Why does PWM create lots of EMI and signal ground noise?

Question

TI had a note to create DAC signals using PWM. http://www.ti.com/lit/ug/tidu027/tidu027.pdf

Whereas I also found an article that says

http://www.mosaic-industries.com/embedded-systems/instrumentation/pulse-width-modulation-pwm-controller-driver-actuator/noise-reduction

"High current, high frequency PWM signals are notorious for creating and radiating electromagnetic interference (EMI and RFI).

Consequently, if you use PWM currents for controlling power devices, you should make your connections in a way that minimizes or limits the radiated noise.

In general, single point grounding and using shielded twisted pair cable are essential to prevent EMI or mitigate PWM noise."

• Why does high current / high frequency PWM signals can create EMIs and signal ground noise? It looks like this is related to electromagnetics, but I cannot clearly explain the reason.

• Does this also mean that PWM based buck converter, like the LM3671/-Q1 2-MHz, 600-mA Step-Down DC-DC Converter, can create EMI to nearby circuits? It makes me feel like I have to populate the Bluetooth/Wi-Fi microcontroller far away from this buck converter.

In the LM3671 datasheet, it says

http://www.ti.com/lit/ds/symlink/lm3671.pdf

• 2-MHz PWM Fixed Switching Frequency (Typical)
• Automatic PFM-PWM Mode Switching

Answer 1

Here is one example

If you combine Biot_Savart and Faraday Law of Induction, you will get this math to predict voltage induced, from a long straight wire that is in the plane of a rectangular loop

Vinduce = [MUo * MUr * Loop_Area / (2 * pi * Distance_wire_loop)] * dI/dT

where this is fully accurate, without need to write and evaluate a 1/R integration, for small areas or long distances. Given the 1/R, integrated, becomes something like

ln(longer distance to loop) - ln(shorter distance to loop)

I just ignore the "ln", and insert MU0=4*pi*e-7 and MUr=1, to have

Vinduce = 2e-7 * Area/Distance * dI/dT

and remember I've dropped the LN math so people won't balk at using this APPROXIMATION.

Here is my favorite EMI example: 10,000 horsepower train speed controller, with 2000 amp bus about 40mm from the IGBT-trigger PCB. The customer would not give details on the switching speed of the PWM pulses, but were comfortable with my estimate (for thermal survival of the IGBTs) of 1uS. Given the power was 6000 volts and 2000 amps (max), or 12 Million Watts, the IGBTs only survive if very quickly switched.

Given the PCB had various ICs being slowly destroyed because of (closer) 40mm distance, and the new PCB layout, I assumed a vulnerability area of 0.1meter by 0.1meter in the Ground (there was no single Ground plane layer).

The math becomes:

Vinduce = [2e-7 * (0.1meter * 0.1meter) / 40mm ] * 2,000 amps / 1,000 nanoSec

Vinduce = 2e-7 * 0.01/0.04 * 2e+9 amps/second

Vinduce = 0.5 * e-7 * e+9 = 0.5 * 100 = 50 volts [error: forgot the 2nd 2]

Correct

Vinduce = 2 * 1/4 * 2 * e-7 * e+9 = 100 volts

Thus this 10,000 horsepower speed controller was (attempting to) induce 100 volts on the GROUND.

Massive eddy currents. And GNDA is not the same as GNDB or GNDC or GNDD.

In the phonecall, I asked the customer's engineer to make a 1 square inch loop, at end of Coax cable, and go measure the induced voltages. About 2 weeks later, a fine PDF appeared. He was measuring 1.5 volts in that 1inch^2. Does this correlate with the prediction of 100 volts? Check, by scaling down the 100volts (4" by 4" assumed loop area), thus 100/16 = 6 volts. And we don't know if they were running at 2,000 amps (nominal is 1,000 amps). Given the various unknown params: (the PCB layout), the PCB eddy currents, the unknown positioning of the 1inch probe, and actual buss current being switched, I felt we have rather good correlation between theory and concepts.

I asked the user what had changed (they'd have no prior problems) and turned out the (large multinational) had hired a programmer to update the code, and the company had directed the "coder" to modify the PCB and move the PCB closer to the 2,000 amp bus.

From what I recall ------ a fine example of Biot_Savart and Faraday, and a "coder" working well past his level of understanding.

Answer 2

I'll spare you the math, but a truly square wave of 50% duty cycle at frequency f1, is effectively made up of the sum of sine-waves starting with the fundamental at frequency f1, plus the odd harmonics at fractional amplitudes:

1.f1 + 1/3(3.f1) + 1/5(5.f1) + 1/7(7.f1) +...

Other (non-50%) duty cycles will introduce other even harmonic frequency multiples, and at different fractional amplitudes. More detail of the math here and here.

So that square wave of, say, 1 MHz in a switch-mode regulator isn't just 1 MHz, but also 3 Mhz, 5Mhz, 7Mhz, etc... albeit at successively lower amplitudes at each higher frequency.

But what happens to a current through a conductor as frequency goes up? It's more and more likely to radiate off the wire, as "radio", EM waves.

So this is why "PWM creates a lot of RF noise". There are various techniques to mitigate this reality, but those are answers for other questions :-)

As for "ground noise" - "ground" isn't magic; it's just the other half of the circuit, with the same characteristics of resistance and impedance. What you do on the "positive" happens equally on the "negative/ground" return conductor, even if that's a 'nice thick ground-plane' on a PCB. That's ground noise and "ground bounce"; your ground is developing a voltage across it because of the same current running through it as for the 'positive' side of your circuits. It's just the nature of simple switch-mode power supplies (as distinct from some switch-mode topologies that don't slam current on & off like a dunny door on a windy day) that tends to involve high amplitude short duration transient currents, due to the PWM switching nature.

Answer 3

Any flow of electrons will generate a magnetic field. That field can induce currents into near by conductors. Which is called EMI.

PWM (Pulse Width Modulation) sends pulses of current down the conductor. This means it's a potentially very 'noisy' signal. Why? Because it can induce multiple frequencies into near by conductors as the pulse changes. If you look at any of the graphs in a PWM devices datasheet you can see the signal is noisy.

The higher the power, the larger the magnetic field can be and therefore affect things further away.

A buck converter will certainly generate noise. The LM3671 is not high current, but it's still very noisy. If you read the datasheet, it actually tells you how you should place and connect it to reduce EMI. There is even an example of a PCB layout.

But yes, you should physically separate noise sensitive things from noisy things. Shielding also exists. You can get shielded versions of some components.