Measuring Ethernet Impedance

Question

I have a PCB, which was not designed by me. The Ethernet works intermittently between multiple copies, except for on one, where it works just fine.

I want to start troubleshooting at layer 1, and when I looked at the design for this PCB, I noticed that there is no impedance control, or length matching on the traces going from the RJ45 to the PHY (which is integrated into the microcontroller). These traces are routed like GPIO, not like a differential pair. There is also no ground plane on this board, so calculating the actual impedance if the traces is going to be very difficult.

I suspect there is a signal integrity issue that is causing the ethernet to work intermittently, but how do I go about proving it? I was thinking of using my VNA and measuring the traces individually (single ended), but I was hoping to get some different ideas from this post.

The ethernet jack in question: Mouser Link

So, I did end up hooking my VNA up to the boards, and here is the results I got. Note this was on raw boards, no RJ45 or MCU attached. Measured at 66MHz.

Line	Resistance	Reactance	SWR	Smith Chart	Delay	Complex
TX+	16.1Ω	-64.2Ω	7.2	16.0Ω 37.2pF	4.0ns	.21-j.75
TX-	18.3Ω	-64.2Ω	8.5	18.3Ω 38.4pF	4.2ns	.19-j.76
RX+	16.5Ω	-62.0Ω	7.7	16.5Ω 39.2pF	4.0ns	.19-j.75
RX-	15.8Ω	-64.0Ω	8.5	15.7Ω 37.5pF	4.0ns	.20-j.75

Answer 1

I suspect there is a signal integrity issue that is causing the ethernet to work intermittently, but how do I go about proving it?

First and foremost - there is not differential pair or impedance for the differential pair, you could estimate it with a calculator. But looking at the traces it probably isn't going to be a deal breaker with such a short run. If I have a diff pair with 10mils traces with a height of 15mil above the reference plane and 5mil spacing, I get close to 100Ω of impedance. Increasing the spacing to 50mil gives me 150Ω but for such a short run it probably won't affect the wavelength too much at ethernet frequencies.

Another thing would be the layer stackup, if you aren't getting a 50Ω impedance for the single ended traces then that could be an issue, but then again it looks like a short run and well under 1/16th of a wavelength.

The best thing to do would be to check the eye diagram with a differential pair and see if you are getting overshoots or something the not-valid area. If you are then re route. Also check the VCC and other area of the board for noise if you are seeing eye problems.