Originally posted by photoptimist Astrotracer performance is a complicated issue. The first big issue is the calibration which some people have trouble doing for various reasons.
But even with perfect calibration, astrotracer has time limits that depend what part of the sky you are looking at and on focal length.
First is the issue that stars on the celestial equator appear to move faster across the sky and the sensor than do stars at the north or south pole. Thus, photos that include the celestial equator somewhere in the image have shorter time limits than photos pointed exactly at the poles.
Second is a more complex issue with focal length in which both very long and very short focal lengths have shorter time limits. For long focal lengths the problem is obvious -- a longer focal length magnifies motion so the sensor hits the limit sooner. But a different problem happens with very short focal length lenses that is caused by discrepancies between spherical geometry (the rotation of the celestial sphere) and rectilinear geometry (the way most high-quality lenses image the world). The same effect that makes people and objects seem weirdly stretched at the edges and corners of UWA images affects astro images, too. The net result is that stars at the edge of the frame move faster than stars in the middle of the frame and the astrotracer correction cannot simultaneously remove all the motions. The exact time limit isn't easy to estimate because it very sensitive to exactly where the camera is pointed, the lens focal length, lens distortion, lens sharpness, and the photographer's subjective standards for residual trailing.
P.S. Technically speaking, the astrotracer does use data from all five axes to make it's corrections. It measures the X-Y location of the camera on Earth with GPS as well as the 3 rotation angles of the camera orientation relative to the Earth (using the compass and accelerometers). It then uses all that data to correct for the effect of the rotation of the Earth by moving the sensor in 3-axes (sensor X, Y, and rotation). (If astrotracer did not correct for rotation, then every astrotracer image would have some swirl of star trailing.)
I actually agree with everything posted, however there are a couple of additions.
First, the astrotracer is a consumer item, it's not going to be absolutely perfect. For what Pentax implemented, it is pretty good. For my purposes, untracked I can get about 13 sec of exposure time (maintaining a star in a 4 pixel box). With the aid of the astrotracer, that time can be increased to 60 seconds. Actually not bad - that is a 4.6x improvement, at the cost of about 2oz in weight and $175 in cost. I just think that it could be a bit better - and that comes from trying to understand Pentax's implementation, and how it could be improved.
Sources of error - the compass and accelerometers - I do think that the accelerometers are pretty good. That said, the compass appears to have up to a 10 degree error +/- at times (after calibration) - which has been touched on by others here in the forum in other threads. Again consumer electronics - but for all essential purposes the astrotracer is a computationally based simulated equatorial tracker. As in any tracker, polar alignment (pointing North) is critical, and a compass error is not going to help matters any.
I see and read (here on PF) all the time about wide angle lenses with their inherent distortion across a wider than normal field of view - and due to the WA and UWA focal lengths, you can't track properly. However, if you consider a camera and lens set mounted on an equatorial tracker (or even using a barn door implementation) that is properly aligned - you will obtain excellent tracking for quite some minutes (greater than 1 minute). Then, taking the exact same camera and lens set and using it in the astrotracer environment, for longer than 1 minute up to 5 minutes - you will encounter the star trailing (the longer the exposure, the more trailing). So what is the difference? With the equatorial tracking mount, you are tracking in a environment that physically moves with the spherical geometry - you are moving the entire camera/lens as a monolithic unit. In the case of the astrotracer - only the sensor is moved while the surrounding camera/lens remains stationary. The implementation as I understand it and others have stated, the sensor only moves in the plane of the sensor (X and Y and rotation within the defined XY plane - essentially yaw - no pitch and roll). There are limits to the physical movement of the sensor, but again as I understand it, with the new 5 axis design, there can be additional rotational movements implemented (pitch and roll to a very small degree). I do believe that if implemented, the astrotracer could be improved, and this star trailing problem could be mitigated to some extent. It all boils down to an accumulation of error, that manifests itself as the tracking time is increased. Basically, Pentax has done itself no favors by "advertising up to 5 minutes". It's an accurate statement, but everyone just reads "5 minutes" and comes away a bit disappointed.
I think I do understand the problem pretty well. I helped design and implement a
little telescope down in Texas - over 20 years ago. I did the
star tracker. The design has also been used over on a
second telescope in Africa.