One thing that a lot of the UAV guys are doing, is to use a magnetic
heading sensor coupled with a MEMS gyro, or if more money is available,
a FOG. The magnetic sensor can be used to periodically determine heading
and to correct for signal drift from the gyro. For instance, the gyro
can be used as the primary source for heading info, while the magnetic
sensor is used during long, stable glides. Very similar to how the DG
and compass are used in power planes. Of course, the gyro would have to
be either mechanically gimbaled to remain parallel to the horizon, or,
alternatively, several gyros on different axes can be used together to
determine actual heading change, rather than yaw rate.

Using some basic Kalman filtering, it should be possible to get fairly
accurate (5-8 degree) heading info. This will, of course, be magnetic
heading, and subject to the usual accuracy issues. However, it might be
possible to correct the data to true heading by loading a map of lines
of deviation and magnetic anomalies (which can also be fed into the
Kalman filter). Simply feed the wind component data from the GPS, the
raw heading info from the magnetic sensor, and the stabilized data from
the gyro into a Kalman filter, and pretty stable and accurate heading
data should be available.

Definitely do-able. Let's do it! :O)

BTW, check out www.rotomotion.com. These guys have several fairly cheap
AHRS', as well as several cheap IMU's. They've also published a bunch of
open source code on Sourceforge under the Autopilot project
http://autopilot.sourceforge.net/ with an effective Kalman filter to
integrate GPS, IMU, and magnetic sensor data.

- Robert