Originally Posted by
Fitter2
For good geometrical reasons, the algorithm which determines height above geoid is separate from that which outputs lat/long co-ordinates.
Please clarify. I believe you might be referring to the coordinate transformation stage, from ECEF to whichever horizontal/vertical/combined reference systems the user chooses to have. The navigation solution itself is obtained by solving all four unknowns simultaneously and does not even involve the concept of local vertical (and hence altitude). Also the geoid correction does not enter the stage until much latter, if at all.
The reason why the vertical component is less precise than the horizontal component is indeed geometrical however: as you can only see satellites above the horizon, you have no observations at all coming from below (the ones coming from low elevations are also less reliable due to atmospheric effects), which increases the measurement uncertainty in the vertical direction. Note incidentally that this effect diminishes with height above ground due to the lower horizon elevation amongst other things.
The relative "goodness" of the position solution in different planes/components is given by the dimensionless DOP (dilution of precision) indicators.
Since the primary interest of GPS users is lat/long
Do not generalise please. Timing is a major field of use.
Lastly, the geoidal correction bit, which seems to mystify a lot of people is
not a major factor in recreational GPS use, aviation included. I have geoidal correction source code at hand, originally from a consumer grade GPS receiver which is less than 7Kb (7072 bytes to be exact
) and computes a "rough" correction based on bilinear interpolation on a 10 degree grid. This achieves accuracy in the order of 2-3m worldwide (with the possible exception of the great mountain ranges, I haven't checked). Clearly, that's unacceptable for geodetic or many engineering uses, and for Cat III autolands (where the radio altimeter controls the flare anyway), but for general aviation it doesn't matter a bugger.