Increase PWM frequency from 5 Hz to 100 kHz to suppress EMI crosstalk


I use a 24 V, 5 Hz PWM voltage source for driving a 50 Ohm Pt heater integrated with a sensor that has an RC filter with a 1.5 Hz cut-off frequency.

The PWM current on the heater is giving a strong 5 Hz EMI to the integrated sensor (this EMI is reduced if the heater and the sensor are far apart in space). I am trying to increase the PWM frequency to 100 kHz such that the EMI can be filtered out significantly when the heater and sensor are near to each other.

On the board which is operated using an NI DAQ card, we use a 24V input, and right after the 24 V input, is a MOSFET used as the switch, and that is directly controlled using an NI DAQ card (Labview) where 5 Hz is the maximum frequency that it can be assigned to. Given the limited frequency (5 Hz) the NI DAQ card can supply, I wonder if there is another way to supply the high-frequency PWM (100 kHz) to the MOSFET.

And accordingly, Do I need to step down the voltage of 24V to a lower voltage as the PWM frequency is changed vastly from 5 to 100k? Or maybe as long as the duty cycle is the same, I can still maintain the 24 V input.

Please advise!
Thank you, Xiaheng

How do the source and received signals compare?

If the interference signal bears direct resemblance to the source, the problem is most likely related to conduction phenomenon of some sort, such as the heater current passing through a resistance that is involved in the sensor measurement, or causing a bias or reference voltage to shift, or something like that. 5Hz is simply too low a frequency to be well-coupled through the electrostatic mechanism that would be typical in such a case.

If the interference signal resembles the edges of the source signal however, this would be characteristic of capacitively-coupled interference; depending on the rise time, the edges of a 5Hz digital PWM signal can have spectral content well into the MHz range, which can pass through a parasitic capacitance quite easily.

If those high-frequency edges are already passing through your 1.5Hz filter however, making more such edges isn’t going to improve the situation.

I would suggest that further diagnosis of the problem is in order, in order to avoid pursuing a solution that actually makes the problem worse.

Hi Rick,

Thanks for checking on the problem first!

I do check that EMI is the issue before I ask this specific question -

See the experimental data below -

I caged the heater with a copper mesh (Faraday cage) that can shield itself from the sensor. This cage will only be effective if it is grounded properly.

In the figure, for the first 25 s, there is no PWM current to the heater, after 25 s, the PWM starts - and we see the associated PWM noise, remember at this point, the cage is not grounded, therefore the EMI is not shielded. Then after 60 sec, while the PWM never tunrs off, but I grounded the copper cage, and the PWN noise is gone. This clearly tells me that, the PWM generated EMI is the major noise source, but not something from the source directly.

Please comment : )


Some observations:

  • There appear to be changes in DC offset occurring around T=25 and the levels at T=10 and T=70. Is there an explanation for this?
  • The ripple present from T=25 to 45 has a frequency inconsistent with the 5Hz given as the fundamental for the PWM signal, indicating that it is not the direct cause.
  • Information regarding the duty cycle of the PWM waveform with respect to time is not shown. Plotting this alongside the sensor output may prove informative.
  • Two parallel square plates 10cm on a side separated by 1mm of air result in a capacitance of approximately 1pF, with smaller area or greater distance reducing the capacitance proportionately. For this to be the coupling mechanism, the transferred signals must be explainable using modeled components on this scale. Given the long time scales shown, an extremely high impedance of the vulnerable node/net in the receptive circuit would seem necessary.

My suspicion is that the PWM duty cycle applied to the heater will be found to oscillate between high and low values in correlation with the ripple between T=25 and 45, suggesting that the coupling phenomenon is temperature-mediated in some manner.

Further information on the apparatus in question would be useful.

1 Like

I am not sure at this point. My guess is that the PWM-generated EMI generates some static current in the readout wire before it goes into the amplifier. My sensor’s baseline current is in the range of nA so it is very small, any small static charge might be equivalent in this range. My readout wires are not co-axial type, but just normal wires.

Good catch. The reason is that the data sampling rate is 4 Hz here. If I increase the data sampling rate to 100 Hz, below reflects exactly 5 Hz from the PWM - see the data below -

The duty cycle here is likely to be 20% but I don’t remember exactly. I have tried to see the dependence of the duty cycle to the noise, the higher the duty cycle, the larger the amplitude of the 5 Hz noise waveform.

I am sure it is not really a temperature effect - as long as the PWM starts, the noise kicks in, the sensor and the heater is still just RT, say at T = 27 s, 2 s after T = 25 s when PWM starts. And when it is grounded, at T = 40 s, the temperature on chip is alrady high (40 C), there is no noise, because the heater itself is shielded. Please pay attention to the cage effect here.

Here is a photo of the setup. You can ignore the small transformer and a lamp sitting on top of my sensor. It is unrelated, I already tested with them off, and the same effect persists.

I will wait for another round of responses after you see these facts. As you might revise your other arguments here a bit : ) Thanks


Ah. Under-sampled data does indeed paint a different-looking picture.

That would suggest your sensor’s source impedance is approaching GΩ territory, in which case interference from electrostatic coupling @ 5 Hz is a lot more plausible. Things like surface leakage, fly flatulence, and salacious thoughts are also potential noise sources in that sort of context.

Since shielding your heater solved the issue, that’s probably a good solution. It seems to have been demonstrated that the heater signal edges will couple into your signal otherwise without it in spite of whatever filtering is present, so a kHz PWM signal is probably the last thing you’d want to be introducing to the environment.

If shielding is unworkable for some reason, linear regulation of your heater may be an alternative. A circuit like the below could be a useful tool for that.

Hi Rick,

Thanks for the valuable responses!

Exactly, I have a 24V bias (this is the same sensor that I am asking questions about in the other post). May I ask how to correlate the impedance with the electrostatic interference? In other words, do you think that increasing the 5 Hz to 100 kHz can isolate it from the Gohm sensor source?

Yes. That will do for the apparatus I show you where the heater and the sensor are essentially separate from each other so I can wrap a copper mesh toe shield the heater. This is just the easier case that I am showing you to facilitate our discussion. In reality, the heater is monolithically integrated with the sensor on a single silicon substrate, which means there is no way I can wrap a copper mesh to shield the heater like what I am showing here.

So you might notice that my readout wires are just normal wires leading to the amplifier, I am wondering do you think if I would replace it with a coaxial wire where I can ground the metal layer inside the coaxial wire for the readout will help or not? The underlying question here is - do you think it is possible that this noise might be simply due to the unshielded wire that I am using (the surface leakage static charge can pin onto), or it is due to the capacitive sensor with the Gohm impedance (this will mean that even if I change to the coaxial wire, the noise will be intrinsically there.)

I will study the linear regulation a bit before I can ask some questions relatable.

No; current flow through a capacitance increases with signal frequency. If this heating element has an unshieldable capacitive coupling with a sensitive node in your sensor apparatus, increasing the signal frequency will only increase the amount of noise injected into the system.

If you’re not already familiar, I would suggest some quick reading on the concept of a parallel plate capacitor. While unshielded lead wires could create an opportunity for noise pick-up and coaxial connections might be helpful, a conductor integrated onto the same silicon as your sensor seems more likely to be the cause of the problem.

No, it has an oxide layer to do the insulation between the heater and the sensor.

Can you redirect me to some material regarding this and how this is helping the scenario here? Thanks!

Are you suggesting if the sensor is not a capacitor, then this strategy should work - isolating the heater frequency with the sensor’s sampling rate and using an RC filter?

By “integrated onto the same silicon” I mean combined into a monolithic whole, not implying a direct connection to the semiconducting material.

It appears that this heater has a capacitive coupling to a sensitive node in your apparatus; I would suggest looking up the formula for a parallel plate capacitor, and estimating the value of this coupling effect based on your sensor’s geometry and the dielectric constant of the insulating material.

To the extent this capacitive coupling exists, signals applied to the heating element will result in current being injected into your measurements, as C*dV/dt. It seems quite possible that these currents could outweigh the signals you’d be trying to detect.

Linear regulation would mean varying the power applied to the heating element in analog fashion rather than using PWM techniques, in order to minimize the dV/dt term above and reduce the amount of noise added to the system.

Quite to the contrary; I’m suggesting that if your sensor is like a microphone used to capture the sound of a butterfly’s wings, this PWM signal may be similar to a jackhammer. To shut the jackhammer off entirely seems a more likely solution than simply running it faster.

That said, it’s not obvious to me what the purpose of this heating element would be, given that we were recently discussing cooling methods. I’m missing quite a bit of information regarding your apparatus and experiment, and as such much of the advice I’m offering is based on inference. Your organization has an electrical engineering program if I’m not mistaken; I’d suggest that seeking collaboration from that department may allow you greater freedom of disclosure and more rapid progress toward your goal, as compared to discussion via a public forum.

1 Like

Hi Rick,

Thanks for the reply! I do reach out for EE folks to help. The increased PWM is what they suggested. I actually feel that this forum has more veterans that can give me a different perspective, which I do appreciate!

These all two completely different topic. I do try to ask a good question on the forum. Will try better next time!