Ping Jim T. precision current mirrors + multiple additons?

I assume you're asking about bipolar mirrors?

What kind of precision are you looking for? ...Jim Thompson

"Ballast" resistors in the emitters greatly improve matching AND greatly reduce Early effect. Generally you want 0.25V drop or larger... the larger the better the output impedance AND the reduction of offset mis-matching.

You can run the math, but it's transcendental, so numerical solution is required. ...Jim Thompson

accuracy).

The ballast resistors also limit the compliance voltage. In my continuing headphone amplifier explorations I ended up looking at the internal schematic for the BA5406 power amp chip - the way they get around the compliance voltage of the current mirrors compromising positive output swing with low voltage supplies is they just bootstrap the current source, current mirrors, everything!

nt

nce

, is

ied

ccuracy).

:)

=A0 =A0 =A0...Jim Thompson

I'd go for at least two Vts, probably 3, so 75mv. But 25mv is good enough to get around serious current hogging.

Note that if you have a string of BJT mirrors and you saturate a BJT, the whole string is effected. MOS is much more forgiving.

accuracy).

No I cannot run the math, I could draw the circuit into LTSpice?

Not tried that yet, might enlighten me with the LED drivers.

Accuracy now is within a few percent, okay for LED strings from simple boost switcher with current feedback from one string, and a zener from boosted output to stop over-voltage if the strings open.

One string has the current feedback, other strings' current mirrors attach bases to the first transistor. I'm wondering if I can get reliable proportional currents to the controlled current by changing the emitter resistor, and need only max about 16V compliance, too much and the poor mirror transistor is cooking with the voltage across it.

Another thing is to dim some of the strings, but with duty-cycle dimming that's when the current mirrors go crazy, as the base current from an unloaded mirror transistor (acting like a simple diode) goes to the reference string and upsets the remaining LED strings.

So I'm looking for options, add proportional currents and/or duty cycle drive each or groups of strings independently. Answer for that is to stop the simple mirror base current doing a damage when one turns off a slave string.

Here's a pencil circuit of a basic LED driver, no magic, just an NCP3063 switcher on boost configuration with current feedback and overvoltage limit, current mirrors for additional strings of series connected LEDs:

Grant.

...

I've used 20 to 39 ohms without much performance difference, so I was in your ballpark :)

Grant.

Not by that much in this application, I suppose I could stand up to a couple volts before being worried, current circuit conversion efficiency is over 80%, so there's a little room for better control.

Not quite where I'm at ;)

Grant.

...

Thompson

Yes, but the string voltage is known, so that can be avoided, apart from the issue of turning off the string. Which may be exactly what you're referring to, in which case it is the problem I'm looking to avoid, as the mirror transistor transforms to a diode upsetting the control string, took a while to see what was happening too, possibly a case where simulation would've been a lot clearer, faster :)

Not sure I want to use MOSFETs, much higher drops compared with bipolar?

This particular application is not something I want to add opamps or a lot of stuff to, perhaps another transistor per string or something?

Unsure what topologies to try, so I looked again at Jim's comparison page for ideas, and am a bit lost for ideas on this at the moment.

After all, it's only about making some light with lots of cheap LEDs rather than those nice big expensive ones I'd be a lot more upset about blowing up.

Surprising how warm a white LED runs with 20mA through it, when to have a hundred mounted on matrix board. I'll add photos to the web page mentioned upthread too:

Grant.

accuracy).

Certainly! Anymore, I "solve" most of my transcendental equation in PSpice ;-)

Nice! ...Jim Thompson

...

Doesn't suggest much to me? better that terrible :)

But it's funny, looking at a newly drawn circuit, I can put a pnp switch in series with any mirror emitter to open circuit it, no effect on the rest, not see that before, plus I've had some weeks away from this project doing other things.

Grant.

on

=A0 =A0 =A0 =A0...Jim Thompson

er

up.

oned

Upsetting the diode (mirror) reference is exactly what I am referring to. Mosfets don't have this problem, but they only way to get a decent mosfet mirror is to do it on a chip.

There are "sat catcher" circuits to stop the excessive base current from the saturated BJT. You would find them in LDO designs.

The nice thing about MOS is it fails softly. You clip something, it generally doesn't propagate a failure down the chain. Mostfets are more of a one way device then BJTs. However, they are less precisely manufactured, i.e. parameters are not as tight. When the dust settles, I really rather have a MOS amp than bipolar, simply due to how it fails softly and needs less protection. BJT junctions aren't all that rugged, hence SOA.

Thanks.

But you can blow a gate by looking at it cross-eyed ;-) ...Jim Thompson

I think what you need is a form of mirror that has a "helper"... an extra transistor that supports the base current in bipolar mirrors.

See page 7 of this presentation...

(Add emitter ballast resistors to improve matching.)

For amusement of the student: Duplicate the helper device, base and collector connections the same, but have the emitter go to a mirror output collector. Voila! Mirror can't saturate and upset (to first order) other mirror outputs ;-)

...Jim Thompson

Grant a écrit :

You can save a bit of power when reversing your current mirror (upside down and with NPNs) then using the emitter ballasting resistor as your feedback current sense resistor (your 1.25V ref drop is wasted anyway as VBC on your mirror non FB branches).

Yes, thanks. Reading through that I remember the original problem was to scale the reference current up to LED current, for example run 2mA reference, then have many 20mA mirrors hanging off that 2mA reference -- didn't work, and using 1:1 reference current produced a lot of heat, from there I figured to run a LED string as a load for the reference current.

After that effort, scaling LED currents is lesser requirement, so what I'm chasing is to not disturb the reference current from the mirrors, when switching strings at different PWM rates for dimming.

Some ideas to work on...

Guess I'm the student here ;)

accuracy).

You're right. Originally I was trying to 'hide' that 1.25V with scaling the current mirrors, but it didn't work out for me.

The circuit grew from complex directions to what it is now. In hindsight I would've been better off simulating some of that. I'm much more at home using opamps and microcontrollers than the simple transistor circuits.

And npn sinks will be so much easier to run with a series PWM npn (or maybe easier with a MOSFET switch, 2N7000) in the mirror emitters, from a uController for my dimmer version. A different way to do what Jan P. did with his uC controlled big LEDs. Similar end result. Not sure if I want to put a PIC on the Internet though ;) Serial via USB probably the way I go.

All this because inductors cost much more than the switcher chip, so I try to get most from a single inductor.

Grant.

Ahh, the beauty of two-stripe emitters ;)

Actually, this trick has limitations on the output collector voltage, if it gets too high the reverse-biased emitter junction changes from protection to problem, whether IC or discrete. There's a better way...

steve

accuracy).

In my IC world lateral PNP's have enormous BVeb_ ;-)

accuracy).

...Jim Thompson