Finally, (some) technical details, though still not as many as I'd like.
https://www.facebook.com/permalink.php?story_fbid=740971779281171&id=178...
The original analysis was based either on a round trip time measurement or a reported transmitter power, which depends on satellite elevation. Either way, it could only yield a locus of positions that fell on a circle of constant distance (or elevation angle) to the satellite. Erasing the part of the circle that could not be reached with the available fuel left the arcs. I don't understand why there would have been a gap between the two around the equator.
The big problem was to solve the ambiguity -- northern hemisphere or southern?
I believe the answer relies on the fact that Inmarsat 3 F1 is an old satellite launched in 1996. Its inclination has risen to 1.67 degrees, probably because its stationkeeping fuel has run out.
The total measured Doppler would be the sum of the Doppler caused by the plane's motion with respect to the satellite and that due to the satellite's non-zero motion with respect to the earth. Judging from the diagrams on heavens-above.com, I'd say the satellite was moving southward during the flight, which took place between roughly midnight and dawn local time. Since the plane was in the southern hemisphere, the satellite was moving toward it, resulting in a higher net frequency shift than if the plane had been in the northern hemisphere.
Note the shapes of the curves in the plot titled "MH370: Burst Frequency Offset". The southerly track has the Doppler increasing with time, while the northerly track would have it decreasing with time.
The analysis does have to assume a certain fixed speed for the aircraft but it seems to produce consistent results.
It should be possible to replicate Inmarsat's graphs using the three line elements for the satellite and their presumed ground tracks for the aircraft.
Phil