On Mon, 18 Jan 2010 22:19:11 -0600, flipper wrote:


Or put R1 under the original NPN emitter.

Yes, that removes any concern about variation of LED voltage drops
affecting low-end brightness. On the other hand, you'll need a
transistor per led (or pair of led's)

Yep, that's the downside.

You also don't need to worry about 'low bat' and regulator dropout,
though.

Make a trimmable Vref for the top end of the pot and size the
resistors for 18mA at max and you have 0-18mA irrespective of 'LED
drop', temperature drift, etc.

JFs circuit has the virtue of destroying the pot and/or transistor at
maximum pot rotation.

I still like the LM317 feeding multiple LED+resistor branches. It has
regulation, current limiting, thermal limiting, and is bog simple.

Yeah, those are it's advantages.

It strikes me we're, pardon the pun, shooting in the dark a bit not
knowing just how 'low' it needs to go.

Another point is the eye's response is logarithmic so it might be best
with a log pot (same goes for PWM) and that might mean the individual
transistor approach works best for the low end. Either that or custom
shape a linear pot with a resistor from wiper to one end.

Really need to do some testing to see what the desired parameters
really are.

I admit I haven't tried linear current dimming white LEDs so I don't
know how much the color temp shifts.




Not to change the subject, NO!, but I tested some super-efficient
T1-3/4 Agilent green LEDs to see the minimum current that would make
light. Dark adapted, staring straight into the LED, the visual
threshold was just under 1 nA. These things are decently bright in
room light at 1 uA.

John