There is no dedicated GND in the classical Ethernet 8P8C ("RJ45") pinout.[1]
Why does the Ethernet spec not include a ground, unlike many other cable types used for interconnecting devices that may both have its independent power source as well, e.g. RS-232 or USB?
If you just ignore the POE 48 Volts in the image below, you can see Ethernet uses transformers on both sides.
This way there is no need for common ground as long as the common mode voltage stays below 1500V generally. The isolation specification of the transformers.
And as a bonus you now also know how POE works. (802.3at)
However, CAT6A often has a shielded connector. The shield is then grounded to chassis using the little flaps inside the socket.
Why is Ethernet not grounded? There are two reasons:
The reason Ethernet is more susceptible to a ground loop is because:
The companies and engineers that designed the ethernet spec had this in mind (there is a lot of thought that goes into specs)
If you had a ground in between the transmitter and receiver it would create a ground loop. This ground loop would be formed by the cable and the return path would be mains ground as shown below. Any magnetic fields that flow through the loop would create a current along the cable (and the rest of the loop). Even if you isolated the signal wires this would be a problem because of mutual inductance between wires (wires run along side each other can couple currents from one to the other). This would inject noise (and cause potential bit error and packet loss).
So If you add an isolation transformer between devices, you break the loop and are still able to transmit a high speed signal between the transmitter and receiver. Another benefit of galvanic isolation, also increases the impedance of the cable to the device in the event of a large electrostatic discharge.
This is an example of isolation between two devices, ethernet has two isolation transformers, but the result is the same, it breaks the ground loop (and reduces common mode noise in conjunction with the twisted pair and common mode choke).
Images from wikipedia on ground loops
/edit: as Tom Carpenter pointed out, properly implemented POE, with isolated DC-DC converters, still retains the galvanic isolation and "no wire at ground potential" property. (Ok, POE breaks the galvanic separation part and adds kind of grounding, not clearly visible, yet it's there. But POE is a hack on top of the Ethernet, not an original part of the specs. Non-POE devices retain the original advantages.)
USB has differential signalling as well, but it also carries DC power. Mere existence of the powering possibility makes the common ground necessary.
RS-232 doesn't carry power, but the signal is not self-contained differential pair, it's a single wire referenced against the ground - that makes the common ground necessary.
Grounding is often misunderstood to be the end-all solution for connecting things together. However, in most cases, even on short runs, grounding adds more issues than it solves.
The issue with sharing a ground over any distance is you are assuming both ends are at the same ground potential. In a perfect world that may be true, but in real life it almost never is.
Whether it be due to bad wiring, ground leakage, or EMI effects the ground over here on this monitor is different from the ground over there on your TV. As such, when you run a cable that includes ground between them, there will be a current running through that ground.
Moreover the common ground becomes the current return path for your signal. That means you are actually adding noise into the ground line. If your communication system uses multiple lines, they effectively share the same return path and the currents get far more complex, and the noise much worse, in the common ground.
The longer the cable the more voltage difference will be evident along that ground cable. If there is enough difference the delta between ground and the signal voltage will drop so low you are no longer able to distinguish the signal.
The image below demonstrates this. Note you have two lights above your ground in the distance. You can see that with a nice solid ground you can fairly easily tell in the middle two which light is turned on. However, in the right hand situation where the ground boundary is hard to identify, it is no longer possible to tell if that is a high or low light.
Standards like ETHERNET and other differential communication systems use a different technique that eliminates the need for a ground entirely.
By sending as a plus and minus signal over two dedicated wires the receiver can pick out the signal by examining the difference between those wires rather than comparing it to a passed reference voltage. (i.e. "Ground"). The image below indicates how this works. Notice even with the noisy signals on the right you can still tell which signal is being sent.
This technique not only allows the signal to be transmitted over much greater distance but also reduces the systems susceptibility to common mode noise. Since each signals current paths are also restricted to those two dedicated wires the inter-signal return path sharing is eliminated.
For Ethernet in particular, transformers are used to connect to the wires which provides complete isolation between the transmission medium and the sender/receiver.
To the mix of questions and comments of USB and Ethernet, such as why is Ethernet galvanically isolated, and USB is not:
Read up on USB history, and its mandates. It was to be a "low cost", "short distance" (5 meters) signaling port fit for computers at home and business, and low cost above all other needs. USB also had the mandate to improve data rate performance over parallel-printer and RS232 ports.
USB was to be put on all PC's manufactured. Whether the user needed it or not. That means it must be low cost. Parallel-printer ports and RS232 ports and the large associated connectors, had a serious cost penalty on all popular computers. And that made PC's and laptops, costlier, bigger, heavier and more power consumptive. USB, thus to be very "low cost", has no transformers to accomplish galvanic isolation. And it makes it easy to provide DC power to the peripheral. USB data signaling, for lack of a better phrase is "semi-differential". That is the current in the + and - lines of the cable, are about 95% numerically opposite matched (the + and -, are always slightly in error, of not being a perfect opposite current value), as a different transistor set drives each net, + and -.
Ethernet's mandate was and is; "reliable", "medium distance" communication, and low cost. But reliable and medium distance come first. Medium distance of 100 meters needs galvanic isolation very much. If two devices (such as a switch and a PC) were connected across two buildings with a few volts of ground potential difference, that's a rather bad thing, and unintended, unwanted, ground current will flow in that data cable. And that unwanted ground flow, can have all sorts of bad effects of harming the data quality and harming the equipment's, to possibly even endangering people.
Ethernet, also has different transistor sets driving each + and -, however the signal transformer shorts the + and - together, and thus the final + and - current flow is nearly a perfect, opposite match, down to nearly a single electron. Thus true differential signaling is achieved. True differential signaling allows the signal voltage levels to be further reduced, and the cable distances traveled, to be increased, and for unwanted EMI to be reduced.
Later came PoE for Ethernet. PoE's mandate was to bring "low cost" DC power to peripheral devices, ie, VoIP phones, Cameras, & Door-Access units. The PoE typically power exits the common Ethernet switch, to multiple devices, up to 100 meters in opposite directions. That PoE (48 to 57) VDC is a "star" connection to all the devices. That means the multiple power using devices "PD", share a common supply (this is NOT isolated power, per RJ45 connector at the PSE). Therefore it is a GUILT-Edged "shall" that the PD's shall maintain power isolation (as per the IEEE 802.3 standard), even on the PoE power inputs, by way of either an DC-to-DC isolated converter supply in the PD, or the PD is entirely in a non-conductive case, and never are its circuits ground plane connected to the local ground of the building or other nearby equipments (such as really low end cheap peripherals). Unfortunately, IEEE 802.3 at PoE standard does not spell this out very clearly.
Summary: Ethernet has transformers at both ends. If even a transformer failure happened, galvanic isolation from PD remote device, to the PSE in the Ethernet switch is not lost.
PoE, gives up DC power isolation, (for the sake of low cost) at the Ethernet switch, and leaves this isolation "shall" up to the PD peripheral manufacturer. Nobody is really checking on these manufactured items. If IEEE put a reward bounty on the violators, that would improve the situation.
The new PoE standard, IEEE is pondering for even higher voltages and currents, for more PoE power, should move toward some improved quality and safety. These should be on only commercial/industrial grade or better installations: 1) full power isolation at the PSE, for each connector. 2) required test reports for PSE & PD power isolation, that are filed, and downloadable for the public. To include the wiring diagram of the PI. 3) at the manufacturers cost, maintain a server, with a list of all PD's that meet the new standard. 4) consider making an industrial grade standard, if the cost of these improvements are too much for the low end consumer markets, yet still provide the serious level of standards, safety and traceability of industrial needs.
The answer thus far have missed one key element: shielding couples noise.
The Ethernet standard has included both UTP and STP (unshielded/shielded twisted pair) for decades.
IBM heavily influenced the original inclusion of STP, for its compatibility with Token Ring. The claim was that the STP shielding provides an extra layer of noise protection for the differential pairs (a bargain at only 5x the price!) However real-life experience quickly demonstrated that the shielding provided worse performance. Point sources of electrical noise were coupled to the shielding, where the noise then had the entire length of the cable to be coupled to the twisted pairs.
Shielding can also increase cross-talk. Pairs are twisted at slightly different rates -- a typical schedule is 11/12/13/14 twists per foot. This way they don't physically nest together to form a parasitic transformer when the cable is twisted and pulled during installation. This works great. Way better than you might expect. But the wiggling will stretch the shield between the pairs, coupling the signal to the shield and to the other pairs.
The Ethernet specification calls for devices to be galvanically isolated from each other - this post explains the isolation implications in more detail.
Since you mentioned power sources and grounding, the Wikipedia article on Power over Ethernet (PoE) may be relevant to you.
One reason WHY TP ethernet, thank $DEITY!, is not using a common ground reference between stations has not been mentioned clearly enough:
An ethernet cable can run in excess of a hundred meters, probably even connecting two buildings.
The potential difference on even a well-executed grounding system, if measured at two points tens or hundreds of meters apart, is far from guaranteed to be even close to 0V - there can be DC or low frequency AC voltage due to current flow (due to leakage currents, actual fault currents, transients, or wiring mistakes/defects...) in the grounding system, and all kinds of interference.
AC potential differences will readily cause interference, while strong current flow across a grounded cable shield could actually turn into a fire hazard.
And all this assumes equipment being properly grounded in the first place, even worse things can happen if that is no longer the case.
Not worst, but bad case example what could happen with grounded (on both sides) shields : One PC has accidentally been connected with a two wire IEC cable (as has been found in the wild before!) on an RCD-less (old TN-C-to-the-socket ...) wiring system. This PC develops an internal short connecting live mains to the metal case, to which all the ground connections in that PC are connected. The other end of that connection is to a device where the grounding has been correctly implemented. The ethernet cable used is of the lightest build possible, with thin shielding material. And it is a 30 meter piece wound up in a coil since you were out of 5m cables. At that length, this shielding could have just the right resistance (around 15 ohms would be "perfect" on a 240V system) to pass a current that is small enough not to blow any fuses or automats, but well large enough to dissipate in excess of 1000 watts on the cable shield. Which will mean a lot of smoke and not unlikely an ignition hazard to stuff around it.
Both RS-232 and USB have signals referenced to ground, that's why you need one. (Yes, D+ and D- signals in USB are used independently from each other with GND as a reference during device discovery). Ethernet signals are purely differential, so no GND reference is needed.
As you may know, all laptop/tablets have floating ground unless connected to some port with an external earth ground such as a VGA cable to a 3 pronged LCD monitor due to the galvanic transformer isolation in the battery charger.
Ethernet also has no signals in the lower spectrum including DC because of the Bi-Phase encoding methods.
More important is that for signal integrity and EMI reduction for egress and ingress, the signals are transmission lines with 75 Ω termination and balanced with CM transformer and 1:1 centre tapped transformer. This raises the CM impedance on the user side for isolation while maintaining the differential impedance on the cable side to local ground in the upper spectrum where signals exist.
