5

I have a crystal associated with an Ethernet-to-PCIe bridge operating at 25 MHz.

I observe spikes at 62.5 MHz and 312.5 MHz. Those harmonics are 2.5 times and 12.5 times the crystal fundamental frequency.

My questions:

  1. Can the harmonics be 2.5 times or 1.5 times the fundamental or should it be only 2 or 1 times the fundamental? Can it be half or can it be a whole number only?
  2. Any ideas or good practices that can be followed to reduce the spikes at those harmonics?
JYelton
  • 32,302
  • 33
  • 134
  • 249

2 Answers2

6

Can the harmonics be 2.5 times or 1.5 times the fundamental or should it be only 2 or 1 times the fundamental? Like, can it be half or can it be a whole number only?

The EMI won't be related directly to the crystal frequency but, to something that is used on a bus or a data line that is at a fraction of 25 MHz. Crystals themselves won't directly produce significant levels of EMI so it has to be a circuit that divides the crystal frequency by 2 (for instance).

  • Half of 25 MHz is 12.5 MHz and, 62.5 MHz is the fifth harmonic.

  • 312.5 MHz is the 25th harmonic of 12.5 MHz

Try looking for other harmonics in between that may be a lower level and you have not noticed. Anyway, the EMI will be related to sub-harmonics of 25 MHz and almost certainly not due to the basic crystal frequency.

Any ideas or good practices that can be followed to reduce the spikes at those harmonics

Use a spectrum analyser and a near-field sniffer probe to establish where the interference is primarily coming from on your PCB. It can take a while to establish this so do it carefully. Look at data lines and busses.

Near-field sniffer probe: -

enter image description here

Image from here.

DIY probes: -

enter image description here

Andy aka
  • 434,556
  • 28
  • 351
  • 777
  • Thank you for your answer. There's a PCIe reference clock with PCIe signals. And also, on the ethernet copper lines there are common mode chokes with magnetics. The PCIe reference clock is 100MHz input to the bridge ic –  Jan 11 '23 at 17:14
  • It could be there or. it could be somewhere else. – Andy aka Jan 11 '23 at 17:16
  • Thank you. Do you think the common mode choke or the ethernet magnetics might contribute to it? –  Jan 11 '23 at 17:17
  • In my experience, the noise can come from several places or one place and, it can come from places you didn't expect it to. When I have found those unexpected places in the past, I have cursed myself for not recognizing it and wasting money in labs. – Andy aka Jan 11 '23 at 17:19
  • 1
    Thank you for the answer @Andy aka. Much appreciate your help :) –  Jan 11 '23 at 17:19
2

Assuming you have standard 100Base-TX Ethernet, it has a symbol rate of 125 Mbaud and bandwidth of 31.25 MHz.

The measurements indicate the 62.5 MHz spike is at exactly half the symbol rate and double the bandwidth frequency.

And the second 312.5 MHz spike is 5th harmonic of 62.5 MHz.

And like said above, 62.5 MHz is also 5th harmonic of half the crystal.

So somehow the Ethernet chip manages to take some sub-division and multiple of the 25 MHz and radiate them out.

Justme
  • 127,425
  • 3
  • 97
  • 261
  • Thank you for your answer. Let me check and come back –  Jan 11 '23 at 18:11
  • Could you please just expand a bit more on, " Ethernet chip manages to take some sub-division and multiple of the 25 MHz and radiate them out." , please? –  Jan 11 '23 at 18:14
  • So, to conclude, you suspect, it is due to 12.5MHz fundamental (due to crystal) or due to the 100Base-TX Ethernet datalines? –  Jan 11 '23 at 18:22
  • It can be anything, but somehow they all point to the Ethernet design. Bad system design, bad schematic design, bad PCB layout design, bad selection of componets, missing, damaged or malfunctioning components, PCB manufacturing error. You have not given any other info than 1) It's Ethernet 2) There's a 25 MHz crystal 3) Two largest harmonic frequencies. That is not enough to start debugging a design. – Justme Jan 11 '23 at 18:47