From 9e452d1ef02b78bf2fafae8152a6f839b200c911 Mon Sep 17 00:00:00 2001 From: Noah Smith Date: Sun, 19 May 2024 02:53:17 +0200 Subject: [PATCH] starid93 --- _drafts/starid77.md | 24 ------------------------ _drafts/starid93.md | 20 ++++++++++++++++++++ _posts/2024-05-20-starid93.md | 20 ++++++++++++++++++++ 3 files changed, 40 insertions(+), 24 deletions(-) delete mode 100644 _drafts/starid77.md create mode 100644 _drafts/starid93.md create mode 100644 _posts/2024-05-20-starid93.md diff --git a/_drafts/starid77.md b/_drafts/starid77.md deleted file mode 100644 index fd8643f..0000000 --- a/_drafts/starid77.md +++ /dev/null @@ -1,24 +0,0 @@ ---- -layout: post -title: "starid77" -categories: aerospace starid ---- -[paper 1977](https://statespace.dev/docs/papers/1977%20junkins.pdf){:target="_blank" rel="noopener"} - -this paper is a time-capsule from the seventies, capturing the era when integrated-circuit 'chips' were fully coming on-line, at least for state of the art aerospace hardware, in two separate domains - microprocessors or 'systems on a chip', and charge-coupled device imagers or 'digital cameras'. - -the original was published in 77, and this is a pdf of an extended version from 78. 77 is a reminder of fortran77. fortran was born in 57, the same year as sputnik, and this paper is a window into the twentieth year of the space age and the fortran era. - -the star recognition algorithm is limited by the first-generation hardware, and nowhere near tackling the lost in space problem. it assumes an attitude estimate is available and sufficient to limit the pattern match to a 'sub-catalog' of stars, in other words a small region of the sky. it's an excellent historical review of the types of work done from the fifties thru the seventies. - -the discussion of the [fairchild semiconductor](https://en.wikipedia.org/wiki/Charge-coupled_device#History) ccd camera is of special interest. fairchild played an important role in the research and development phase of digital imaging. - -_Several companies, including Fairchild Semiconductor, RCA and Texas Instruments, picked up on the invention and began development programs. Fairchild's effort, led by ex-Bell researcher Gil Amelio, was the first with commercial devices, and by 1974 had a linear 500-element device and a 2D 100 × 100 pixel device._ - -the paper describes a 488 x 380 pixel device, so possibly fairchild's second generation ccd imager. the larger story around the transistor, shockley, fairchild, chips, and the birth of intel is too far afield for project starid, at least for now. - -beyond the purely technical history, this paper also captures a certain 'spirit of the times'. one element of the fortran era was the cold war sense of urgency around engineering and science. some of that legacy survives today in enclaves like the national science foundation, the [berkeley physics course](https://en.wikipedia.org/wiki/Berkeley_Physics_Course), the [feynman lectures on physics](https://en.wikipedia.org/wiki/The_Feynman_Lectures_on_Physics), and the remaining engineers and scientists from that generation. - -by the eighties and nineties, the fortran generation were becoming the teachers and role models. this paper captures a part of what that was like. it was created on a manual typewriter, and then poorly xeroxed multiple times. at the same time, it was discussing dramatic digital advances. it's on the divide between the analog and digital worlds. - -john junkins became a leading professor at texas a&m university and interacted directly with byron tapley, bob schutz, george born, the university of texas at austin center for space research, and the university of colorado at boulder laboratory for atmospheric and space physics. parts of that enter the project starid story, and the intention is to explore that over time. diff --git a/_drafts/starid93.md b/_drafts/starid93.md new file mode 100644 index 0000000..ce733ff --- /dev/null +++ b/_drafts/starid93.md @@ -0,0 +1,20 @@ +--- +layout: post +title: "starid93" +categories: aerospace starid +--- +[paper 1993](https://statespace.dev/docs/papers/1993%20liebe.pdf){:target="_blank" rel="noopener"} + +this has been the earliest known paper that directly tackles the lost in space problem. the author doesn't use the term 'lost in space'. this could be a result of being outside the sphere of american aerospace jargon, or it could be evidence that the lost in space problem wasn't yet widely recognized as practical. + + the topic of 'triangular features' is a clue that the lost in space problem is the subject. roughly speaking, if the lost in space problem is involved, the concept of star triangles will necessarily appear. and conversely, if star triangles don't appear, there's probably a finite attitude uncertainty involved so that they're not needed. + +a trio of three-dimensional unit vectors, or equivalently their three dot products, are a fundamental representation of a star triangle. surprisingly, unit vectors and dot products aren't directly invoked in this paper. a 'spherical angle' does make an appearance, along with a pair of 'distance angles', so a shadow of three dimensionality is present. indications are that the thinking remained essentially two dimensional, concerning camera images and basic astronomical-style practices. in other words, this paper is a good example of an 'astronomical approach', as opposed to an 'aerospace approach'. + +another aspect of the lost in space problem is that the runtime sky information has to cover the entire sky and be searchable as a whole. the sky necessarily has to be an indivisible whole. narrowing search to subregions of the sky defeats the purpose of the lost in space problem, which is to cope with an infinite attitude uncertainty. + +what this paper does is build a searchable set whose members are triplets of floats. each of the the floats in a triplet is a dot product. the first is between stars a and b. the second and third are between stars b and c, and stars c and a. star b is the 'nearest neighbor' to star a, and star c is the 'second nearest neighbor'. + +this effectively increases the 'hi-fi' high resolution requirements on the star tracker and the measurements it feeds into the starid algorithm. essentially, the 'nearest neighbor' and 'second nearest neighbor' have to be highly accurate and reliable. this is the opposite of the 'lo-fi' objectives of project starid, where the star tracker and its outputs are as primitive as possible, on the level of a [children's camera from the eighties](https://en.wikipedia.org/wiki/PXL2000). lo-fi forces the algorithms to work with global features, rather than high accuracy localized features such as ordering of neighbors and accurate numerical values. + +for a concrete example, assume the true triplet of floats for a star in some artificial units is (2173.89, 937.49, 2366.03). the measured triplet from a hi-fi star tracker is (2173.7, 937.6, 2366.0). for a lo-fi star tracker, with pixelation 'rounding' effects, the measured triplet is (2, 1, 2). the lo-fi measurement might be effective for distinguishing the star's signature as long as there's only a handful of other stars, but as the number of stars grows, more stars will have the same lo-fi signature. the lo-fi star tracker forces recognition algorithms to use additional information. diff --git a/_posts/2024-05-20-starid93.md b/_posts/2024-05-20-starid93.md new file mode 100644 index 0000000..faa3b5b --- /dev/null +++ b/_posts/2024-05-20-starid93.md @@ -0,0 +1,20 @@ +--- +layout: post +title: "starid93" +categories: aerospace starid +--- +[paper 1993](https://statespace.dev/docs/papers/1993%20liebe.pdf){:target="_blank" rel="noopener"} + +this has been the earliest known paper that directly tackles the lost in space problem. the author doesn't use the term 'lost in space'. this could be a result of being outside the sphere of american aerospace jargon, or it could be evidence that the lost in space problem wasn't yet widely recognized as practical. + + the topic of 'triangular features' is a clue that the lost is space problem is the subject. roughly speaking, if the lost in space problem is involved, the concept of star triangles will necessarily appear. and conversely, if star triangles don't appear, there's probably a finite attitude uncertainty involved so that they're not needed. + +a trio of three-dimensional unit vectors, or equivalently their three dot products, are a fundamental representation of a star triangle. surprisingly, unit vectors and dot products aren't directly invoked in this paper. a 'spherical angle' does make an appearance, along with a pair of 'distance angles', so a shadow of three dimensionality is present. indications are that the thinking remained essentially two dimensional, concerning camera images and basic astronomical-style practices. in other words, this paper is a good example of an 'astronomical approach', as opposed to an 'aerospace approach'. + +another aspect of the lost in space problem is that the runtime sky information has to cover the entire sky and be searchable as a whole. the sky necessarily has to be an indivisible whole. narrowing search to subregions of the sky defeats the purpose of the lost in space problem, which is to cope with an infinite attitude uncertainty. + +what this paper does is build a searchable set whose members are triplets of floats. each of the the floats in a triplet is a dot product. the first is between stars a and b. the second and third are between stars b and c, and stars c and a. star b is the 'nearest neighbor' to star a, and star c is the 'second nearest neighbor'. + +this effectively increases the 'hi-fi' high resolution requirements on the star tracker and the measurements it feeds into the starid algorithm. essentially, the 'nearest neighbor' and 'second nearest neighbor' have to be highly accurate and reliable. this is the opposite of the 'lo-fi' objectives of project starid, where the star tracker and its outputs are as primitive as possible, on the level of a [children's camera from the eighties](https://en.wikipedia.org/wiki/PXL2000). lo-fi forces the algorithms to work with global features, rather than high accuracy localized features such as ordering of neighbors and accurate numerical values. + +for a concrete example, assume the true triplet of floats for a star in some artificial units is (2173.89, 937.49, 2366.03). the measured triplet from a hi-fi star tracker is (2173.7, 937.6, 2366.0). for a lo-fi star tracker, with pixelation 'rounding' effects, the measured triplet is (2, 1, 2). the lo-fi measurement might be effective for distinguishing the star's signature as long as there's only a handful of other stars, but as the number of stars grows, more stars will have the same lo-fi signature. the lo-fi star tracker forces recognition algorithms to use additional information.