LED drivers with very high analog dim range?

Most of the time analog. But for pre-programmed scenarios this would come from a DAC. SPI, I2C, CAN would be ok as well but an additional analog tie-in would be nice. However, right now I'd only like to know whether there is any chip with more than 250:1 analog dimming range, regardless how it's controlled.

op

if

usual

and

That

I think you may have made an invalid assumption.

?-)

Quote "Both are claiming +/-5% current regulation, ...". Well, that ain't quite cut'n it :-)

Output current regulation typically 10%, max 15% (page 3). That's what I saw a lot with other chips as well. I always wondered why that is so large a tolerance, all the discrete stuff I did is in the very low single digits worst case and could be under a percent if a narrow-tolerance reference is used.

The datasheet kind of hints why, the current sense comparator has a major offset issue, 10mV:

They don't state such data for the others but probably not better. So you end up doing your own reference and comparator, just as I said, and as shown in the schematic on page 1 of the datasheet.

That is one busy circuit. Exactly what I said, external reference, external comparator, external everything. Tolerances don't look very enticing and Digikey has already obsoleted it. Kind of early for such a young chip.

Why? For visual brightness, even relative brightness,

A 20% difference in current is barely noticable. Try it.

Bin codes are only used where the application is fussy. It would be insane to use binned LEDs for Christmas tree lights. They're usually unbranded floor sweepings to start with. Probably 3:1 variation in brightness.

LEDs age anyway, and change significantly with temperature (both brighness and wavelength). upscale commercial/aerospace. 5% clearly was not an acceptable trade-off

I have a tough time with imposed specifications that greatly increase cost and complexity (reducing reliability) without providing any benefit whatsoever. Simple, conservative and always works is what I like.

I did :-)

Side by side is what you have by default in RGB-LED lighting. I am not particularly fussy about color spectra and my color vision isn't that stellar to begin with. But some people are.

Same here. But there are markets where that just isn't enough. I remember the words of an automotive guy when something mundane was tested. Pulling out the ash tray or whatever, where you'd think nobody cares. "That makes a not so noble sound, this is not acceptable!". And the design was corrected.

an op

that.

Now if

To

usual

and

chopper

That

PFC

to

OK i misunderstood the requirements then. Linearity and accuracy aren't the same thing though.

??-/

In lighting, they kind of are. This is because a gross inaccuracy will badly skew the spectral response over the dim range in a RGB lamp. There, three drivers must be ganged rather precisely.

te:

rote:

g an op

hat.

Now if

To

usual

y and

per

That

PFC

to

..

n't

I wonder if a battery charger like for example:

-Lasse

I wish, but unfortunately not. The Ifb has an offset tolerance that is roughly 10% of full scale.

rote:

ing an op

that.

. Now if

V. To

e usual

lly and

opper

t. That

ing

me PFC

ed to

L...

at I

and

h a

ren't

I see that, it is also way over your budget. chargers seemed like a good candidate since there are several available with ~0.5% references but I guess they only care about voltage accuracy so you'd need to add you own current sense amp, and you would still wouldn't have a way to control the current

btw. is the reference accuracy even important? if you want to control the current with a voltage won't that be you reference?

-Lasse

op

if

usual

and

That

It is important. Way it works is roughly like this: There is a switcher of suitable architecture. Which one will depend on the available input voltage. For example, a SEPIC for mobile applications and something like a PFC-equipped buck for 120/230VAC grid operation.

This converter sinks current into a bunch of series-connected LEDs and that is the current that must be precisely controlled and dimmed. Voltage will also be monitored but only as a safety stop if something in the chain of LEDs breaks, so it doesn't go sky high.

There could be three of these for RGB applications, mood lighting and such, Precision is important because many people are picky regarding the color spectrum. There could be even more if you have an array where sometimes they have to be absolutely at same brightness and sometimes on purpose not.

All this may sound a bit "over the top" but one must keep in mind that for many years to come LED lighting will be pricey and thus it'll be posh boutique applications. It's like buying a Bentley where, after plunking down a hefty six-digit amount of money, customers expect everything to be perfect.

dding an op

an that.

er. Now if

the usual

ially and

chopper

set. That

bring

some PFC

y.

ng a

nted to

e-L...

what I

so

So

, and

ry

uch a

aren't

ll

I think I understand why it needs to be accurate, my point was that if you control the dimming with an external control voltage, it is that voltage that needs to be accurate not the reference in the switcher The switcher "just" need to regulate the current accurately to the external control voltage

-Lasse

an op

Now if

To

usual

and

chopper

That

to

The external voltage will always be accurate, that is the easy part. The reference in the switcher will also have to be accurate and it shall be no problem to do that with a divided-down bandgap, as resistor ratios are very accurate in ICs. The main problems pop up because of the poor offset specs of opamps and comparators in such ICs. They are usually

I can do much better but then we are back to lots of external parts, and then I can as well roll my own switcher as usual. That's what I wanted to avoid.

Joerg:

I am disappointed in you. :)

First, You spend years as the hands-down "design that circuit" contest winner, with statements like "I can design that with two transistors and a cap!" and now you're looking for an integrated circuit??

Second, I'm pretty sure that the 250:1 range is a consequence of simple device physics: the power transistor has a minimum pulse width that it will support, and the inductor has a minimum frequency that it will support. The product of that minimum pulse width and minimum frequency pretty much defines the analog dimming range of the circuit.

If you have the room, consider using a circuit like that for the first

If that's too big, and if you really need the space savings, then have Jim Thompson design you a chip...

This stuff has to be super small. To give you an example of another design I am involved in right now: It has to fit into a space of 0.500" by 0.100" by 0.040". Plus lots of channls. Try that with discretes :-)

True, but there are several extensions to that which have not been exploited at all to date. But this is the point where I'll have to keep my mouth zipped. Suffice it to say, from the steering side of things I am quite sure we can go to a few thousand to one, more than we need to.

Yes, this will definitely have to be an IC. There is no way around that.

=A0Adding an op

per

than that.

ower. Now if

h the usual

ntially and

a chopper

ffset. That

n bring

t some PFC

day.

ding a

wanted to

ine-L...

hat

s what I

s so

ow

s a

r. So

id, and

e,

very

such a

cy aren't

will

ter

is

ke

in

he

on

yeh 10mV doesn't matter when the reference is the usual 1.25V, not even that many switchers where you can actually get the open to set the reference

seem to have tighter specs and option for external reference, but you'll need some gain in the current sensing to get down to a reasonable sense voltage

I guess external fets are trouble enough so adding an opamp is out of the question?

-Lasse

What I fail to understand, still to this point, is that Jörg is doing some kind of special RGB lighting, with human color vision as the recipient -- so far as I've been able to gather anyway. I've done a lot on this score in the past, working with Seimens and Osram. Custom ICs were fabricated and I programmed them for testing, binning, and so on. I know a little about the requirements and I know a little about just how much difference, and at what wavelengths there is more and less sensitivity with humans noticing. And PWM was in the picture and worked fine. We used both current control and PWM within that. But it was very high quality as far as users were concerned. And these were also high-demand users

-- aircraft instrumentation observed in darkened situations as well as lit. Current control by itself, even at the level of 1:1000, was entirely insufficient if the LEDs themselves weren't either binned or else had their current settings calibrated as well.

I know he's not disclosing all the details. But the narrow focus he's exhibiting suggests a lack of visibility on other, more important issues which in combination could make cheaper approaches work well. At least given my years of experience with human perceptions and the state of the art in LED manufacturing a few years back. But I'm ignorant, so I'll leave it there.

Jon

an op

that.

Now if

usual

differentially and

chopper

offset. That

bring

PFC

wanted to

what I

and

aren't

That goes into the right direction. 2.5mV is still way too much though.

Nice reference indeed.

That gain should be on the chip. It would require some sort of internal chopper to push the offset down to the uV level, followed by a filter to remove the chop frequency so it won't mess with loop stability. I guess that's already asking too much in that market :-)

Yeah, pretty much. In an LED industry product this kind of stuff should be integrated. It's not rocket science, we even do similar things in ultrasound ICs where quantities are orders of magnitude lower. Why it's not done for LED, beats me.