FEDEVEL
Platform forum

, 04-29-2025, 06:58 PM

I am an amateur and don't have basic testing equipment. Could you analyze and suggest the possible reasons for the EMC test failure when using a 4-layer PCB (SIG/PWR+GND/GND/SIG), with a signal end impedance of 50 ohms and a differential impedance of 90 ohms, using the stackup JLC04161H-3313?

Sniper2 , 04-30-2025, 03:28 AM

I'll take a look

Sniper2 , 04-30-2025, 04:09 AM

I can't see any blatant problem

Sniper2 , 04-30-2025, 04:10 AM

I was expecting split gnd or some odd rmii rgmii that are routed in a odd way

Sniper2 , 04-30-2025, 04:11 AM

So the emi is probably frome somewhere in the esp32, not sure how it comes out tho maybe the vcc pin

Robert Feranec , 04-30-2025, 06:15 AM

interesting. how did you power up the board during this test? What power supply did you use?

QDrives , 04-30-2025, 01:50 PM

It is a broadband noise from 50 to 80Mhz.
Personally I would not have power on L2. Put it on L3 or route power.
Make sure that the ESP thermal pad has enough vias connecting it to L2 Gnd.
Add more stitching vias between the 2 Gnd planes.
Add Gnd return via in proximity to signal vias.
Place decoupling capacitors at the connectors for the power.

, 05-01-2025, 03:29 AM

@deefer5694 I powered the board via USB-C connected to a computer running the software for this board.

, 05-01-2025, 03:31 AM

Thank you for the suggestions. I’ll use them to improve the design. This was my first test, and I followed what I learned from textbooks about EMC. Do you think the result is very bad?

Robert Feranec , 05-01-2025, 12:21 PM

are you sure it's not coming out of the computer?

QDrives , 05-01-2025, 10:20 PM

The problem spike is at 60MHz, next high is about 72MHz, that is in 12MHz steps. 12MHz is USB2 full speed clock. Your USB cable is generating the spikes.
But USB does not cause the broadband noise. At least not the communication, it may be the antenna for the noise.
In the screenshot you notice all the spikes at 12MHz steps, starting at 36MHz, 48, 60, 72, 84, 96, 108 and 120MHz.
One way to test this is putting snap-on ferrite cores on the USB cable **at the PC end**. If the noise is gone (broadband only!), it is the PC and not your device.

Important question: was the USB cable mainly vertical?

Kalaslas , 05-07-2025, 08:18 AM

Did you test the PCB standalone, or was there other components & cables involved? Peaks below 100 MHz are often related to switching power supplies for example. Did you have an enclosure or any other metal structure involved during the test? It is sometimes worthwile calculating the wavelength involved and look for what measurements in the product that fits into that measurement, if there's a cable with a similar length, or a slit in a metal enclosure that fits.

Kalaslas , 05-07-2025, 08:23 AM

I agree with what @QDrives said, most likely the USB cable is the source of the emissions, I would also recommend testing this by adding snap-on ferrites to the cable, and if that works then add a suitable CM choke onto the USB signal lines on the PCB.

Also, your design doesn't look superdense in terms of routing, if possible I would suggest to move all signals to a single layer (either top or bottom), dedicate L2 & L3 to nothing but GND, and then dedicate the top or bottom layer (the one not used for signals) to power. Mixing power & GND on L2 is not really optimal.