.

If you know what you are doing, you can make it work by adding smal resistor or utilizing your distribution line resistance for same to reduce effects of junction NTC with ESR matching resistors. I do this all the time using a cheap $35 Laptop charger wit 19.5V to drive long strings of 6 x 1W LEDs in parallel... The operative thing is knowing what you are doing. Most people don't thats why Mfg do not recommend it. But Basically it is the ESR differences that make the voltage difference at rated loads. The ESR is MOST accurate at the center point of operation on the load curve., but is sufficiently accurate for to predict using asymptote plots the way I do. So I can evaluate my luminaries of 6 power LEDs to see if they are sufficiently matched , compensation for ESR means adding loss. Using a current source is easiest and also most inefficient with larger headroom.

Just ask me how they parallel 64 MOSFETs in the Tesla Roadster to share 850 AMPs. They also know what they are doing.

Sunnyskyguy:-

The ESR is MOST accurate at the center point of operation on the load curve

Click to expand...

..but the load curves can vary from LED to LED.

Sunnyskyguy you write so enthusiastically and so ferverently and interestingly that it is intruiging to keep reading you.

-but your words are only uttered by yourself, -nobody else makes such statements, nobody from a LED foundry makes such statements.

With greatest of respect to you you have declared that you are a specialist distributor of LEDs.
...so if people can parallel LEDs then they'll buy more of them because they can just keep stuffing extra parallel lines of LEDs into their lamps.

You point so readily to the pot of gold at the end of the rainbow (i.e. parallel LEDs).....we all want parallelable LEDs at cheap price, ..because it allows us to just stuff the schematic with our LEDs without having to add more drivers............but who is corroborating your evidence that LEDs can be paralleled.

So far, every bit of evidence which you have presented, might just as well be fake, because it all depends on whichever parameter you refer to being very similar between LEDs..........and how can an individual such as yourself possibly have tested enough LEDs from a foundry to be able to know the statistical distributions of the parameters.

My guess is that most people who are paralleling LEDs are doing it, then selling the product to people such as police departments etc etc (eg police car flashing blue lights)......i.e. selling to people who have no care or concern if the product lasts a long time, BECAUSE THEY THEMSELVES ARE NOT PAYING FOR IT!

...so if they fail regularly they just call the maintenance team, who replace the flashing lights, with more of the same, and so the cycle continues.

If you see page one of this Osram document called "LED diagnosis in automotive applications", you will see the following......

"Because a LED is a device optimized for
light output the electrical parameters have a
wider variation than for other diodes with
pure electrical function. Some properties of
LEDs have to be considered in detail
especially for the diagnostic function.

A LED is mainly electrically characterized by
its forward current (IF), reverse current (IR),
the forward voltage (VF), reverse breakdown
voltage (VR) and the junction temperature
(Tj). The characteristic is similar to a silicon
diode, but has several differences:
- a higher VF
- a higher VF production distribution
- a higher negative VF coefficient
- a higher leakage current

https://www.google.co.uk/url?sa=t&r...nqSFBA&usg=AFQjCNE9sFwwJ2AUj3J8fxK7XVESLvlS1g



Did you see the bit where it said that LED Vf has a worse production distribution than other types of diodes?...Did you also see the bit where it said that all the electrical parameters of LEDs have worse tolerance than other types of diodes becuase LEDs are optimized for light output?

...this is from the horses mouth.....the LED foundry......so why should anyone believe anything else.?



I think another point here is that you are in USA/Canada..................so you have a captive market......USA/Canada customers will patriotially buy USA -even if the product fails regular.............European or Far eastern customers will not buy local, but will by from anywhere if its cheaper etc.....so i am saying that a USA company that sells products with parallel LEDs that fails at a high rate, will not go out of business because fellow Americans will still buy from them and stay with them.

Sunnyskyguy:

Just ask me how they parallel 64 MOSFETs in the Tesla Roadster to share 850 AMPs. They also know what they are doing.

Click to expand...

Sorry but that is irrelevant.

Paralleling MOSFETs is nothing whatsoever like paralleling LEDs....you dont get thermal runaway by paralleling mosfets.
I've done SMPS's before and paralleled FETs...thats not a problem

I beg to differ on many of your experiences grizedale.
in reverse order...
You need to get more experience and get familiar with what I am trying to tell you. Everything I have said is relevant.

1) ESR and temperature coefficient of junction diode and the tempcoeff. of the bulk resistance (ESR) are key variables for ALL power semiconductors.
2) Some MOSFETS have significantly lower conduction losses yet classic NTC like power LED's
3) The power MOSFETS which say "easy paralleling" have a PTC bulk resistance to compenaste for NTC of the junction and hence more lossy but temperature stable for paralleling devices.
4) Using lower loss bulk resistance (again read ESR) requires using devices from the same batch of wafers and verified by ATE with sorting. This method is used for ALL LEDs and is done for some customers of MOSFETS that demand the ultimate performance ( like TESLA).
5) They do not spec ESR on LED's for a reason, which I won't get into, but like MOSFETS you can compare them for performance which are spec'd for ESR or ON resistance.
6)ANother key parameter forLED's is efficacy vs current. Lab LED's with 200Lm/W drop to 100Lm/W at max current with some customer's cooling methods. 140Lm/W at rated current are available now but after AC drivers are typically the same as Fluorescent tubes ~100Lm/W or worse 60Lm/W ( not bad considering it was half of that a few years ago .. again all due to lowering ESR and self heating. It will get better as ESR drops !!!
7) Modern LED's now have < 1milli-ohm ESR.@ 350mA which remains fairly constant to 1A and extended to 3A with some variances due self heating and junction to heatsink to ambient thermal resistance and junction NTC effects.
8) Your comments for regional bias on purchasing may be valid but not relevant to me. My main clients are in USA & New Zealand, and I only supply the best diodes tolerances, factory built to my specifications in order to meet customer requirements. e.g. 4000~4500 'K, Vf tolerance 0.1, high CRI, custom beam angle. I got the biz in the first place because I could get things done from spec to delivery of build to order parts in < 2wks if wafer stock on hand, otherwise 3 wks. in Quantities > 10K. One key to regional bias is the decision to avoid LED's from companies that violate Nichia's patents for example.
9) Don't misunderstand everything you read, the assumptions are vague about Vf. I can validate my statements. Vf can vary significantly from vendor to vendor wafers. But as you control your process and source of wafers, the distribution gets smaller. Again on a single batch I can get small power devices (75mW indicators on 5mm) to be within a few millivolts in a bag of 500 parts rated at 20mA. rated at say 3.0 to 3.2V That is just the tight process controls on the wafer, not by design of a certain ESR and hence Vf. Indeed you can get a distribution from many wafers and on high ESR ( read small chip) LEDs 3.1V +/-0.1 bins, 3.3V +/-0.1 and higher for 350mA LED's going to 3.6, 3.8V.
I can specify any bin I choose or expect a few % outside that range. It all depends what you want. The key here is I know what I am getting before it ships. If you don't care and use lossy current sources in a string, no matter. If you want higher efficiency in say a wireless power luminaire, then it matters. ( don't assume all LEDs have poor Vf distribution. So if you blindly choose any white LED from any supplier, you cannot predict it's parameters without reading the generalized specs. That does not contradict anything I have said. It is by design and spec, thus you can get tighter parameters albeit for a difference in cost, which I do. So it depends on your requirements, be it cheap and dirty, or best of the best or average but low standard deviation.
10) "optimized for light output" is true but also takes into consideration how effective a client's heat management is. ie. max Tj Since all Semiconductors are tested at 25'C ( and 85'C) using nitrogen, H2O @25'C or pulsed so narrow that there is no heat rise, the reader must understand Vf is also a function of your heat sink. THis is such a huge variable! With learning curve on power LED production, you will see significant improvements in time , that I have seen already.
10) higher leakage is normal for higher junctions. But the ratio of reverse effective resistance (leakage) to forward conductance resistance (ESR) will not degrade. In fact it is improving significantly. The writer of that article had a different perspective. He was comparing white LEDs from company orange to company apple, because each buys wafers from different sources and not always from the same source. Even CREE, Philips who now have trade agreements and many others.

I beg to differ on many of your experiences grizedale.
in reverse order... but I need to backup my statements (again) it seems for you. I hope this is constructive.

You need to get more experience and get familiar with what I am trying to tell you. Everything I have said is relevant.

1) ESR and temperature coefficient of junction diode and the tempcoeff. of the bulk resistance (ESR) are key variables for ALL power semiconductors.
2) Some MOSFETS have significantly lower conduction losses yet classic NTC like power LED's
3) The power MOSFETS which say "easy paralleling" have a PTC bulk resistance to compenaste for NTC of the junction and hence more lossy but temperature stable for paralleling devices.
4) Using lower loss bulk resistance (again read ESR) requires using devices from the same batch of wafers and verified by ATE with sorting. This method is used for ALL LEDs and is done for some customers of MOSFETS that demand the ultimate performance ( like TESLA).
5) They do no spec ESR on LED's for a reason, which I won't get into, but like MOSFETS you can compare them for performance which are spec'd for ESR or ON resistance.
6)ANother key parameter forLED's is efficacy vs current. Lab LED's with 200Lm/W drop to 100Lm/W at max current with liquid cooling. 140Lm/W at rated current is common now for clients who can afford them. ( not bad considering it was half of that a few years ago .. again all due to lowering ESR and self heating.
7) Modern LED's now have < 1milli-ohm ESR.@ 350mA which remains fairly constant to 1A and extended to 3A with some variances due self heating and junction to heatsink to ambient thermal resistance and junction NTC effects.
8) Your comments for regional bias on purchasing may be valid but not relevant to me. My main clients are in USA & New Zealand, and I only supply the best diodes tolerances, factory built to my specifications in order to meet customer requirements. e.g. 4000~4500 'K, Vf tolerance 0.1, high CRI, custom beam angle. I got the biz in the first place because I could get things done from spec to delivery of build to order parts in < 2wks if wafer stock on hand, otherwise 3 wks. in Quantities > 10K.
9) Don't misunderstand everything you read, the assumptions are vague about Vf. I can validate my statements. Vf can vary significantly from vendor to vendor wafers. But as you control your process and source of wafers, the distribution gets smaller. Again on a single batch I can get small power devices (75mW indicators on 5mm) to be within a few millivolts in a bag of 500 parts. rated at say 3.0 to 3.2V That is just the tight process controls on the wafer, not by design of a certain ESR and hence Vf. Indeed you can get a distribution from many wafers and on high ESR ( read small chip) LEDs 3.1V +/-0.1 bins, 3.3V +/-0.1 and I can specify any bin I choose or expect a few % outside that range. It all depends what you want. If you don't care and use lossy current sources in a string, no matter. If you want higher efficiency in say a wireless power luminaire, then it matters. ( don't assume all LEDs have poor Vf distribution. So if you blindly choose any white LED from any supplier, you cannot predict it's parameters without reading the generalized specs. That does not contradict anything I have said. It is by design and spec you can get tighter parameters albeit for a difference in cost. So it depends on your requirements, be it cheap and dirty, or best of the best or average but low standard deviation.
10) optimized for light output also takes into consideration who effective a client's heat management is. Since all Semiconductors are testing on a nitrogen or water cooled bed @25'C or pulsed so narrow that there is no heat rise, the reader must understand Vf is also a function of your heat sink. THis is such a huge variable together with learning curve on power LED production, that you will see significant improvements in time , that I have seen already.
10) higher leakage is normal for higher junctions. But the ratio of reverse effective resistance (leakage) to forward conductance resistance (ESR) will not degrade. In fact it is improving significantly. The writer of that article had a different perspective. He was comparing white LEDs from company orange to company apple, because each buys wafers from different sources and not always from the same source.

Sunnyskyguy,

Suppose for one minute that it is possible to buy a large batch of LEDs which have Vf's matching within 0.1%.

-You're still not in the clear because if these LEDs are not well thermally coupled in the product, then you just go straight back to thermal runaway.


Also, you have declared you are now retired EE and are selling LEDs as a job.
-so you should have a good idea about prices....

what is the typical price difference between a batch of LEDs with Vf's matching to 0.1%.....compared to the price of the same number of LEDs, but with far more variant Vf's?

(take whichever LED you like as an example, but what about Cree XPEWHT?.....

1....price of a batch of Cree XPEWHT with Vf's matching to 0.1%
2...price of a batch of Cree XPEWHT with Vf's variable over the range (eg varying over 3.1V to 3.9V)


Also, are you saying that you know of MOSFETs whose drain-source voltage *decreases* as they conduct more current?
i.e., a Mosfet whose drain-source voltage *decreases* as its temperature goes up.?

..please can you provide a link to the datasheet, so it can look at its price and marvel at it......i have looked at hundreds of fet datasheets of fets, but never saw one like that.
...i'm not saying such a thing doesnt exist, but are we talking galactical prices here?


Please try and get the prices, i'm serious when i say that if you have good pricing for this then i may know people who may be interested.
-And even better if you can provide statistical data from the LED foundry themselves.

I only deal with serious buyers who know what they want and you request lacks the details of a dozen other parameters I would need to know so that apples are compared to apples. I only deal with clients with MO$ of $10K and each supplier does not do custom orders on the spot, you have to go thru normal disti. channels and they wont do this. You need to have an existing demand with steady demand to negotiate best pricing. Cost depends on how much demand there is for yield loss. If the loss is normal product for other customs then consider 10% normal. If the product that falls out of range say for Iv +/-10% in a colour that is non standard where distribution might be +/50%, then price is a function of yield. For high end products such as autobahn tunnel sidewalks and wireless road studs, a premium on LEDs is smaller compared to packaging costs when compared to say PAR lamps which are cheaply produced compared to luminaires made to be run over by trucks. There is no secret formulae but yield and demand, and customer relationship are key factors on pricing.

- - - Updated - - -

Here is the NTC MOSFET IXG(K,X)400N30A3 but since it has a NTC vs PTC , not suitable to gang without ESR equalization.

**broken link removed**
If you want to pulse a 400A load , you may need to pulse the gate with its max currrent or source impedance that can deliver a few amps on gate charge.
Notice the Tempco is near zero at 200A and negative for < 200A and positive for > 200A.


Naturally all LEDs on a common heatsink need to factor thermal resistance and variation of same which affects coupling of adjacent devices for Shockley voltage characteristics.

Sunnyskyguy:-

I only deal with serious buyers who know what they want and you request lacks the details of a dozen other parameters I would need to know

Click to expand...

I know that LEDs can suffer thermal runaway in parallel and that paralleling them is definetely not recomended by LED foundrys and by most people in the know about it.

Can you supply names of companies who are putting your LEDs in parallel in their product?.....after all, if it really is an OK thing to do, they shouldnt mind opening up about it.

As far as i can see, there's no gaurantees given by the LED foundrys concerning Vf tolerance distribution..none whatsoever.

If i have engineers who can design buck-boost SMPS, boost SMPS and SEPIC SMPS plus flyback etc etc..(i.e. smps's which will facilitate longer series strings from a low input voltage) ...........then why would i want to put leds in parallel and risk thermal runaway?

if i can put leds in parallel and reduce the chances of thermal runaway by "conservative" design.......then why would i want to do that if conservative design is more expensive?

- - - Updated - - -

I do not disclose my clients, but they must pass VDE tests and are installed in Swiss tunnels and where they expect long life in their products. Initially there were some field failures 6 yrs ago, due to handing from ESD solder stress and thermal ESD from molten encapsulation but when I added zener protection and a dozen other process improvements to their design, the reliability issues were resolved.

Let me correct your understanding of "..but the load curves can vary from LED to LED."

The load curve is the sum of the Shockley ideal voltage curve vs T and the ESR .period. So ESR is the key variable.

The fact is the load line is curved due to the Shockley effect and is straight or asymptotic due to bulk junction resistor, we call ESR. It is one of my key metric to comparing one LED vendor with another because it is an indicator or Vf , max current and self heating issues.

In addition to efficacy, Thermal resistance, beamwidth, CRQ, CRI, sidelobe glare etc etc. ESR is one quality factor that stands out when you increase power.

I agree you won't find LED foundaries promoting parallel operation, because thermal canbe an issue from current hogging. But if you know what you are doing and what you are getting, it can be managed ( e.g. with distribution losses to add ESR equalization for example) or to improve system losses. The value of total system ESR and its variation must be greater than the NTC effects of Shockley voltage reduction to prevent thermal runaway.
This is most easily accomplished with a lossy Current source and a wide range of input voltages. But again I accomplish this using precise voltage source, precise LED ESR and effective thermal management with minor additional ESR due to exterior wiring in residential applications. For example my universal laptop chargers are 19V at rated 4.5A and 19.5V open circuit giving a source ESR of 111mΩ which is much greater than the ESR of the LEDs of 10 parallel strings of 6 series 1W LEDs on Alum clad PWB. With commercial current source drivers being 1~2$/W, I can easily drive 60 watts of 10 luminaires @80% efficiency or I can use my design with 65W laptop chargers that are only $35 in qty 1 and $10 in large qty.

I add a 150uH 5A torroidal choke and external low ESR cap to smoothen current or increase dynamic ESR at SMPS rate to reduce stress and smoothen current for regulation of a non-linear load on PSU. The distribution lines for power can be up typ. 100ft of 18 AWG also add dynamic ESR at SMPS freq. from inductance.

I believe my conservative design is low cost. But I do not recommend you suggest what I do to anyone who does not understand all the variables. That is why no one else does.

OK Sunnyskyguy, thankyou for your explanations.

I think it's only a side aspect of the discussion, but regarding said MOSFET with "NTC" characteristic, it should be mentioned that IXGK400N30A3 is an IGBT, a bipolar semiconductor, not a MOSFET. MOSFETs have a "PTC" on-resistance characteristic by nature. A MOSFET current source can nevertheless have an opposite temperature characteristic due to the Vth temperature dependency.

ty FvM for correcting my Senior's Moment

Wow- I've been a bit out of the loop following this thread, as I have been taken under other tasks... (new baby girl on the way!!)

Thanks for all the replies, this is really great stuff.

Let me give you some background as to why current matching is so important in my field. I design light fixtures for machine vision. Machine vision is an industry that uses cameras and lighting to inspect products in a manufacturing line. It could be anything from beer bottles, to yogurt containers, to car tires. The lights and cameras inspect anything from your beer level fill, the yogurt container labels, read bar codes, 2d and 3d matrixes, etc..etc..etc. The applications go on an on.

What happens when a light has a failed section? Well, dozens, hundreds (thousands!) of parts will be automatically failed by an automated inspection machine. The camera software uses algorithms to locate and inspect parts, and if the lighting fails, the camera cannot inspect the parts properly.

We're working on a side-fired backlight design which has incredible premise if we pull it off. The backlight has single rows of CREE XTE LEDs on opposite sides. If the current matching is off, this backlight will have total shitty uniformity.

So anyways... I must go back and read all these posts. Thanks all for the amazing input.

- - - Updated - - -

I would like to try to get this thread back on track in order to help my original request.

I didn't intend to have much over-thinking going on with this thread, but this is all good information and I am very impressed by the responses.

Basically I want to do 1 thing -- just have hella-good current matching across parallel strings ( 2 strings, 10 strings, 100 strings, whatever it may be).



We parallel tons of LEDs, and typically just drive them with series resistors for continuous which is "OK"
We also parallel LEDs to use with switching buck converters (minus the series resistors), which is also "OK"

We've been doing this for years, without much problem. We buy LEDs by the thousands in reels, so they are fairly well matched to begin with.

The problem with new high current LEDs is that their dynamic resistance is so low - a tiny change in voltage is a large current change. Even though the LEDs are "binned", they could be off 50-100mV and it makes a huge (10-20%) current change.

Moving ahead and trying to innovate our product line, the goal was to implement current matching in order to improve our uniformity. Right now we end up around 5-10% current matching with no current mirroring (using the standard methods above). Every once in awhile we get a bad string (either caused by an out of spec LED or bad soldering job) which puts it up to 20% or so, so... this current mirror would minimize any problems like this that are seen.

Overall I want to get down to the 3-5% current matching range.

Simple breadboard prototypes of 100mA or less, using standard NPN (2N2222 or similar) yield good results with 5MM leds. When I move to higher currents, things get weird.

My last attempt yielded worse results with transistors installed (about 15% current matching) and better results without (7% in this case).



Essentially a prototype i have now, has 3 strings of 6, (one string has 8 ohms in series to provide a higher voltage drop than the other strings), with the base tied as a reference string. All other bases tied to the reference string, just like the original circuit.

I haven't tried low-ohm emitter resistors yet, but plan to next.

Other Transistors I've tried using are high power NPNs:
1. ZX5T853
2. ZX5T851

I picked these because they have high collector current limits (6A +) and high pulsed current limits (15A+).

What is the nominal & variance for ESR & Rja for your LED matrix? Also what duty cycle and/or PWM rate are you using?

I just installed LED's for a friend's outdoor garden stairs on 60 feet of handrail going to the lower back yard. I installed 13 flat PCA's with 6 x1W LEDs on each 1x6" Aluminum clad boards = matrrix of 78 = 6 x 13 luminaires of 1Watt Cree LEDs. It worked perfectly as usual. He said it was bright like daylight, but he was exaggerating.:razz:

On each 8 foot section of railing is a 1/4"x1" L shaped aluminum strip to block glare and act as heat spreader. I have developed a formulae to prevent thermal run-away on voltage sourced parallel strings of series diodes and each string is the same brightness. In this case I used 6 one Watt LEDs per MCPCB Luminaire because I just happened to have an 85W power source with the right voltage to drive 6 LED's at maximum power, without any need for lossy current source. That is 4.5A @ 19V that works perfectly to drive each string of 6 LED's in a "Luminaire" with 13 such units equally spaced in parallel, then wired from the center through the concrete basement wall to the Laptop charger. This meant I could drive 4.5A/0.3A = 15 lamps., so I chose 13 luminaries on 80 ft of 18AWG speaker wire.


I want to share my formula, but it is rather complex to type here but simplified is based on Ohm's law of ratios. The critical threshold for current sharing stability is determined by the ESR to Rja ratio

As current (I) is increased, there is a voltage rise due to bulk ESR losses (Rs) and a junction voltage drop (Vt) due to temperature rise. These are somewhat independent variables but share the same current within the diode.

If the voltage drop from temperature rise matched the voltage rise due to ESR for the same step in current, the net voltage drop is constant for small~medium changes in current. This is the threshold of instability for thermal runaway if run in parallel. I do not have these issues by design. The key is for a given family of semicondutors (eg. Silicon IGBT, Schottky diodes, White phosphor InGaN LED, Red AlInGaAs LED , each with unique thresholds (Vt), I have a target ratio for ESR (Rs) to Thermal resistance (Tja) value for each product with enough margin to allow for deviation in ESR and Tja for ALL parts. If it were in production I would ensure there is 5 sigma margin. (5σ)


- The Tja is the overall Thermal resistance in deg C/W from Junction to ambient Rja = Rj-tsp+ Rtsp-a or the sum of diode junction to solder point (TSP) , +TSP to heatsink + heatsink to ambient

- The Diode voltage Vd=Vt (T,I,Rja,Rs) + Vs ( T,I,Rja,Rs) ,,,, the diode Shockley diode law for Vt with a coefficient of -0.0015 V/ deg C,

- the effective series resistance of the load ( in this case 6 x ESR ), the diode threshold voltage @25'C ( in this case ~2.6).

As you know mux'd LED's are less efficient due to lower efficacy and higher ESR losses, but then the driver losses are improved.


So the value and variance of ESR is the biggest question then Rja. The Rja/ESR ratio is must be less than a certain value.

**added**
for those with poor Rja such as a parallel string of LEDs on a 12V or 24V, 5meter flex cct reel of SMD LEDs , you may find the mfg has added a small Rs e.g. two SMD for each string approx equal to the ESR of the string. THis lowers the Rja/ESR ratio or raises the ESR/Rja ratio to compensate for a wide variation of ESR in production so no sorting is needed for ESR ( measured as Vf at rated current) such that it can be driven with a constant 12/24V voltage source easily. The value depends on quality factor of ESR and size of LED.

But I did not need this because with better thermal management of Alum. clad PCB on power LED's which have an ESR under 1 ohm and better Tja with heat spreader that also acts as Glare Blocking rail edge so the emitters are not visible from the side when cast down to the ground. THis Alum. strip is commonly used to edge 1/4" ceramic tile for finishing and is found in any hardware store near the tiles.

So please verify the ESR of your LED string and ESR of your transistor driver and if possible the TJa. I measure ESR at 10% of rated current but depending on Vt used it can be done anywhere from 10% t0 300% of rated current for a good linear approximation. I have verified my values empirically on a spreadsheet to agree with test results. The problem with CREE testing is they do all tests for a fixed temperature ( artificially controlled ) rather than a fixed thermal resistance (real world example) and do not state ESR, but you can determine ESR & Vt either way. Rja depends on your thermal design skills with surface flatness, roughness, heat spreader etc..If you are not good at it, it is better to buy the LED's already mounted on a suitable AL PCB. BTW aluminum has higher thermal resistance than copper, but due to lower mass, has faster heat spreading velocity. **

If you could get you constant current source working, these issues would not be as important but then the driver is less efficient.

BTW CUI brands a chip for driving 700mA strings here https://www.ledlighting-eetimes.com...700-ma.html?cmp_id=7&news_id=222907963&vID=44 with pot. R or PWM dimming from 0~100%. When you switch @200Hz the low ESR of power DC source and low ESR of LED's effectively clamps any inductive spikes with 1uF added. but costs more than 20W of LED's ($8 + other parts) so ease of design has a cost. ** Whereas my design is both cheap with SMPS cost < $0.30/Watt and more efficient with prudent thermal design and CV management of Resistance ratios.. I don't recommend you do this unless you learn what I know.

Maybe this is a silly question, but I'm curious why you need multiple parallel strings, rather than a single, long string with all the LEDs in series. Safety? Scalability? If everything's in series then it's guaranteed to have excellent matching.

If you insist on parallel strings, depending on just how precisely you want the currents to match, it's trivial and cheap to use op-amps to control the current using 1% current sense resistors. It should even be possible to use a linear regulator as a current sink, by configuring it to output a fixed voltage and feeding it to a resistor (you'd want to use an LDO controller which controls external pass transistors, like this).

Even easier, you can allow the currents to be mismatched, and interleave the LEDs (so half of each string is on each fixture). Every other LED on the fixture will have the same brightness, but because the brighter and dimmer LEDs are right next to each other they'll average each other out. You'd need less components (at the cost of more wire).

Sunnyskyguy:

So the value and variance of ESR is the biggest question then Rja. The ratio is must be less than a certain value

Click to expand...

The LED foundrys do not provide statistical data on ESR or Vf spreads. So i dont see how its workable, what you suggest.

I'll come forward now and say it........Parallel LEDs are for Engineers who do not know how to design step-up type SMPS LED drivers.

Sunnyskyguy, you have told that you are selling LEDs to your clients, i cannot help think, that with all these technical explanations, you are merely protecting your customers who are not capable of designing step-up converters , and so are putting LEDs in parallel. -they have to put LEDs in parallel, because they know no other way.

I am sure your customers are delighted that you are defending their putting LEDs in parallel........i believe your customers (the ones who put leds in parallel) dont know how to design step-ups.

I had to come out and say this, because saying that LEDs in parallel with no form of current equalisation whatsoever is OK, is akin to saying that the Earth is flat.



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 !!

The only way you can put LEDs in parallel with no form of current equalisation is to buy specific matched parallel LED modules from the LED foundry......they make sure the LEDs are matched, and then mount them on special MCPCB and ensure excellent thermal coupling.......................and by the way, these LED modules are EXTREMELY expensive.

I do it this way because it is cheaper now to buy universal CV sources of 85W laptop chargers than design and build it. $0.30/W The cost of high efficacy , high power LED's is ~ $0.50 ~0.75/Watt which will drop 10%/yr for small qty until both are $0.10/Watt. I expect 42V or 48V to become a standard interface voltage for high power light engines. At present this is too rapidly evolving for the Zhaga consortium to standardize, but check it out for yourself. I only chose 19V for readily available cheap reliable power sources that are high efficiency. I chose this for several other reasons.. I could design 6 LED string luminaires for easy distributed lighting.that requires 19V. My next home will be wired for DC power distribution of power for LED lighting in CV mode so I can use this method of design, until such time when CC mode power supplies becomes more economical.

Zhaga consortium is an industry standard for mechanical light engine interface. In the future electrical driver interface will be decided. If one understands the Rja/Rs ratio then CV can be used. otherwise CC is the safe approach. Also if there is a solder joint or gold wire bond failure in a series string, you need a shunt zener to protect the entire string from failure. internal Zener only protects the internal failures and not interconnect failures from environmental abuse. Mechanical, climactic and EMC , also rodents love to eat PVC insulation during breaks in hibernation.



I challenge anyone who can back up their words, ( maybe even Grizedale) to disclose their design and compare costs of driver, LED's and heatsink, efficiency and reliability. Efficiency can be improved in my case with a 85% typical purchased SMPS AC to DC which can be improved to 95% for high voltage strings around 48V ( common telephony SMPS ).

- - - Updated - - -

I will reply to criticism if it is based on facts rather than opinion. BTW at risk of repeating myself too many times. I get tight control of ESR from using 1 or 2 bins of voltage for my customers, on 5mm LEDs and I know the consistency of any supplier's high performance ahead of time for unsorted LED's. Every batch is matched by virtue of the process. Batch variations are reducing every year.

Sunnyskyguy,

you needn't challenge someone like me......if your ground-breaking statements really are correct then you are going to revolutionise the entire LED lighting industry world-wide.

Tridonic, the multi-million pound LED lighting organisation , would be very interested to hear about your fantastic suggestions, if there really is great substance to them......they would literally be re-designing their entire product range of LED lights starting now, to incorporate your great ideas.

But i have not heard a single other person on this forum agreeing with your statements, let alone external multi-million pound lighting organisations.

When going for an interview as a LED lighting engineer, if you say you put them in parallel with no current equalisation, its end-of-interview, and next candidate in.

Consider GE Lighting in USA, will you tell us that they are putting their LEDs in parallel?......i strongly doubt it..........i strongly doubt that you would name a single LED luminaire company using LEDs in parallel, unless they were using LED modules expensively manufactured at the LED foundry.......but you cant stay in business and pay that much for your parts.


I think Tridonic would like to know if there are parallel LEDs (with no current equalisation) in Swiss tunnels, because they have the Swiss market mostly covered, and they would be interested to know that some "bodge-it-and-scarper" lighting company had taken some of their market share with parallel LEDs.

Heres a book on LED driving by Steve Winder:
"Power supplikes for LED driving"
https://www.amazon.co.uk/Power-Supplies-Driving-Steve-Winder/dp/0750683414

Does any reader here think that we can now contact Mr Winder and tell him he's missing pages about how to drive LEDs in parallel with no current equalisation?......i somehow dont think so.....Mr Winders book mentions nothing about the "Rja/Resr" ratio.

In the link to this book supplied, there are free pages to view, search on the word "parallel" and you can read for yourself Mr Winder saying that putting LEDs in parallel is a mistake.

-But in this thread we are stating the opposite.

Is Mr Winder wrong?.....will he now have to refund all buyers of his book?

I was watching the movie Game Change about Sarah Palin's Campaign starring Woody Harrelson and it reminded me of the intelligence in the response to my comments. I wonder if anyone else has any intelligent questions or comments.
At the risk of explaining the obvious to someone who lacks design experience but is quick to criticize without theory or practical experience, I should point out that not all diode geometries of PN junctions lend them selves to uncompensated parallel operation. In fact very few. But that does not negate the criteria for prevention of thermal runaway nor the real world examples I shared of examples that work. If anyone does not have the courtesy to understand by asking rather than seek the comfort that comes from being righteous and wrong at the same time, think again before you speak.

Inteligent questions are welcomed.

Would anyone be interested photos of my recent installation of 85W 19V CV LED engines in parallel? and measurements? Actually they are so bright, friends requested dimmer control which illuminates entire side stairs and patio deck. All equal brightness and cool, but too bright at night!!


*added*
Just read Steve Winder's published book, "Power Supplies for LED Driving", so I have some editorial relevant comments to make.
His Preface was written in 2007 , so his research is now > 5 yrs old.

It is a well written book based on solid principles of SMPS Design by Keith Billings book ( who I worked with during my Burroughs days in the mid 80's ) and is easy to read without too much complicated math. At the time he was a Field Eng for Supertex and was not a Design Engineer but worked with many designers and customers alike to solve MOSFET issues with LEDs and SMPS. There are significant improvements in the quality factors of LEDs and SPMS in the last 5 years, although some existed 5 years ago. Mr Winder's logic is good, but his assumptions were different from my experience.

I consider his book excellent overall with a brief overview for young designers and advanced design info for SMPS designers with Common Problems in each section. His technical references are fairly accurate and informative to those interested with the following differences from my specific application;

He reports ;

1. On page 13, "The ESR (equivalent series resistance) of a low power 20 mA LED is about 20 ohms" I ones I supply happen to be around 10 ohms since 5 years ago ( and getting down to 5~8Ω. for HB White.) However he agrees Zener diodes have a higher ESR than similar voltage LEDs.

2. On page 14, "Typical Forward Voltage, Vf Red = 2 V Blue=3.5V, " This may be true for 1W LEDs but he shows a graph with a sharp knee where he indicates Vf.

This, unfortunately is not quite correct. The Vf is the forward drop at rated current, so it includes the threshold voltage and IR drop due to ESR above the threshold Vt using a linear approximation. In any case , he uses Vf to mean both threshold and total forward voltage at rated current , where it should be Vt (threshold) and Vf (forward at rated I) Many manufacturers do the same, so it is not his fault.

3. The same Fig. 2.5 for diode curves show effectively the same ESR for all colour LEDs. This is not true but for his purpose to show differences in Vt with a sharp knee of the curve is acceptable. ESR for different colours at same power are not the same, but not relevant for this discussion. Furthermore, I did not indicate previously, but ESR also has a PTC characteristic while the PN junction usually has a larger NTC curve, and I have factored that into my designs and recommendations, just not shown all the formulae as that is a p/n unique characteristic which is not specified typically. Although Cree does indicate this in their webpages for parameterization design guide of certain power diodes, but not generally disclosed as it is a proprietary quality factor of their devices. (read as ... Intellectual Property)

4. on page 15 "Common Mistakes", Mr Winter indicates "For example, a 1 W white Luxeon Star has a typical Vf = 3.42 V, but the minimum voltage is 2.79 V and the maximum is 3.99 V." They have improved as has everyone else.

Today , the spec for the Rebel ES ranges from for Vf = 2.5min ~ 3.5max on pg.6 https://www.luxeonstar.com/v/vspfiles/downloadables/DS61.pdfhttps://www.luxeonstar.com/v/vspfiles/downloadables/DS61.pdf

Also ∆Vf/∆Tj = -2.0 to -4.0 [mV/°C] ... but it is critical to note the Vf range above is given for Tj = 25'C~110°C @ 700 mA.
This is a common mistake in misunderstanding differences in LED Vf's the reason I believe it is widely misunderstood and triggers fears in the minds of those who are unaware of the test conditions when applied in methods I have used.

5. ESR of all diodes has a small function of temperature function. However on Page 6 of Luxeon's spec above it indicates this device has Vf= 2.85V @ 350mA, @25'C which is indeed a tighter value at constant temperature ( as expected ), but tolerance is not given. The table on page 6, indicates Vf=3.00V @ 700mA, we can roughly estimate average ESR over this interval is 0.15V/0.35A= 430mΩ @25'C.ESR drops to ~300mΩ from 0.75~1.0A

6. On page 17 of Mr. Winder's book, in the Voltage Source chapter, "Driving a constant voltage load from a constant voltage supply is very difficult, because it is only the difference between the supply voltage and the load voltage that is dropped across the ESR. But the ESR is very low value, so the voltage drop will also be low. A slight variation in the supply voltage, or the load voltage, will cause a very large change in current"

I disagree, it is not difficult, if you know the ESR of your driver SMPS, cable to LED wire resistance and LED ESR, it is not that hard. Of course it IS hard, if you do not know how to get these variables.

7. On page 18, "Most supply voltages from a regulated supply have a 5% tolerance, but from unregulated supplies like automotive power, the tolerance is far greater"
This may be true from his perspective, but most PC SMPS I see are +/-1~3% and the Laptop charger I choose is <±2% from no load to full load or 19.25 ±0.25 @50% power ±50%... ie. load regulation. I believe this range is due to the DC cable wire resistance. Tolerance on a single supply is the sum of temperature, initial error and load regulation losses as the internal reference voltage is no longer a zener these days but rather a precision 100ppm band-gap 1.25V reference diode embedded in some chip. My recent installation was 0.1% initial error at rated load and 0.5V drop from ESR of cable and SMPS. effectively giving ESR of my SMPS tested @4A ≈ 125 mΩ.

Consider a 4A load @19V. Each series string of 6x 0.5 Ω LEDs per MCPCB appeared as 3 Ω total which is much higher but with 13 "luminaires" or bare aluminum clad PCB's mounted on Alum edge strips for heat spread and glare block appeared as 3/13≈230 mΩ to the SMPS. So you might understand now how effective driver resistance ESR total in outdoor distributed lighting can help equalize small variations in Vf for CV operation of some devices with a a well regulated inexpensive SMPS that sold for $30 (+ tax) in qty one. Prices may vary for your sources as much as $50~$75 for 85W.

I think you all can agree if you add a series resistance large enough any parallel operation is stable. I just have found the optimum minimal resistance and use wiring to create that resistance to be greater than my variation in ESR between devices and hence equalize the load for equal current. I do not expect everyone to have this level of awareness. Even Field Engineers who write books about the subject, so I am not surprised the subject is taboo among those who do not know how it is possible. Yet in many installations it is. If the wires losses were zero, then it might be more difficult and depend on perfect matching of LEDs, but I do not and my customer who uses a wireless inductive power sender does not since the effective ESR of that system is high enough so that CV power works well, but one can sacrifice more losses and increase the margins using a CC source.

The higher the distribution voltage and lower the current, it should be possible for a 100V 10A network giving 1kW of distributed power for lighting possible using CV DC power from a central source with very high (98%) efficiency with losses being just copper wire distribution losses.

If one matches the light engine voltage to a high voltage battery such as 48V for lead acid, one could have emergency lighting for a building from central battery power and use existing AC 14AWG wiring, modified for DC only and keep the battery charged to drive the LEDs during power failure for many hours. With my design the biggest cost is the LED's.. With present commercial & consumer Luminaires costing $5~$10/Watt vs my < $1/Watt, you can see why I choose my design. But actually my biggest choice is not the cost, but the quality of light and lack of glare in the luminares with polished aluminum L edge strips that block the glare AND spread the heat to a mild temperature.

Next I have to install a motion sensor to the timer and dim the LED's because when my friend asked for bright, he did not expect it to be "that bright" so will drop the voltage with a a diode or two as a quick & dirty solution with a cheap LED as a photodiode for daylight sensing and another for motion sensing to control the brightness with a series pass transistor to control the light when you need it. Ah say yes to Green Power. Actually the biggest job is hand soldering the wires during installation and I am looking forward to using gel-filled watertight IDC crimp connectors to tap into the stranded wire... except they cost as much as 3 LEDs...So if anyone as a reliable moisture sealing IDC crimp connector for $0.50 let me know...

Mr Winder goes on to describe how to use series resistance to match Vf in each LED or parallel string on page 19 I see now, but he is assuming getting tight tolerance is impossible. It is, if you do not know it is not that hard to achieve when you can work closely with a supplier. But I am not that difficult to work with, just need enough volume to make it worth the effort. :wink:

Keep those CC designs flowing with low cost SMPS and try to get your costs down for LED lighting our highways, bujildings and homes. If you can get anywhere near $1/W for crude designs (read .. hidden luminaires not ornamental ones with glare) let me know. :grin: