I used to work at a precision GPS company, on differential GPS. What they would do is have a 'base station' that would stay in a single spot, and average the position from the GPS signal for a period of time. This is because, due to atmospheric interference, the position 'wobbles'.. Once we have an average position, we use that position to come up with correctors to send to the mobile units (via radio modem usually, though other means are possible). This got us to be on par with what this article is claiming.
I wonder if they account for the 'position wobble' of atmospheric interference. I suspect its possible, as they just pick one position they've received, and use the inertial adjust for the correctors. Not much more work than we were doing.
(I didnt write any of the algorithms or anything, just shuffled data around to different devices and libraries.)