I keep bringing it up because this thread is called “Fiberoptic Ethernet and bits on the wire”. Bit errors is how noise is seen in digital networks and if you don’t have bit errors, going to optical ethernet is pointless. You argue that CM noise is still there, and I keep telling you that it is attenuated and inaudible in well designed audio equipment.
To be clear, CM attenuation and 'inaudibility' is not something that 'well designed' audio equipment is immune from. That is simply handwaving and dismissing the topic. The topic is not the title, rather the post. No joke that both fiberoptic and copper ethernet perfectly transmit bits on the wire. No question. That isn't the purpose of the post.
Common mode noise is related but not the same as a ground loop. It is often not the fault of a single piece of equipment (but may be) rather the interactions between boxes. Let me spell this out:
A ground loop is the physical mechanism that most often creates common-mode noise on an Ethernet cable, so the two concepts are tightly linked but not identical. Common-mode noise is the symptom — a voltage or current that appears equally on both conductors of a pair with respect to some reference. A ground loop is one of the most common sources of that symptom.
The setup is simple. Two pieces of equipment — say a switch in a wiring closet and a PC on a desk — each have their own connection to "ground." That ground might be the safety earth at the wall outlet, the building steel, a rack ground bar, or just the chassis bonded to mains earth through the power supply. In an ideal world every one of those ground points sits at exactly the same potential. In the real world they don't. The earth conductor in the building's wiring has resistance and inductance, other equipment on the same circuits is dumping return current into it, nearby transformers and motors are inducing voltages in the loops it forms, and the result is that the chassis of the switch and the chassis of the PC sit at slightly different potentials. The difference might be a few millivolts of 60 Hz hum in a benign installation, or several volts of broadband noise in an industrial plant, or kilovolts during a lightning event.
Now you connect those two chassis together with a copper Ethernet cable. The cable's conductors, the magnetics' center taps, the Bob Smith termination, the cable shield if you have one, and any parasitic capacitance to chassis all form paths between the two ground references. A current flows around the loop: out one end's ground, along the cable, into the other end's ground, back through the building's earth conductors to where it started. That current is by definition common-mode on the Ethernet cable, because it flows in the same direction on every conductor in the bundle. The voltage that drives it appears, again in common-mode, across the magnetics.
So the relationship is causal and direct: a ground potential difference plus a conductive path between the two grounds equals a ground loop, and a ground loop forces common-mode current through whatever conductors complete the loop — including the Ethernet pair. The Ethernet magnetics are specifically designed to break this loop at DC and low frequencies by providing galvanic isolation, which is why a properly working transformer-isolated PHY tolerates a couple of volts of inter-chassis ground difference without any signal degradation at all. The differential signal rides through the transformer; the common-mode loop current is blocked by the isolation and by the common-mode choke.
The trouble starts when the loop's drive voltage contains energy in bands the magnetics can't reject, or when the loop has a path that bypasses the magnetics. Mains-frequency hum from a ground loop is below the choke's useful impedance, so it shows up as a slow CM offset; the transformer's galvanic isolation keeps it from doing harm at the signal level, but it can still cause problems for audio or video equipment sharing the same chassis. Fast transients from the loop — switching noise from a nearby VFD, ESD events, lightning surges — contain energy well above the transformer's parasitic-capacitance corner, and those couple across the isolation barrier and show up on the PHY side. And if you've bonded the cable shield to chassis at both ends, you've created a deliberate low-impedance path for the loop current that runs in parallel with the magnetics rather than through them; the shield carries the CM current happily and radiates it into the pair through any imbalance in the cable construction.
This is why the standard mitigations all attack the loop rather than the noise. Bonding the shield at only one end breaks the conductive path so no DC or low-frequency loop current can flow. Fiber between buildings or between equipment at very different ground potentials eliminates the copper path entirely. Isolation transformers on power feeds, common-bonded ground bars in a rack, and star grounding all aim to reduce the potential difference that drives the loop in the first place. The Bob Smith termination doesn't break the loop — it terminates the CM path in its characteristic impedance so any residual CM current sees a matched load rather than reflecting and ringing. And the PHY's own CMRR, plus careful pair balance on the PCB, ensures that whatever CM current does flow doesn't get converted into differential noise that the receiver would mistake for signal.
The clean way to think about it: common-mode noise is what you measure, a ground loop is one mechanism that produces it, and the Ethernet magnetics are designed on the assumption that ground loops exist and must be tolerated. They handle the slow, low-energy case by galvanic isolation and the fast, moderate-energy case by choke impedance and pair balance. They stop being adequate exactly when the loop's drive voltage contains energy outside the magnetics' working band, or when something in the system gives the loop current a path that doesn't go through the magnetics.
