There was a question I raised earlier over here where an app note suggested to have no ground plane underneath the area between the ethernet connector and magnetics.

So, in that case, how will the differential impedance of 100ohms be achieved between the pair?

Am I routing the tracks from RJ45 to magnetic as a differential pair but there will be no impedance control as I thought the distance between tracks and plane played a part in this? So in effect im just routing two tracks for each pair with no special constraints?

Also are the ethernet pairs meant to be length matched? This appnote says, it has to be length matched. Any expections?

  • \$\begingroup\$ @TimWilliams, no. sorry. could you explain how the differential impedance can be achieved if no ground plane underneath the signals \$\endgroup\$
    – Freshman
    Sep 28 at 14:07
  • \$\begingroup\$ There are better, and more interesting, questions you could be asking. Since you have provided context (it's Ethernet), a much simpler answer suffices. You would have a much better time asking questions that develop your own (and others') understanding, of transmission lines, electromagnetism, network theory, signal quality, etc. Why not ask why short lines are acceptable? You could drill into the reasons for the above answer. You could even ask why appnotes contradict each other (answer: they suck), or what they're really saying, and epistemic and didactic topics like that. \$\endgroup\$ Sep 28 at 14:14
  • \$\begingroup\$ (Well, not really so much the latter. That would be more of a teaching/learning or philosophy Stack. But with regards to locating and parsing information within this specific domain, might still prove interesting.) \$\endgroup\$ Sep 28 at 14:15
  • \$\begingroup\$ I thought this was already mentioned in your previous question. If you think of an unshielded twisted pair, such as Ethernet cable, it does not have a ground anywhere near it, but it still has 100 ohm impedance. You still seem to insist there must be a ground plane. And, this has nothing that is specific to Ethernet. \$\endgroup\$
    – Justme
    Sep 28 at 16:17

3 Answers 3


Differential impedance can be achieved by referencing two lines to a common ground plane (so-called uncoupled lines) by referencing them to one another and a ground plane (so-called loosely coupled lines) or by coupling only between themselves (so-called tightly coupled lines). Eric Bogatin provides a nice diagram showing electric field lines for 3 different ways that a 100 ohm differential impedance can be achieved:

enter image description here source: https://www.signalintegrityjournal.com/blogs/4-eric-bogatin-signal-integrity-journal-technical-editor/post/402-pop-quiz-use-tight-or-loosely-coupled-differential-pairs-to-reduce-cross-talk

On the left is the tightly coupled case, where the lines are close to each other relative to both their width and their height above the ground plane. In this case, nearly all of the field lines go between the two lines of the differential pair. On the right is uncoupled case, where the lines are moved 3 times their width apart and then moved much closer to a ground plane. In this case, nearly all of the field lines go to the ground plane. In the middle is the intermediate case where the lines are coupled to one another and the ground.

As a designer, you get to pick how much you couple the lines. As you move them closer together more of the impedance is determined by their mutual coupling, as you move them further apart, more will be determined by their coupling to ground. In the case of a line with no ground, all of the coupling is to the opposite line, which is just the case on the left of that figure.


Differential signaling means there's no ground reference by definition. You can have a grounded shield around the twisted pair, but if you ground a wire the differential signal becomes single ended.

how will the differential impedance of 100ohms be achieved between the pair?

It's achieved by balancing the impedance of the twisted-pair cable with that of the receiver - the magnetics.

Also are the Ethernet pairs meant to be length matched?

Yes. IEEE 802.3 clause defines a maximum skew between pairs of 50 ns for 1000BASE-T. The skew must not vary more than 10 ns in use. For 10GBASE-T it's the same, for 100BASE-TX and 10BASE-T it's 60 ns/20 ns.

  • \$\begingroup\$ How does "balancing the impedance of the twisted-pair cable with that of the receiver" help? What does it mean; how it's done, and does it answer how to perform the PCB routing to get 100 ohm impedance? Also, the magnetics isn't the receiver though, signal passes through the magnetics. And you are referring to intra-pair skew, there can be a total of 50mm between pairs, and 100M Ethernet only uses one pair per direction so I doubt 1 or 2 meters extra on e.g. TX pair would make any difference. \$\endgroup\$
    – Justme
    Sep 28 at 18:50

Two wires running in parallel in free space have a set differential impedance that's determined only by the geometry of the cross-section of the wire pair. No ground planes needed.

That's why it's called a differential impedance: you start with two wires, it's all between those two. You can add planes to change it to your needs, make it more practical to manufacture, etc. But those extra conductors are entirely optional. Two wires is enough.


Not the answer you're looking for? Browse other questions tagged or ask your own question.