The reason that you need to sync the timeserver to a external source, is just that, auditability. As others here have pointed out, a GPS receiver is a valid external source. Using a GPS receiver or a time radio source is in fact, a preferred way to managed time servers in air-gapped scenarios, as this means no sensitive data can leak out from the air-gapped area, and no malicious data can leak into the air-gapped area.
Yes, a malicious adversiary might be able to modify the time inside the air-gapped area by sending false GPS or radio signals, but nothing more.
It does NOT say that you need to let time server connections out from the PCI DSS boundary, it just says you need to, in some way keep the clock in sync with world time.
The reason you cannot just "run your own, internal, air-gapped timeserver" with no means to receive anything from the outside at all, is because imagine a breach occurs.
Then you cooperate with law enforcement, and the hacker is caught. Imagine then that all your logs are fully in sync internally, but your internal timeserver is 1 minute off from external time.
And now, the criminal have a aliby, lets say the hacker was, according to official records, detained by the police for suspected drunk driving but then was found innocent, at the exact time the breach occured.
Now, your evidence is basically invalid, because it shows something that couldn't happen.
Thats why you need to sync your time server to a external source. Which external source is used, does not matter, as long as its "industry-accepted", eg a common way to sync timeservers, that will be sound from a law-enforcement perspective.
If you ABSOLUTELY must keep the PCI DSS area air-gapped for some reason, lets say inside a thick vault, you can provide time services anyways, by having a external time receiver (that receives signals from a GPS receiver or similiar), that will then output PPS (Pulse per second) via a laser that are pulsed into a small window* in the vault, and inside the vault, you have a receiver, that will sync the internal time server. (and thus, the internal time server will only step forward on laser pulse, and if external time server drift from world time, it will correct by sending faster or slower pulses)
*This window must also be made in a way so it does not compromise the security of the vault.
Note that you also need a means to initially sync the internal time server if you go on this route.
This has nothing to do with separation of duties. Separation of duties are required only if you handle such a amount of transactions that make this neccessary.
If a single-man company process credit card details, single-duty will be accepted, but only for a small amount of transactions, where the risk of corruption is low.
What Trey Blalock said about having 2 internal time servers logging each other, not managed by the same system admins, would not get accepted as compensating control, because you have no means to ensure this time is correctly in sync with world time. Yes, you could sync them from a external source manually, but that would mean the time server in the meantime, drift from world-time.
One compensating control would however be to have a OCXO (Oven compensated crystal oscillator) that is once synced to external time, before brought into the PCI DSS area. This OCXO could also be sealed with tamper-resistant seals.
But note that also a OCXO drift with time, and needs to be periodically checked and resynced.