A bit about GPS and GPS attitude
Differential GPS: This requires an accurately surveyed site to broadcast correction data – typically it needs to be within 100km of the user to be effective. It is therefore mostly used for maritime navigation close to land. A more useful variant is to the link the correction data from a host of accurately surveyed sites across a landmass to a central site and then generate a contour map of errors, a bit like pressure isobars. This map is unlinked to a satellite and rebroadcast on one of the spare GPS channels for use in the GPS receiver to give significantly improved accuracy. This is known as a Satellite Based Augmentation System (SBAS) but has only limited landmass coverage (WAAS in the USA, EGNOS in Europe and MSAS in Japan). These are intended for aviation use including Category I approaches. For obvious reasons it will not be available over large oceans, but then it is not necessary to know your location that accurately in cruise.
GPS Attitude: This does not require accurate GPS location information (e.g. Differential GPS or SBAS) but relies on the phase differential of the received carrier as seen by a number of closely spaced antenna. For an example of equipment capable of generating GPS attitude data refer (opens as .pdf):-
http://www.promagellangps.com/assets...en_16Jan08.pdf
It should be noted that GPS attitude measurement requires a number of antenna on the aircraft whose 3D location geometry is accurately known. The further apart the antenna are the better, but any unintended variation in their relative position will result in attitude calculation errors, so mounting them on wing tips is not ideal due to flexing. I have heard engineers talking of errors resulting from flexing of the fuselage in flight and these variations may be difficult to characterise and thus correct. However, attitude accuracies of about 0.2 degrees can be achieved with a separation of only about 2m and adequate structural stability at this distance should be readily achievable.
Use of standard GPS for attitude determination: The concept proposed by Peter H should work in theory, but two matched GPS receivers will not produce identical results. A long baseline (separation) would be required e.g. wing tips, and with residual GPS errors and wing flexing I would expect a roll error of typically 1 to 5 degrees. Differential carrier phase measurement technology would improve on this by at least an order of magnitude.