I have a scenario similar to Tuning a constant-current LED driver, per channel:

copied link of schematic

also considering use of two TLC5928 but with

  • If <= 20 mA per channel
  • 28 channels have Vf ~ 8V < 9.6V
  • 4 channels have Vf ~ 4V < 4.8 V
  • Vled driven by regulating 12V down to 10V (0.4V headroom for turn-on region of CC driver) for most channels, 5.2V for four channels
  • Rref = 2.52 k

based on

Rref vs I

In the other question, it's suggested that dimming be done by either:

  • moving to a TLC5948A that has built-in PWM dimming; or
  • PWM Vled [by using] a high-side switch circuit which (depending on the rest of your circuit) could be as simple as a P-channel MOSFET

I'll do one of those if I have to, but I'm somewhat uncomfortable with the idea of PWM dimming on this board as it will also have a 2.4 GHz RF section (even though it's under a can). For that reason I want the broad design to reduce EMI where that's easy. Could I, instead of PWM dimming, fake a higher reference resistor by inserting an NPN pair in series with the minimum resistances? Something like


simulate this circuit – Schematic created using CircuitLab

Would probably need a matched pair like the BCM847BS. Linearity is not very important so long as the left and right drivers are matched enough for the human eye; the brightness ramp can have an arbitrary compensation curve in software.

DAC ideally non-switched, maybe an R2R ladder since resolution and linearity aren't very important. Any DAC output and transistor drive changes would be trivially slow, on the order of a couple seconds. I'm on the fence about the regulators being SMPS or linear and am now leaning toward linear, because it won't produce switching noise, and there's lots of room in the heat-sinking and power budgets.

Is this approach practical? Will NPN saturation get in the way of maximum brightness? The BCM847 claims gain matching of 10% and Vbe matching of 2mV. Would the worst-case mismatch be visible to the eye?

  • \$\begingroup\$ I don't want to bother performing a sensitivity analysis on this, mostly because I'd need to ask more questions (some I'm likely not aware of, yet, and would develop out of a discussion which is likely inappropriate here) just in order to figure out what partials I cared to consider. But I don't like it. And I've had long experience working with calibration of large LED panels on a pixel by pixel basis for commercial systems that had to meet human color and brightness expectations. PWM is the way to go for dimming and those are the results from a large number of people looking at this problem. \$\endgroup\$
    – jonk
    Jan 16, 2022 at 19:29
  • \$\begingroup\$ Most higher power panels built up from driving lots of LEDs also have dissipation as another aspect to consider. I don't know if you have heat as a consideration here (that's one of those questions.) But you may. In any case, when you ask the question, "The BCM847 claims gain matching of 10% and Vbe matching of 2mV. Would the worst-case mismatch be visible to the eye?", for example, you are asking a complex question requiring good data on the CIE models (which I have here, both for color and brightness, having derived such tables years ago) and LED wavelengths, just for starters. \$\endgroup\$
    – jonk
    Jan 16, 2022 at 19:39
  • \$\begingroup\$ But unless I were prepared to spend the time and effort necessary to validate your thoughts, I would instead follow through with a traditional approach that has survived many tests over time and pull forward some tests to evaluate the RF section questions you have as early as practical, hoping they aren't a significant issue. If it turns out they are as bad as you may think, then I'd come back to this idea and prepare and execute a plan for it. But not before the RF section test results were in. \$\endgroup\$
    – jonk
    Jan 16, 2022 at 19:44
  • \$\begingroup\$ Do you even need the transistor? The reference pin sources negligible current, so you could use a DAC output and a resistor in parallel with each Rref. As you raise or lower the DAC output you'll provide more/less current through the parallel resistor, which has the same effect as changing bthe series resistance. \$\endgroup\$ Jan 16, 2022 at 21:47
  • 1
    \$\begingroup\$ Trimming DC converters with a DAC or even digipot is a well established technique (you can find white papers on it), so "discouraged" is maybe too strong. You do have to look at what makes sense for your application though. PWM is much better for making large brightness adjustments, but as you point out it also has downsides. \$\endgroup\$ Jan 17, 2022 at 2:32

1 Answer 1


Yes, it can. With an adjustable current source.

Maxim Integrated makes current sink/source DACs that could do this with I2C control. You need up to 833uA for each driver for full scale; the DS4412 provides two channels at up to 2mA. Linky: https://datasheets.maximintegrated.com/en/ds/DS4412.pdf

If you use a current mirror technique you can control more than one TLC5498 with a single current-source DAC.

Here's how that looks (simulate it here):

enter image description here

You can also use NPNs in place of the FETs.

Could you do this with PWM chop? Sure, and I've done such a thing using drivers like this, dynamically updating the LEDs with PWM duty cycle control. It works, but I don't like PWM strobing when there is motion involved (LED taillights with PWM - argh!)


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.