My apologies, I was in a rush, so I should probably have phrased my answer differently.

Rather than producing corrections in the way WAAS/DGPS would do (I know I'm here, so this is the error), the AGPS system provides the Ephermeris data which allows the AGPS chipset to know which satellites are in view and how quickly they are moving (to allow it to a) work out which frequency to try to record data from for each satellite, b) work out how much of a Doppler shift to apply to said frequencies due to the satellites movement).

I take back what I said about ionospheric disturbance, perhaps. For this reason: The satellites broadcast Almanac (gross pos'n/speed data) and Ephemeris (fine corrections), the GPS chipset must presumably then use this data to generate a list of frequencies (for the satellites it knows are in view) and frequency offsets (for the Doppler shifts of said satellites knowing their positions and speeds). The SUPL (AGPS) server presumably either provides the AGPS chipset with the raw Alamanac/Ephemeris data from which it does these calculations *or* with the pre-processed data (the former approach would require no processing on the SUPL server, it would simply forward the data it has received from the satellites/DoD; the latter approach would require that it models the location of the satellites and can provide the pos'n/speed for any position on Earth - wherever an AGPS chipset is sat and asking for directions).

From what I've read, there are a few different types of AGPS service, and one of these (at least) allows the AGPS chipset to send fragmentary signals back to the SUPL server and it will use brute force to work out the receiver's position (without the receiver needing to capture all of the signals, which contain the Almanac/ephemeris data).

So to sum up, I'm not completely clear on what is/can be done either, but for the last case, knowing ionispheric conditions is probably necessary.

It's an interesting topic, perhaps we should branch it off into a seperate thread to avoid polluting this one?