>>3 satellites would be enough
Global positioning system(s) assume an 'earth centered, earth fixed, x-y-z 3D cartesian coordinate system'. Any location in this 3D space requires no more than 3 components to be completely identified. So, even though 3 spheres we obtain by 3 distance measurements intersect at two different points, one of those points is rendered useless by [ earth centered + earth fixed ] characteristic of the coordinate system GPS assumes; we are interested in locations below the earth's atmosphere. 3 satellites could be used to determine 3 position dimensions with a 'perfect' receiver clock (with an expensive atomic/optical clock).
!YES!, you !could have gotten! a 3D position fix with 3 satellites IF the GPS receiver you are using was equipped with an atomic clock. (The ELIMINATION of the second point, on the bottom left figure of the illustration above, is done "intuitively" as it corresponds to someplace in DEEP SPACE. BECAUSE, there is a reason why GPS satellites are at their specific constellation (~their setup in the sky): !more than! 24 GPS satellites, on 6 orbital planes that are ~20,000 kms above you, and 4 satellites on each plane, 60 degrees between these planes, and 55 degrees inclination with respect to equatorial plane, GIVES YOU 5-8 satellites that you can "connect to" from (almost) any place on earth, and 3 SATELLITES TO GIVE A 3D POSITIONAL FIX ON EARTH. If we are talking about locating things "inside AND outside" of earth, WELL THEN YES, you need at least 1 more satellite to eliminate one of two possible intersection points in the last step. This wasn't the question, was it?
In practice, placing expensive clocks in GPS receivers is rarely possible/feasible and 3 space vehicles (SVs, i.e. satellites) can be, instead, used to compute a 2D, horizontal fix (in latitude and longitude) when a certain height (e.g. z-dimension) measurement is ASSUMED; so you get rid of 1 dimensional measurement out of 4 that were originally required. Assumed height can be either the sea level or the altitude of a (normally) altimeter equipped aircraft.
It is the height dimension that is chosen to be discarded, because it is the (relatively) least important one among others. Among the 4 required dimensional mesaurements (x,y,z,time), time always needs to be resolved BECAUSE satellite signals (electromagnetic waves) travel at the speed of light and reach receiver in ~0.07 atomic seconds; and thus, a slight inaccuracy in the GPS receiver's relatively cheap internal clock would give a "very wrong" locational fix because of the extra distance the signal is assumed to travel at the extreme speed of light. And, well, the other two dimensions will place the GPS receiver on some (longitude,latitude) pair on the surface of the planet.
More than 4 satellites provide better accuracy by introducing additional 'time difference pairs'. 4 dimensional requirements remain, yet number of independent equations increase and exceed 4. This will result in an over determined system of equations with multiple solutions. Over determined systems are !approximated! with numeric methods, e.g. least squares. In this case, the least squares method will give the position (of GPS receiver) that best fits all of the time measurements (with extra dimensions) by minimizing the sum of the squares of errors.
(1)
Global Positioning System Overview, Peter H. Dana, Department of Geography, University of Texas at Austin, 1994.
http://www.colorado.edu/geography/gcraft/notes/gps/gps_f.html
(The Master GPS Control facility is located in Colorado, Schriever Air Force Base)
(2)
Position Determination with GPS,Dr. Anja Koehne, Michael Wößner,Öko-Institut (Institute for Applied Ecology),Freiburg im Breisgau, Germany
http://www.kowoma.de/en/gps/positioning.htm
(3)
An Underdetermined Linear System for GPS, Dan Kalman
https://www.maa.org/sites/default/files/pdf/upload_library/22/Polya/Kalman.pdf
(4)
For the colorful illustrations
http://www.colorado.edu/geography/gcraft/notes/gps/gif/figure09.gif
http://www.colorado.edu/geography/gcraft/notes/gps/gif/ecefxyz.gif
http://www.colorado.edu/geography/gcraft/notes/gps/gif/gpsxyz.gif
http://www.colorado.edu/geography/gcraft/notes/gps/gif/navigate.gif
>> INaccuracy
" Four sphere surfaces typically do NOT intersect. Because of this we can say with confidence that when we solve the navigation equations to find an intersection, this solution gives us the position of the receiver along with accurate time thereby eliminating the need for a very large, expensive, and power hungry clock. "
http://en.wikipedia.org/wiki/Global_Positioning_System#Basic_concept_of_GPS
It says "typically" BECAUSE the measurements are Inaccurate; otherwise they would intersect at exactly one point. From 4 satellites, you get 4 Inaccurate distance measurements. INaccuracy in all these 4 measurements are SAME (=in the same amount) BECAUSE satellites use atomic clocks which keeps them perfectly synchronized among themselves (and accurate with respect to GPS time scale), additionally, the INaccurate clock in the measurements remain the same too, because we are talking about one particular GPS receiver. Since accurate and INaccurate clocks, and thus the INaccuracy, is constant in our measurements, there can be only one correction value that reduces the volume of intersection of 4 spheres to a single point of intersection. That value represents the time INaccuracy.
(5) The UTC clock is currently (2012-11-14) 16 seconds behind the GPS clock.
http://www.leapsecond.com/java/gpsclock.htm
(6) How a GPS Receiver Locks, Thomas A. Clark, NASA's Goddard Space Flight Center
http://gpsinformation.net/main/gpslock.htm
(7) How Accurate is a Radio Controlled Clock?, Michael A Lombardi, NIST-Time and Frequency Division, Maryland
http://tf.nist.gov/general/pdf/2429.pdf