A TDoA system should in theory be most accurate.
To achieve the best possible accuracy all the anchors should be on a wired common clock with good synchronisation between them and calibrated antenna delays. You can manage without that wired clock but it drops the accuracy which seems to be what you care about. Also for a solution you need one more anchor than for TWR, there is one more unknown that needs to be solved, the tag transmit time.
TDoA is also better for movement since all measurements are made at the same time rather than one at a time so there is no opportunity to move between each measurement.
TWR will be a little less accurate but doesn’t require the clock synchronisation between the anchors making it a lot simpler to deploy.
There are also differences in things like where locations are calculated and the maximum number of tags supported but it sounds like you don’t care about those details so much.
PDoA can be added to either system type to give more information per anchor. This either allows you to decrease the number of anchors or increase the accuracy for a fixed number of anchors.
By far the best way to improve accuracy for either system is to add anchors. Even if you manage to dial out all the biases and offsets in the measurements there will be some random noise. The more anchors to average between the more the effects of this noise can be decreased. But more anchors means more complexity, especially for a TDoA system if you need to worry about clock synchronisation.
If things are fairly static then averaging over time will also decrease noise significantly. However this isn’t as good as adding anchors. All anchors will end up having a slight bias in their measurements, averaging won’t remove these but if you add more anchors then the influence of individual anchor biases will be reduced and there is more chance of the biases averaging out.
When you say 1 cm accuracy do you mean absolute or relative? Relative accuracy would be if you move 50 cm then the two locations are 50 cm apart but both outputs could offset from the correct positions… Absolute accuracy is that the real world location for any given point is also correct.
Getting good relative accuracy over short distances is relatively simple. Getting absolute accuracy or relative accuracy over larger distances can be a lot harder.
We run a TWR system and can achieve ~3 cm RMS absolute accuracy using 8 anchors, dropping closer to 2 cm if you use 12 anchors. But a lot of work has gone into maintaining that accuracy over large areas, initial tests had a pincushion effect, if you travelled in a line through the middle it was straight, if you travelled in a line down the side it was slightly curved.
I remember someone posting here that they had a TDoA system deployed that achieved sub-cm accuracy over a basketball court but that was using something like 50 anchors.
So to answer your question:
TDoA + PDoA is in theory going to give the best accuracy but is also more complex to implement. To achieve the best possible accuracy you will need to put a lot of work into tuning and calibrating the system. 1 cm accuracies are probably not possible with 4 anchors but you may come close if you can average over time. Adding more anchors will improve accuracy.