In reply to Jonny2vests:
> For a good height, you need some satellites near the horizon, geometry is the key
Acually, for good height, you want satellites away from the horizon; consider two satellites on the horizon, one W and one E. Now think about how the range spheres intersect; for a small range ambiguity, you get a very large intersect circle, because your intersect points climb up the steep (near vertical) slope of the intersecting sphere. Geometry, as you say, is the key.
Not only that, but satellites on the horizon are subject to much larger iono and tropo degradation, because the signal passes through a lot more atmosphere, and is further away. The receiver computation engine should be weighting out low elevation satellites from the solution.
The best satellite position for good height values is one directly overhead, since the height error it gives is simply the range error.
Generally speaking, the best constellation would be one in which satellites were mutually orthogonal...