3
\$\begingroup\$

I am currently developing my first product which needs to fulfill EU EMC requirements.

Below you can see the first result out of the EMC lab (radiated emission test):

EMC test - failed

One can clearly see the 16 MHz clock frequency and the radiation out of the product seem to go over the power lines through multiple connectors from the main PCB.

This assumption also comes from the following results of a measurement without having the connectors attached:

EMC test - passed without wires

So, now I am designing an update and beside optimizing the layout regarding the crystal, uC decoupling, trace length reductions, and provisions for ferrite beads on the signal lines, but I am not really sure about one countermeasure - an LC filter.

Here I would be very thankful to get some help.

Questions:

  1. I want to add an LC filter on every power line right before the connector. But here I have the first issue - which cut-off frequency needs to be used? Is it the frequency of the first peak (32 MHz) or, as I read a lot, 1/10th of the first peak (3.2 MHz)?

  2. Filter: I have a connector which has 1 x 12 V power and 3x switched low-side power (PWM/MOSFET). Does it make any sense to add filters to these lines, or is it just the power line which makes sense to filter?

\$\endgroup\$
20
  • 1
    \$\begingroup\$ You need to do one of two things: either 1) analyze where the noise is coming from in your circuit and address each cause individually, or 2) slap ferrite beads all over the place in the hope that it'll clean up the noise enough. Using a lossy ferrite bead instead of a standard inductor for an LC filter should give you somewhat better performance than just the inductor. \$\endgroup\$
    – Hearth
    Commented Sep 30, 2022 at 19:59
  • 3
    \$\begingroup\$ I'm sure you don't have a full EMC test lab at your disposal, but figuring out where noise is coming from can be a lot easier if you have an E-field or H-field probe. Even better if you have both. \$\endgroup\$
    – Hearth
    Commented Sep 30, 2022 at 20:00
  • \$\begingroup\$ Thank you for the fast reply. Actually I was in full EMC lab and we are 99% sure that it's the 16MHz crystal on power which is creating the noise. So ideally with better decoupling of the uC (currently only one 100nF cap at each VDD/VSS pin pair) we will already get rid of the issue. But I want to do a "just in case this is not enough" thing. Ferrit beads are also considered to be taken instead/in addition to the LC filter, but in this question it's really about the LC filter I want to at least provide the footprints for, even if I do not populate them and first trying with a ferrit bead. \$\endgroup\$
    – Daniel
    Commented Sep 30, 2022 at 20:18
  • 6
    \$\begingroup\$ Is the 16MHz noise from the clock, or from logic switching at that frequency? Because a clock is pretty high-impedance... \$\endgroup\$
    – rdtsc
    Commented Sep 30, 2022 at 21:05
  • 2
    \$\begingroup\$ @Daniel Yes, you were in an EMC lab for that testing, but using an EMC lab is incredibly expensive. I doubt you have the ability to just make use of one whenever you want. E-field and H-field probes are types of probe you can get for any ordinary oscilloscope--ones good for proper quantitative measurements will run you a few thousand dollars, but you can make ones good for qualitative measurements from coax and tape. That'll be helpful to narrow down where the noise is coming from, even if you can't properly measure the noise with them. \$\endgroup\$
    – Hearth
    Commented Sep 30, 2022 at 21:50

2 Answers 2

2
\$\begingroup\$

Question 1 - If I'm interpreting your test results correctly, you are missing your spec by approximately 18 dB at (I think) 32 MHz. That means you need to attenuate that frequency by at least 18 dB. Hence the filter needs to be designed for a cutoff frequency of much less than 32 MHz, depending upon the order of the filter.

The peaks at 140 MHz and around 200 MHz are not directly related to the 16 MHz crystal oscillator. The spectrum of the oscillator should roll off with increasing frequency. So those frequencies are indicative of other parts of the circuit that are switching at ~140 MHz and ~200 MHz.

I know this solution probably isn't in your solution space, but this is one of the reasons we routinely shield cables, especially signal cables. Input power filtering is usually taken care of with appropriate common mode filters.

\$\endgroup\$
5
  • \$\begingroup\$ Hey SteveSh, thank you for your answer, your point "Hence the filter needs to be designed for a cutoff frequency of much less than 32 MHz, depending upon the order of the filter." goes exactly in the direction I am asking for - I am currently thinking of a filter 2nd order. What would be the correct target frequency in this case and beside telling me if you know I of course would be interested to know how I can calculate this or if it's simple experience. A lot of times I read the cut off frequency should be 1/10th of the frequency you want to filter but without explanation. Thanks in advance! \$\endgroup\$
    – Daniel
    Commented Oct 1, 2022 at 17:59
  • \$\begingroup\$ @Daniel a filter cut-off point is defined to be the frequency it attenuates 3dB. So obviously if you want to attenuate something more than 3dB, you want the cut-off to happen earlier. The filter order defines the slope, which defines where to put the cut-off to get the requires attenuation at the required frequency. \$\endgroup\$
    – Justme
    Commented Oct 1, 2022 at 21:04
  • \$\begingroup\$ Hello Just me, thank you for your answer. I think I got it now. I found a cool tool which gives you the attenuation at a specific frequency for a defined filter: falstad.com/afilter \$\endgroup\$
    – Daniel
    Commented Oct 1, 2022 at 21:32
  • 1
    \$\begingroup\$ But now I have a following question...for what do I have to watch out in regards to too low cut off frequency? Because my first (probably incorrect) thought now is "ok, I just go extremely low with the cut off frequency, like 100kHz, and then I simply eliminate all spikes in the MHz range" \$\endgroup\$
    – Daniel
    Commented Oct 1, 2022 at 21:56
  • \$\begingroup\$ As information, the main power supply I am using has an fosc of 110kHz...is this maybe something to consider like staying above this f buy then as low as possible? \$\endgroup\$
    – Daniel
    Commented Oct 1, 2022 at 22:04
0
\$\begingroup\$

I want to add an LC filter on every power line right before the connector. But here I have the first issue - which cutoff frequency needs to be used? Is it the frequency of the first peak (32 MHz) or, as I read a lot, 1/10th of the first peak (3.2 MHz).

Do something easier first and add a ceramic 0.1uf cap between the terminals and ground of the connectors while you are testing. If you use wire, use a high gauge with lower inductance (not 30awg), the higher the better. Usually it's a good idea also in future designs to have connectors with the option for shielding in the case of radiating cables. Make sure all cables are twisted to minimize radiation.

Filter: I have a connector which has 1x 12 V power and 3x switched GND (PWM/MOSFET), does it make any sense to add filters to the GND lines or is it just the power line which makes sense to filter?

Have two things you can do. either make the impedance high enough in or before the cable the high frequency signals will not travel down the cable. Another option is to short the high frequency signals to ground with capacitance. One problem with this is you need to make sure that your capacitance does not have parasitic inductance (parasitic inductance arises from any copper traces, vias or even the leads of capacitors, which creates small amounts of series inductance that can block high frequency signals, these effects must be accounted for when creating a shunt to ground, at lol at the impedance graph of the shunt capacitor)

\$\endgroup\$

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.