1

I am using this connector which has the internal magnetics. I am connecting it to an Ethernet PHY device.

In the datasheet, we can see, that it has internal capacitors of 100nF.

I went through this connector (CIRCUIT E) which has internal capacitors of 10nF.

Can someone tell me what would be the impact of the 10nF and 100nF capacitors?

What would be affected if we increase or decrease the value of these capacitors?

JRE
  • 67,678
  • 8
  • 104
  • 179
  • 2
    The capacitor just helps filter noise to ground and reduce EMI susceptibility. I doubt it makes any difference whether it's 10 or 100nF, both seem widely used. – Finbarr Jun 07 '22 at 10:13
  • 1
    Pulse knows a thing or two about magjacks as they've made a few million of them. Besides, if you choose that magjack, you don't have the choice of varying that capacitance. If the PHY chip manufacturer recommends that part, then it has been verified to work. – Kartman Jun 07 '22 at 12:43

1 Answers1

0

The capacitors in question are decouplers for the centre-taps of the ethernet transformers and are not critical in value. Similarly, many designers use 100 nF decouplers on ICs on a PCB and, there are many other designers who use 10 nF decouplers for each chip. The centre-taps do need to be decoupled though: -

enter image description here

Image from here.

What would be affected if we increase or decrease the value of these capacitors?

It's a bit of a goldilocks story really. Too high in value and the capacitors self resonant frequency will turn them into net-inductance at frequencies coincident with data frequencies and this could reduce data integrity. Too low in value and they are ineffective at centre-tap decoupling at data rate frequencies. Here's an example of different capacitors vs their self-resonant frequency: -

enter image description here

Image from here - useful reading.

So, roughly speaking, you want a capacitor that has lowest impedance in the 10 MHz to 100 MHz range for normal ethernet data.

Andy aka
  • 434,556
  • 28
  • 351
  • 777
  • Thank you for the answer. So, any capacitor between 10nF and 100nF will be useful? Am I correct? And how to identify the impedance value that will affect the ethernet data? –  Jun 07 '22 at 10:56
  • It's a tricky one to identify. I guess we all sit on the shoulders of others who have gone before and burned their fingers. If cost isn't an issue, I'd probably put a 100 nF capacitor in parallel with a 1 nF capacitor to get a fuller coverage of the spectrum (belt and braces). But, choose capacitors that are suitable. Kemet's website will give you impedance graphs for many of their capacitors and so will Murata. – Andy aka Jun 07 '22 at 11:04
  • Just to refine the point on resonance: it's measured at the cap body, and is lower with connecting traces included. Further, there's leakage inductance in the transformers themselves, that far outstrips the capacitor's ~2nH. Resonance is a red herring; getting into the inductive region basically means you've won, the impedance isn't getting any lower than this. And most of the time, that impedance is low enough, regardless of its phase, to do the job. So the ESL isn't the enemy. (And stuff like, caps don't need to be staggered or anything.) – Tim Williams Jun 07 '22 at 13:49