Every LoRa gateway have a GPS receiver to give precise timestamping to each LoRa frame received.
GPS has a UART output with the time in seconds and a PPS (Pulse Per Second) 1Hz output that triggers a Timer inside SX1301 that gives microsecond or less precision (I supose 1/32MHz).
The advantage of LoRa modulation is that it is easy to determine the beginning and the end of a bit or a frame, because of the chirp modulation. In FSK it is not easy to determine the beginning or the end of a tone.
In GPS location you know the T=0s and the time of arrival of each satellite and you can triangulate (draw circles with the center in each satellite). The calculation is done in the GPS receiver.
In LoRa you don't know the T=0s and you don't know the time of arrival (TOA) to each gateway. The LoRa Network server knows the time difference of arrival (TDOA) between every pair of gateways. If you have 3 gateways you have 3 TDOAs. And if you have 4 gateways you have 6 TDOAs.
Instead of drawing circles to triangulate, you must draw hyperbolic curves with the two focus points in each pair of gateways. The calculation is done in the LoRa Network server, not in the node.
In page 22 you can see an example with 4 gateways, so 6 TDOAs, so 6 hyperbolic curves colliding in a point that is the estimated location of the node.