help with LED current mirror, pulsed current source

Hello,

I tested a N-MOSFET current mirror to drive two LED strings in parallel. This worked well using a constant current power supply:



Next, I have a "pulsed current source" which can output up to 50A, and drive LED strings with up to 100V. The current is limited via a "current sink" type current source tied to the cathode of the LED strings. The "HV" pin shown in FIGURE 2 below can drive anywhere from 10V-100V depending on what types of LEDs are connected. This voltage is configurable depending on the LED type.

Typical wiring connection looks like this:


Pulsed current between the two strings is not equal - much like it is not equal in continuous when connected this way. The differences can be as much as 30% or more depending on the LED!!


So.....I thought I could connect the same NFET current mirror to the output like shown in figure 3:


But this does not work. Current matching is still way off - and the FETs seem to conduct even when I remove the gate signal(s).

Questions:
- Is this a proper current mirror circuit when used with a "current sink" type current source?
- Would a PFET work if wired slightly differently?
- Why does this not work?
- Why does the transistor conduct without a gate signal even though the VDS rating of the FET is 100v?
-

Thanks for any help!

A disconnected gate wire can respond to nearby static charges, or even ambient 60 cycle magnetic hum.

Are you disconnecting the gate of the Nfet when you remove the gate signal? It should be grounded at such times.

If the insulated gate was exposed to static charge, then it may be perforated. The mosfet may not work properly. It cannot be repaired.

I made a simulation of your circuit, sending clock pulses to the Q3 mosfet. It works as you propose. The led strings turn on and off. Current is equal even if the strings do not have the same amounts of led's.

I am surprised that the mirror in the first circuit worked well without source resistors, unless the MOSFETs were carefully matched. Even in that case if the voltage drop across the LED strings are different, the Vds of the MOSFETs will be different, which should unbalance the currents.

it somewhat works now - I don't know if I had bad MOSFETs or not, or if it was a bad breadboard connection.

I rewired the circuit and it appears to be working> I can short LEDs whilst strobing at 50-60V, 15A per LED string and the current maintains pretty good matching (5-10%). Normal case shorting an LED would cause a string to change 20-30% current or so..


Would adding gate resistor(s) be a better practice?

BJTs better than fets for current mirrors
preferably matched bjts mounted to be well coupled in thermal way.

The reference transistor must have the higher voltage led string....so that usually means you have to add a resistor to that ref string to ensure it is always the highest voltage one.

Why not using a "perfect" dual current source instead of a by design bad performing current mirror?

FvM said:

Why not using a "perfect" dual current source instead of a by design bad performing current mirror?

Click to expand...

The problem is making a mirror that will work with a few different, already designed, current sources (led drivers). The mirror would be placed din the light and is remote to the current source.

And I don't think bjts would work in all cases because some of the loads are high current and high DC voltage in strobe.

you cann arrange it so that the BJT does not see the high voltage.....a BJT in a current mirror would in the way that you describe would not see a high voltage.

i dont think you can get a mosfet current mirror, mosfets even from the same batch rarely have exactly the same VdsTH voltage.

The problem of all current mirrors is that they have good balancing effect only in one direction, for Vout <= Vin. The restriction is even tighter for a BJT current mirror. At the end of the day, a pair of series resistor probably provides more effective current balancing.

FvM said:

The problem of all current mirrors is that they have good balancing effect only in one direction, for Vout <= Vin.

Click to expand...

You can get around that with a 2-way mirror (for want of a better word). e.g. The circuit below ensures equal current in the two LED strings, no matter which has the higher voltage drop.

The separate low voltage supply on the side isn't really necessary - the collector of Q3 and the top of R4 could just be connected to the top of the LED strings, but that would be a bit less efficient.

godfreyl................your above circuit is excellent.................im not criticising your ingenuity here.....thats top marks for you there........and you only designed what the poster would have wanted............but i would say for that price, just use another led driver and have each string have its own led driver, theyre not that expensive...and the heatsinking demands go down.

- - - Updated - - -

godfreyl................your above circuit is excellent.................im not criticising your ingenuity here.....thats top marks for you there........and you only designed what the poster would have wanted............but i would say for that price, just use another led driver and have each string have its own led driver, theyre not that expensive...and the heatsinking demands go down.

but i would say for that price, just use another led driver and have each string have its own led driver, theyre not that expensive...and the heatsinking demands go down.

Click to expand...

Yes, if we presume industry standard switched mode LED drivers. The original post is for some reason referring to a linear LED driver, and the poster claimed in post #7, that a second driver won't be an option.

You may want to describe the situation as finding a silly workaround for a deadend design. Without thinking too much about the specific problem, we all know that similar situations sometimes happen in real engineering.

A solution in search of a problem. :-D

What is the ESR of the switches relative to the load? Measure incremental V/I to get ESR.
You need pretty good matching in gain of devices for mirrors. ROHM is a good source.

That's why the emitter resistors are there, and to cater for differences in Vbe and temperature.

Sorry Godfrey I meant the original design. Iwas trying to figure out why the MOSFET current mirror was failing , assuming @jstamour has had a chance to replace any bad parts. Of course the bipolar design works well and the value of Re of 100mΩ in the range of the total ESR for 5x 1W LED's so that compensates nicely for mismatched hFE's.

Hope U don't mind this off-topic question, but related.
I wonder what there exists a formula to determine the loop gain for runaway, assuming that is defined by a positive feedback loop gain . That ought to determine what the threshold of instability for a series resistor when LED's are in parallel.. and what degree of mismatch poses a risk of thermal runaway. Considering all the LED arrays in product with parallel strings, there must be a stable criteria. AFAIK I can simulate an LED junction with the temperature sensitive PN junction and the bulk resistance. for a White LED, it is approx. Vled ≈ 2.85V * (-1.5mV/'C)*'C + ESR * I.

I have measured many white LED's for ESR where it is in the 10mΩ range for power LEDs and 10Ω range for low power (75mW). ESR is the critical factor for performance where low ESR gives superior performance and also more risk for parallel operation. Also consider that the eye does not really detect any difference in 10% intensity. FWIW.... (for what it's worth)

Hi

The way I'd think about it is:

1. A difference in temperature causes a sifference in Vled (1.5mV/C?).
2. The difference in Vled causes a difference in current (due to ESR and external balancing resistor).
3. The difference in current causes a difference in power dissipation (depending on forward voltage).
4. The difference in power dissipation causes a difference in temperature (depending on the thermal coeeficient of the LEDs and any heatsinking).

So we have a positive feedback loop there, and the trick is to make sure the loop gain is less than unity.

If it wasn't for my current post-lunch stupor, I'd try to roll that up into a neat equation for stability. Shouldn't be too hard.

P.S. When I chose 100mΩ for the emitter resistors, it was with currents of 1 to 10 amps in mind. 100mV drop across the resistors is enough to minimise the effects of differences in Vbe, while up to 1V drop shouldn't hurt the efficiency too badly. The ESR of the LEDs and the hFE of the transistors doesn't really come into it.

I agree , devices with low ESR e.g. < 10mΩ ( yours may be <<1mΩ) need a substantial Rseries if using Voltage regulation (possible but not recommended),

Buit if using current source then it may have an equivalent ESR when saturated and string of Vf equals supply voltage avail, otherwise {Requiv} of Isource >> Rs. So adding 100mΩ may be useful to spread the heat from current source to series load resistor. Adding small Rs is also useful for verifying current in test, as you did.

The reverse mode I suggested is like a string of caps when diodes are not conducting as Vdd collapses quickly to zero the device with the smallest capacitance becomes reverse charged in the series, Otherwise if all matched ( which they are not) the capacitance divider network decays with equal forward voltage. Obviously LED is more like leaky RC ladder and most white LEDs are spec'd 1~10µA @ -5V which is non-linear, so equivalent to 500KΩ~5MΩ with few hundred ~ few thousand puff (pF){depending on size} near 0V.

As diode goes negative voltage capacitance drops and that device changes charge or reverse voltage even more.

Verifying this requires 100MΩ probe or modelling it with bigger values to prove my point.

As I said, I have done this on a client who used AC drive on back to back series strings and got failures from over-looking effects of capacitance.. including dielectric of potting material around LED terminations in their case, which if equal, improves, if not degrades results. THis resulted in my recommendation to spec in back to back zeners included on LED chip.. Most Cree chips now have this built in.. I did this 5 yrs ago. Most small LED's eg < 100mW do NOT have built in Zener protection, but some do now because of risk that clients run them in large series strings with pulse or rapid decay of supply. e.g. noise pulse.

Thanks for your comments.

If the device capacitance current is neglible compared to leakage resistance then a series string of LEDS when pulsed, ( voltage controlled leakage) is a moot point as they all share the same current , so the device that starts to leak more shunts its voltage leaving the others to share the voltage, so it is a moot point. But with capacitance, the memory creates a non-linear reverse voltage effect on a large series string on the best part i.e. the part with the least leakage.. This in turn causes more leakage when Vr> 5V say from a pulsed 20V string of 6 LEDs or 100V string of 30 LEDs or so. (depending on device size and rating). in such cases, zener protection is mandatory, as CREE has done.

Sounds interesting Sunnyskyguy, but its difficult to see how the voltage across an LED chain would fall that suddenly that the reverse voltage effect problem occur severely enough to do any damage.

If its AC mains for the back to back leds (ie anti-parallel) then 60Hz is quite slow.
If it was mains, then i hope the person used transient protection, because a transient may have zapped the leds.
Mains, also, as you know is variable, and so i hope they designed for waste case, highest mains.

It would be interesting to see if any of the semiconductor companies eg Texas etc have any special led drivers which counteract the problem you describe. (the problem of leds developing a reverse voltage higher than their rated reverse voltage when the driver is switched off)

But sorry to OP, as i have detracted, .........

even sorrier OP because aren't current mirrors generally a waste of time unless you want to play with circuits for fun?, as they dissipate plus plus and led drivers are cheap and easy to use. Mirrors need matchd trannys and unless you use godfreyl's intricate (but very clever) circuit above, you need to add a power resistor to ensure the reference string is always the highest voltage one.

Also , I think LEDs in parallel is out of the question. No representative of any LED manufacturer has ever said that general power LEDs can be safely paralleled. LED companies dont care about making sure that Vf's are matched nicely, LED manufacturers are only interested in getting more lumens per watt........they are not interested in getting matching Vf's even if it often happens incidentally. Osram documentation specifically states that because LEDs are optimized for light, the Vf tolerance can be very loose.

Grizedale<<

What current do you think it takes to damage a 5mm LED in reverse bias at say -10V

I was referring to 5mm devices I have tested at customer site, which were rated -5V @10uA rating and how many have you tested and verified such as V-I curve tests.

I was not suggesting LEDs to be spec'd as tight Vf , however they are in a single batch.. So tight it would make you appear generous ( in your giving the benefit of doubt) However a 5mm works far better than a 3V zener because the ESR is lower , meaning sharper knee. How many have you tested for ESR and reverse bias leakage?
:roll: