[mikeash]

Can't find much information about it, but what I did find says that it got significant use in the early years of the 747. It makes sense, as the 747 long predates civilian GPS, and other electronic navigation techniques were not completely reliable for long flights. I don't know that there were many radio beacons available for long over-water flights, and pilots probably liked cross-checking their INS.

While searching, I was reminded of another interesting fact, that the SR-71 had a fully automated celestial navigation system that could get still a position fix on the ground in the middle of the day, and was more accurate than an INS on long flights.

[fancyketchup]

There are some really interesting patents related to this (search for variations on celestial navigation at USPTO or Google Patents). Apparently it is possible to track stars in full daylight using a single photosensitive device behind an IR-pass filter (stars radiate IR, but the atmosphere doesn't scatter significant IR from the sun) and a clever arrangement of shutters. Several of the most interesting patents are refinements to allow the system to work well during dawn and dusk, when the background brightness of the sky varies greatly with elevation above the horizon.

[sbierwagen]

  that could get still a position fix on the ground in 
  the middle of the day

What! Surely the sky would outshine the stars... did it use light outside of the visual spectrum?

[Florin_Andrei]

Jupiter, Venus, Sirius and, at the very limit, even Canopus are visible with the naked eye during the day. It's just hard to "get a lock" on them because contrast is so low. But if they are close to a large, fixed, easily visible reference (the Moon, a tall building, a tree) then suddenly they pop into view. The first two are easy, the third is doable in good conditions, the last is difficult and requires excellent conditions and a keen eye.

You need more than a few stars for navigation, of course, but I would assume with a good sensor, proper filtering, and nice optics, a computer could see enough stars during daytime to perform navigation.

[mikeash]

Check out fancyketchup's sibling comment to yours, which shows far more knowledge than I have. Short version: infrared.

[fancyketchup]

The IR filter helps, but (this is sort-of speculation because I don't have any knowledge of the system outside of reading the patents), I think the major processing gain comes from the shutters. In fact, I think the patents said that it could work (albeit much less effectively) without the IR pass filter.

The basic idea is a form of synchronous detection or lock-in amplification: The shutters are arranged so that a point source is occluded at a fixed frequency, while an area source will be only partially occluded at any given time. Due to the way it is designed, area sources should add little to the time-varying component of the photodetector signal. The lock-in amplifier is essentially a very narrow notch filter (centered at the "star occlusion frequency"), so it completely rejects the DC component from the sky.

There is a little more to it than what I'm writing, mainly because you need to not just detect the point source, but also estimate the star's position within the field of view (which IIRC is done by comparing the phase of the time-varying star signal to a reference point on the shutters).

[mikeash]

You're right, that is by far the more interesting part of the system. Thanks for elaborating, that's really cool.