Add draft of optimization article #17

joshwalawender · 2018-02-20T22:26:25Z

This is a draft of an article explaining the design motivation for PANOPTES and how we optimized for etendue. This was drafted months ago and needs a thorough review and editing.

oguyon · 2018-02-26T06:27:24Z

optimization.md

+
+Choosing DSLR cameras, for example, provides us a very low cost detector.  A comparable pixel count scientific CCD, might cost 5-10 thousand dollars, while a DSLR costs a few hundred dollars.  Pixel count is a limiting factor in how much information can be gathered in a single image, so having a large number of pixels is valuable.
+
+Finding a way to use the images from DSLRs [for accurate photometry](https://www.naoj.org/staff/guyon/09allskysurvey.web/56photometry.web/content.html) despite their [Bayer color filter array](https://en.wikipedia.org/wiki/Bayer_filter) was one of the early breakthroughs in the PANOPTES project.  This enabled us to use detectors which were an order of magnitude cheaper in __pixels per dollar__ than "scientific grade" CCD cameras.  


We need to move the content of this hyperlink to PANOPTES website

oguyon · 2018-02-26T06:30:39Z

optimization.md

+
+When PANOPTES was originally conceived, etendue was used as a guide, more specifically etendue per dollar (you're starting to see by now that "per dollar" is a PANOPTES theme).  Some rough etendue estimates by Olivier Guyon suggested that APS-C sized DSLR sensors combined with fast 85mm focal length camera lenses was a very good etendue per dollar option.
+
+A somewhat more detailed analysis was later attempted by PANOPTES team member Josh Walawender.  The PANOPTES analysis algorithm needs many stars in a single field of view in order to do the many comparisons needed to obtain accurate photometry, so he examined what lens (when used with our chosen DSLR) would provide the highest number of stars per image.  Using a model of the stellar population of the Milky Way Galaxy and assuming a telescope pointing somewhere near, but somewhat off of the galactic plane and combining that with an estimate of what the signal to noise would be for stars of a given brightness observed by various telescope aperture and focal lengths, he plugged in the specifications for a number of commercially available camera lenses and determined that the best combination would be an 85mm f/1.4 Rokinon lens (the same focal length and f-ratio combination guessed at by Olivier Guyon using the simpler etendue metric).


"attempted" sounds negative ... like it failed
-> replace by "performed"

oguyon · 2018-02-26T06:33:15Z

optimization.md

+
+A somewhat more detailed analysis was later attempted by PANOPTES team member Josh Walawender.  The PANOPTES analysis algorithm needs many stars in a single field of view in order to do the many comparisons needed to obtain accurate photometry, so he examined what lens (when used with our chosen DSLR) would provide the highest number of stars per image.  Using a model of the stellar population of the Milky Way Galaxy and assuming a telescope pointing somewhere near, but somewhat off of the galactic plane and combining that with an estimate of what the signal to noise would be for stars of a given brightness observed by various telescope aperture and focal lengths, he plugged in the specifications for a number of commercially available camera lenses and determined that the best combination would be an 85mm f/1.4 Rokinon lens (the same focal length and f-ratio combination guessed at by Olivier Guyon using the simpler etendue metric).
+
+This optimization of the PANOPTES lens (and camera) choice can be taken to additional levels of detail (for example examining the population of faint stars which by chance overlap with the stars you're measuring), but the commercial camera world is such that lenses (even from disparate manufacturers) come in quantized families.  One typically finds prime (non-zoom) lenses in one of a small number of focal length choices (e.g. 35mm, 50mm, 85mm, 135mm, 200mm, etc.), so once the estimate starts to narrow in on one particular focal length value, we can stop going in to more detail.  Zoom lenses are typically of a slower focal ratio (smaller aperture), so even though they would allow us to dial in an exact focal length, the reduced aperture would more than offset that advantage.


this paragraph, in my opinion, can be removed. It's not as important as the explanation above.

oguyon · 2018-02-26T06:43:15Z

optimization.md

+When PANOPTES was originally conceived, etendue was used as a guide, more specifically etendue per dollar (you're starting to see by now that "per dollar" is a PANOPTES theme).  Some rough etendue estimates by Olivier Guyon suggested that APS-C sized DSLR sensors combined with fast 85mm focal length camera lenses was a very good etendue per dollar option.
+
+A somewhat more detailed analysis was later attempted by PANOPTES team member Josh Walawender.  The PANOPTES analysis algorithm needs many stars in a single field of view in order to do the many comparisons needed to obtain accurate photometry, so he examined what lens (when used with our chosen DSLR) would provide the highest number of stars per image.  Using a model of the stellar population of the Milky Way Galaxy and assuming a telescope pointing somewhere near, but somewhat off of the galactic plane and combining that with an estimate of what the signal to noise would be for stars of a given brightness observed by various telescope aperture and focal lengths, he plugged in the specifications for a number of commercially available camera lenses and determined that the best combination would be an 85mm f/1.4 Rokinon lens (the same focal length and f-ratio combination guessed at by Olivier Guyon using the simpler etendue metric).
+


We could add a conclusion here:
In addition to providing nearly optimal etendue per dollar, the "many DSLR camera + lenses" solution offers two practical advantages for the PANOPTES project:

each unit is moderately priced, making it possible for citizen scientists and schools to build and deploy

exoplanet transits are identified by averaging measurements from many individual units, so noises and errors of individual units are efficiently averaged

jamessynge · 2018-08-19T17:52:19Z

optimization.md

+
+Finding a way to use the images from DSLRs [for accurate photometry](https://www.naoj.org/staff/guyon/09allskysurvey.web/56photometry.web/content.html) despite their [Bayer color filter array](https://en.wikipedia.org/wiki/Bayer_filter) was one of the early breakthroughs in the PANOPTES project.  This enabled us to use detectors which were an order of magnitude cheaper in __pixels per dollar__ than "scientific grade" CCD cameras.  
+
+Once cameras with a Bayer filter array are an option for us, the current status of the market means that consumer grade DSLRs with APS-C sized detectors offer a good combination of price and pixel count.  The Canon Rebel series was what we chose initially, but the choice of Canon brand cameras is somewhat arbitrary:  in principle, any DSLR brand and model with a similar pixel count and cost which also provides access to unprocessed, "raw" images would work similarly well.


Regarding "any DSLR brand and model", I'd be inclined to change this to "any interchangeable-lens camera". I expect that we'll soon (next 18 months) be testing out non-DSLR cameras.

jamessynge · 2018-08-19T17:55:38Z

optimization.md

+
+Once cameras with a Bayer filter array are an option for us, the current status of the market means that consumer grade DSLRs with APS-C sized detectors offer a good combination of price and pixel count.  The Canon Rebel series was what we chose initially, but the choice of Canon brand cameras is somewhat arbitrary:  in principle, any DSLR brand and model with a similar pixel count and cost which also provides access to unprocessed, "raw" images would work similarly well.
+
+The choice of consumer DSLR cameras also means we have access to consumer camera lenses.  In astronomical surveys, a simple metric to help evaluate survey speed is called _etendue_.  Etendue is intended to be a "rule of thumb" level metric, not a complete analysis.  Etendue is the product of the survey telescope's aperture (i.e. area in square meters) and its field of view (i.e. area in square degrees).  In simple terms, one might imagine that to survey the night sky at a particular depth (to a particular faintness limit) we might choose a big telescope with a small field of view.  This would mean we take a short exposure of each area of the sky (the large aperture means we collect more light and so need less exposure time), but we need to take many different exposures to cover a large area.  Alternatively, we might choose a smaller telescope with a larger field of view.  This means we would take longer individual exposures, but we would need fewer of them to cover the same area.  Those might work out to be the same total survey efficiency.  The concept of etendue tries to capture that balance.


You might link to these two:

https://en.wikipedia.org/wiki/Etendue

https://www.mikewoodconsulting.com/articles/Protocol%20Winter%202012%20-%20Etendue.pdf

So that folks can learn more.

Add draft of optimization article

36eb9dd

oguyon reviewed Feb 26, 2018

View reviewed changes

Updates based on Olivier’s comments.

74b52a6

jamessynge approved these changes Aug 19, 2018

View reviewed changes

Updates based on comments from James.

55dc9ef

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add draft of optimization article #17

Add draft of optimization article #17

joshwalawender commented Feb 20, 2018

oguyon Feb 26, 2018

oguyon Feb 26, 2018

oguyon Feb 26, 2018

oguyon Feb 26, 2018

jamessynge Aug 19, 2018

jamessynge Aug 19, 2018


		Choosing DSLR cameras, for example, provides us a very low cost detector. A comparable pixel count scientific CCD, might cost 5-10 thousand dollars, while a DSLR costs a few hundred dollars. Pixel count is a limiting factor in how much information can be gathered in a single image, so having a large number of pixels is valuable.

		Finding a way to use the images from DSLRs [for accurate photometry](https://www.naoj.org/staff/guyon/09allskysurvey.web/56photometry.web/content.html) despite their [Bayer color filter array](https://en.wikipedia.org/wiki/Bayer_filter) was one of the early breakthroughs in the PANOPTES project. This enabled us to use detectors which were an order of magnitude cheaper in __pixels per dollar__ than "scientific grade" CCD cameras.


		When PANOPTES was originally conceived, etendue was used as a guide, more specifically etendue per dollar (you're starting to see by now that "per dollar" is a PANOPTES theme). Some rough etendue estimates by Olivier Guyon suggested that APS-C sized DSLR sensors combined with fast 85mm focal length camera lenses was a very good etendue per dollar option.

		A somewhat more detailed analysis was later attempted by PANOPTES team member Josh Walawender. The PANOPTES analysis algorithm needs many stars in a single field of view in order to do the many comparisons needed to obtain accurate photometry, so he examined what lens (when used with our chosen DSLR) would provide the highest number of stars per image. Using a model of the stellar population of the Milky Way Galaxy and assuming a telescope pointing somewhere near, but somewhat off of the galactic plane and combining that with an estimate of what the signal to noise would be for stars of a given brightness observed by various telescope aperture and focal lengths, he plugged in the specifications for a number of commercially available camera lenses and determined that the best combination would be an 85mm f/1.4 Rokinon lens (the same focal length and f-ratio combination guessed at by Olivier Guyon using the simpler etendue metric).


		A somewhat more detailed analysis was later attempted by PANOPTES team member Josh Walawender. The PANOPTES analysis algorithm needs many stars in a single field of view in order to do the many comparisons needed to obtain accurate photometry, so he examined what lens (when used with our chosen DSLR) would provide the highest number of stars per image. Using a model of the stellar population of the Milky Way Galaxy and assuming a telescope pointing somewhere near, but somewhat off of the galactic plane and combining that with an estimate of what the signal to noise would be for stars of a given brightness observed by various telescope aperture and focal lengths, he plugged in the specifications for a number of commercially available camera lenses and determined that the best combination would be an 85mm f/1.4 Rokinon lens (the same focal length and f-ratio combination guessed at by Olivier Guyon using the simpler etendue metric).

		This optimization of the PANOPTES lens (and camera) choice can be taken to additional levels of detail (for example examining the population of faint stars which by chance overlap with the stars you're measuring), but the commercial camera world is such that lenses (even from disparate manufacturers) come in quantized families. One typically finds prime (non-zoom) lenses in one of a small number of focal length choices (e.g. 35mm, 50mm, 85mm, 135mm, 200mm, etc.), so once the estimate starts to narrow in on one particular focal length value, we can stop going in to more detail. Zoom lenses are typically of a slower focal ratio (smaller aperture), so even though they would allow us to dial in an exact focal length, the reduced aperture would more than offset that advantage.


		Finding a way to use the images from DSLRs [for accurate photometry](https://www.naoj.org/staff/guyon/09allskysurvey.web/56photometry.web/content.html) despite their [Bayer color filter array](https://en.wikipedia.org/wiki/Bayer_filter) was one of the early breakthroughs in the PANOPTES project. This enabled us to use detectors which were an order of magnitude cheaper in __pixels per dollar__ than "scientific grade" CCD cameras.

		Once cameras with a Bayer filter array are an option for us, the current status of the market means that consumer grade DSLRs with APS-C sized detectors offer a good combination of price and pixel count. The Canon Rebel series was what we chose initially, but the choice of Canon brand cameras is somewhat arbitrary: in principle, any DSLR brand and model with a similar pixel count and cost which also provides access to unprocessed, "raw" images would work similarly well.


		Once cameras with a Bayer filter array are an option for us, the current status of the market means that consumer grade DSLRs with APS-C sized detectors offer a good combination of price and pixel count. The Canon Rebel series was what we chose initially, but the choice of Canon brand cameras is somewhat arbitrary: in principle, any DSLR brand and model with a similar pixel count and cost which also provides access to unprocessed, "raw" images would work similarly well.

		The choice of consumer DSLR cameras also means we have access to consumer camera lenses. In astronomical surveys, a simple metric to help evaluate survey speed is called _etendue_. Etendue is intended to be a "rule of thumb" level metric, not a complete analysis. Etendue is the product of the survey telescope's aperture (i.e. area in square meters) and its field of view (i.e. area in square degrees). In simple terms, one might imagine that to survey the night sky at a particular depth (to a particular faintness limit) we might choose a big telescope with a small field of view. This would mean we take a short exposure of each area of the sky (the large aperture means we collect more light and so need less exposure time), but we need to take many different exposures to cover a large area. Alternatively, we might choose a smaller telescope with a larger field of view. This means we would take longer individual exposures, but we would need fewer of them to cover the same area. Those might work out to be the same total survey efficiency. The concept of etendue tries to capture that balance.

Add draft of optimization article #17

Are you sure you want to change the base?

Add draft of optimization article #17

Conversation

joshwalawender commented Feb 20, 2018

oguyon Feb 26, 2018

Choose a reason for hiding this comment

oguyon Feb 26, 2018

Choose a reason for hiding this comment

oguyon Feb 26, 2018

Choose a reason for hiding this comment

oguyon Feb 26, 2018

Choose a reason for hiding this comment

jamessynge Aug 19, 2018

Choose a reason for hiding this comment

jamessynge Aug 19, 2018

Choose a reason for hiding this comment