Reducing LED PWM Noise: Which is the best option?

Question

Driving an LED with a microcontroller should be easy. But when looking into noise, things can get complex...

As an instantaneous voltage indicator used in a variable power supply (which will be used to simulate a photovoltaic array of a student's designed satellite), I am using some LEDs controlled by PWM (~31 KHz).

^{simulate this circuit – Schematic created using CircuitLab}

After the first revision of the PCB I have realized that each one of the LEDs is generating 200 mVpp in the 5V line:

My question: which is the best way to reduce this noise? Why?

From my (small) design experience I could come with the following possibilities. Which is the most effective, taking into account real-world components (ESR in capacitors etc)? Any other suggestion? A combination of several?

(Note: changing the PWM frequency is not a good option because that same signal drives other devices)

a) Decoupling Capacitor

^{simulate this circuit}

b) Low-pass filter

(The resistor forms an RC low pass filter with the gate capacitance of the MOSFET, removing the high frequency components of the switching)

^{simulate this circuit}

c) Snubber

^{simulate this circuit}

Answer 1

Your a) solution is a good one, just make sure the capacitor is as close as possible to the resistor/LED/transistor-branch as possible. Start with a 100nF value and see how that works. If ripple is still too large to your liking, add a electrolytic capacitor in parallel to the 100nF. The 100nF will suppress the higher frequency components and the electrolytic capacitor will do better for lower frequency components.

As @pjc50 says in one of the comments, a gate series resistor as shown in b) is good practice too to avoid ringing. I'd personally pick a lower value, say 100Ω. It will suppress ringing and will avoid the transistor spending too much time in linear mode (=dissipating heat).

Answer 2

a) is close to my first choice. With a suitable value of capacitor in place, the current through the trace resistance and inductance should be nearly constant, very little ripple.

However, you'll still have switching current through Rlim. Also, the capacitor must be sized as a lowpass with corner frequency below 30kHz, using Rtrace and Ltrace.

If you put the bypass capacitor after Rlim, then the current through Rlim is also mostly constant, and the bypass capacitor can be smaller and/or have a cutoff frequency much lower without sizing it excessively large.

b) The 2N7000 has an input capacitance of 60pF, according to the datasheet. The 1/RC frequency is about 1.7MHz. The PCB traces will still be radiating RF.

c) This looks like a snubber circuit meant to save switch contacts from inductive kickback. Not really what you need, and it will eat some power and possibly increase radiated RF noise as it will be charging/discharging the capacitor on each cycle.

So my vote is for a modified a) with the bypass capacitor from the anode of the LED to the source of the NMOS. Making the leads as short as possible.

Edit: a comment makes a good point about the capacitor current spike on the first PWM pulse, so I will add this to my answer:

I must admit I was thinking of "steady state" with the PWM already running, in which case the current and voltage will be constant. But you are correct, when it first turns on, that capacitor will be dumped through the LED and NMOS. I suppose it is a trade off. I was attempting to keep switching transients on as short and few wires and components as possible. If you can find a value of C and an LED that together allow those initial current spikes without damage, good. If not, then I supposed I'd go with b) and make the interconnections as short as possible.