I have many IR leds to drive, with MOSFETs and a single constant 60 mA low side current regulator. The leds are PWM at 1 MHz (50% duty cycle), and the farthest leds are 60 cm away from the regulator.

I am not very experienced with EMI, but from what I read:

  • I need to "round" the PWM signal to avoid high resonant frequencies
  • I need to keep the loop area as small as possible
  • I need to keep the return traces to ground as small as possible

It is not a problem to route the return trace of each led just beneath the power traces on the adjacent PCB layer.

My main concern is about the very long return traces (60cm) which are required to reach the low-side current regulator.

At 1 MHz / 60 mA, do you think it would cause EMI issues, should I use multiple current regulators at regular intervals ? Would you have any suggestion to reduce EMI (besides lowering the frequency)

Generally speeking, provided that the topology/layout allows it, what are the benefits of a ground plane for handling return path, over dedicated return traces (in terms of EMI) ?

  • 1
    \$\begingroup\$ 1MHz is a very high-speed pwm for LEDs - in fact, about 10000 times faster than it needs to be for the human eye. Is this really the pwm you use to drive the LEDs, or rather the frequency of a switch mode supply's pwm? The latter would make more sense, because then you gain something by using a high speed pwm. Otherwise, all you get is unnecessary high switching losses and EMI. \$\endgroup\$
    – mmmm
    Commented May 19, 2021 at 0:27
  • \$\begingroup\$ I am sorry, I should have said that these are IR LEDs, not aiming the human eye, but IR receiver, I can not lower this frequency. \$\endgroup\$ Commented May 19, 2021 at 0:30
  • 1
    \$\begingroup\$ Ah but then your current over time wave form is mostly defined by what your receiver expects, and deviation from that reduces your communication robustness. Don't think you'll find out you have a lot freedom -can't just change thee roundness of the wave form if the sharp edges are necessary for receiver clock recovery, for example. \$\endgroup\$
    – mmmm
    Commented May 19, 2021 at 0:48
  • \$\begingroup\$ Hopefully you have circuits to sweep out the carriers since the LED probably isn't going to be able to turn off fast enough if you don't. But 60cm worth of trace inductance is not going to help with that. Can you have the drivers closer to the LED and just send the signal 60cm to the driver instead? \$\endgroup\$
    – DKNguyen
    Commented May 19, 2021 at 0:49
  • \$\begingroup\$ @DKNguyen I am not native, so I may misunderstand your point, but IR LEDs rise/fall time is 15 ns. Receiver uses phase detection to demodulate the signal. 60 cm is already the shortest distance, but I can reduce it using more drivers. However I have computed trace inductance using online calculators and SPICE simulated it, and it did not have noticeable effect on the operating frequency. \$\endgroup\$ Commented May 19, 2021 at 1:18

2 Answers 2


1 MHz is fairly tame, not much above audio from the perspective of EMI. For example it is well below the self-resonance of most discrete components, and its quarter-wavelength on a PCB trace is around 4.5 m.

My instinct is to slug the FETs to limit their output bandwidth to around 5-10 MHz. That looks a little lower than the LEDs' maximum bandwidth, which will also be a kindness to them. Decoupling capacitors (usually ceramics in my day) should do the job.

Otherwise, provided you run the traces sensibly, I can't see EMI being a significant issue. There is little difference in principle between using a separate return line or the ground plane, just make sure it chimes in with the FET supply-and-return arrangement. Run the return on the same level as the signal, or one away (whether trace or ground plane); if you run it two levels down there is the risk of an unrelated track passing through the loop and creating a nasty little coupling transformer.


I think Guy Inchbald has it covered well. I had one thought in addition:

If you run two traces to each LED (anode + cathode) to make a tight supply-return-pair, you will have almost no loop area and be safe from differential mode emission and interference.

However, as the LED has no other potential reference other than those two traces, the trace pair is a pretty long antenna for common mode noise.

Two ideas to counter this:

  • common mode chokes, although it sounds a bit like overkill
  • add a cap of 100pF to the ground plane near the LED ( on the side which is anyway at ~gnd potential). This is just to anchor the RF potential,not for a lot of actual return currents.

Or you could just run the return currents through your ground plane. This might cause slight common impedance noise in some precision circuits though. From EMI point of view, using the plane for return is good though.

  • \$\begingroup\$ Assuming the FET is adequately treated, common-mode noise emissions will be negligible. Susceptibility might be an issue in a noisy environment, e.g. affecting pulse widths, but that assumes the rest of the system has big problems. Either way, treat the source not the path. Run the return on the same level as the signal, or one away (whether trace or ground plane); if you run it two levels down there is the risk of an unrelated track passing through the loop and creating a nasty little coupling transformer. \$\endgroup\$ Commented May 30, 2021 at 9:55

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.