I've got a linear array daughterboard prototyped, based on TAOS's TSL1401R 128-pixel sensor. It has an analog output, though, and is aimed more at the MoBoStamp-pe. I haven't tested it yet with my MoBoProp proto to see if the Propeller A/D can handle it. ('Too many things ahead of it in the queue!)
Another option would be TAOS's TSL3301-LF 102-pixel sensor, which has a serial digital output (10MHz max clock rate, 1MHz pixel rate). It might be a better match for the Prop, and there's a daughterboard on the horizon for it, too.
TAOS also makes sensors with more pixels, but the physical array sizes get quite large, which makes them unsuitable for use with inexpensive lenses.
It's been around since 1998 in various manifestations, starting with the TSL2301. TAOS has an EVM for it that uses a 50MHz SX. The "LF" (lead-free) version is new, though. The main difference between the now-discontinued '2301 and '3301 is the former's greater sensitivity in the near IR.
These are hot performers, BTW. I used the '2301 in an app, along with a 75MHz SX, that required extremely short exposures and high acquisition speeds. But if you have a fast A/D, the '1401 has eight times the top-end pixel rate.
correction, they didn't have it at RS (radio spares) last year [noparse]:)[/noparse]
I used a linear array from Hamamatsu to make a spectrometer that used modulated light, that was going straight into an analogue DAQ. These things definately work at rates that would allow for modulated light techniques and lots of other interesting stuff.
Imaging resolution inside of the eye is generally limited by the total amount of light you can use... Generally speaking, sensitivity is the key to success, which usually means that the larger the sensing element the better... which usually corresponds to fewer pixels in the same form factor. Spectral response is another key... we have structures that are of interest across the near-visual spectrum... uv gets filtered out... but not completely. IR is mostly transmitted but is diffusely absorbed and reflected. We use various combinations of dyes and filters to act as band pass selectors to great effect in the near IR and Yellow-Green. An imaging spectrometer is not currently available. The eye constantly moves... the books are wrong about the motion... the motion is both ballistic and stochastic.
Having said that...
Ignoring all of the issues on the Prop side of the equation... what would you choose?
For an imaging spectrometer...which I intend to hack from a fundus camera or as a second choice an ophthalmic microscope.
To do this, I just need to point a linear array at a rotating mirror and hang the whole thing off the imaging end of a camera. If I do it right, a focused image will move across the array. The key issues are sensitivity and resolution.
I'm looking for a cheap source of near-monochromatic light, not a laser. I have some ideas... but I am very open to suggestions.
I don't know squat about linear array sensors.
After your notes, I looked at TAOS... the problem is that I have 35mm image frames...so 400dpi means I get at best around 700 pixels, which eliminates the fundus camera idea but not the microscope.
The best choise at TAOS for me would be the tslr1406... analog.
I actually started the thread just to see if anyone was working on it right now... if they were, I'd do something else. I'm glad I asked because I was absolutely stunned and elated to find that Graham already built a linear array spectrometer[noparse]:)[/noparse] After one of Grahams recent notes, my curiosity got the best of me and I googled him on the web... Amazing. Is there anything you haven't built? We have similar interests but I have never even heard of an EDN... and you are building one already!!!
I'm also very appreciative of Phils comments.
At this point, learning is an important consideration. I have as much to learn on the analog side as on the digital side. I know that with analog we are possibly going to run into some limits until the nextProp... I wouldn't care. I would take it as far as I could and move over to the rotating mirror.
Comes down to two things, the NA of your optical system and the resolution of the camera. If you work out the size of the focal spot expected from your objective lens then multiply this by your magnification that spot should be sampled by 4 pixels of the camera in a 2D array so make that two pixels for a linear array. This is basically akin to sampling theorem. NA is much like f number its the sin on the maximum angle the light from the objective makes with the sample. The spot size is (1.22lamda)/NA.
Sensitivity:
It depends if you mean low light levels or low bit depths. The key for dealing with low bit depths is large pixels because the signal to noise ratio depends on the size of the quantum well or rather the number of photons it takes to "fill" the pixel. The more the better. I think SNR is propotional to the sqrt(N) where N is the number of photons. You can look at the camera specs if you want to find out quantum efficiency and the variation in sensitivity with wavelength.
One quetion, why not use a 2D array rather than a linear array and mirror?
You say you will have a 35mm image size but there is no reason why you cannot add magnification to your mirror system is there?
I actually had a version of my last note that answered your question... but the whole thing got a little too wordy.
I do have a plan to use a 2D array... two of them, in fact. But right now, it would make more sense to implement that with something other than a Prop.
I have longer range intentions, that would re-use this particular set up for mirror calibration purposes. But it is true that a very inexpensive ocular spectrometer would be an interesting adjunct in research ophthalmology. And I think that goal might be attainable with the prop. AND I would learn a great deal in the process.
Thanks again
Rich
ps... before I became ill with my little cancer... which I no longer am... I was building a double slit set up. You wouldn't have happened to do that already would you?
Also, with regard to 2D. At this point that would be the Propcam... about which Phil is pretty tight lipped right now. So, I don't want to ask him any questions. And I wouldn't want to put him off by starting work on another sensor. But I eventually will need complete control of a 2D sensor... and for example with the Propcam, even though the source code would be right there, in order to actually know what it is doing, I would have to know more than might be made available. And I would want that sort of information in the public domain so that other people could duplicate my work... exactly the opposite instinct of most business people.
In theory, I know that whatever I wish to do... I will eventually be able to do it with the Prop or with the nextProp. But in terms of practical realities, for 2D sensors... as you know the world is full of inexpensive, flexible, well documented and cheap examples. The only reason that I hesitate to go to one of those is that right now is that I have immersed myself in this architecture and I don't want to diffuse myself.
Again,
You do ask the best questions, which is a distinquishing characteristic of all true scientists.
In the early 90's, I had a conversation with a widely respected Russian scientist who told me that Russia was using spectroscopy to measure the oxygen content of the leafs of trees in large forrested areas. I have no idea how old the data was at that point.
Russian scientists were shocked to find large areas of forrests, which looked perfectly normal, but in fact were no longer producing normal amounts of oxygen, which meant that they were no longer consuming normal amounts of CO2. We didn't get into exactly what had caused the change, but I got the sense that it wasn't urban polution, such as carbon monoxide. So, if global warming is a fact... and if CO2 is related to that fact, exactly what happened to those Russian forrests and possibly others around the world might be a key to understanding the problem and fixing it.
rjo_
Posts: 1,825
Comments
I've got a linear array daughterboard prototyped, based on TAOS's TSL1401R 128-pixel sensor. It has an analog output, though, and is aimed more at the MoBoStamp-pe. I haven't tested it yet with my MoBoProp proto to see if the Propeller A/D can handle it. ('Too many things ahead of it in the queue!)
Another option would be TAOS's TSL3301-LF 102-pixel sensor, which has a serial digital output (10MHz max clock rate, 1MHz pixel rate). It might be a better match for the Prop, and there's a daughterboard on the horizon for it, too.
TAOS also makes sensors with more pixels, but the physical array sizes get quite large, which makes them unsuitable for use with inexpensive lenses.
-Phil
Graham
These are hot performers, BTW. I used the '2301 in an app, along with a 75MHz SX, that required extremely short exposures and high acquisition speeds. But if you have a fast A/D, the '1401 has eight times the top-end pixel rate.
-Phil
Thanks guys
I used a linear array from Hamamatsu to make a spectrometer that used modulated light, that was going straight into an analogue DAQ. These things definately work at rates that would allow for modulated light techniques and lots of other interesting stuff.
would it be rude to ask the app Phil?
Graham
You do ask the best questions.
Rich
It's never rude to ask, and I do wish I could talk about it, but I can't.
-Phil
Once again, thanks for your input.
Imaging resolution inside of the eye is generally limited by the total amount of light you can use... Generally speaking, sensitivity is the key to success, which usually means that the larger the sensing element the better... which usually corresponds to fewer pixels in the same form factor. Spectral response is another key... we have structures that are of interest across the near-visual spectrum... uv gets filtered out... but not completely. IR is mostly transmitted but is diffusely absorbed and reflected. We use various combinations of dyes and filters to act as band pass selectors to great effect in the near IR and Yellow-Green. An imaging spectrometer is not currently available. The eye constantly moves... the books are wrong about the motion... the motion is both ballistic and stochastic.
Having said that...
Ignoring all of the issues on the Prop side of the equation... what would you choose?
Rich
For an imaging spectrometer...which I intend to hack from a fundus camera or as a second choice an ophthalmic microscope.
To do this, I just need to point a linear array at a rotating mirror and hang the whole thing off the imaging end of a camera. If I do it right, a focused image will move across the array. The key issues are sensitivity and resolution.
I'm looking for a cheap source of near-monochromatic light, not a laser. I have some ideas... but I am very open to suggestions.
I don't know squat about linear array sensors.
After your notes, I looked at TAOS... the problem is that I have 35mm image frames...so 400dpi means I get at best around 700 pixels, which eliminates the fundus camera idea but not the microscope.
The best choise at TAOS for me would be the tslr1406... analog.
I actually started the thread just to see if anyone was working on it right now... if they were, I'd do something else. I'm glad I asked because I was absolutely stunned and elated to find that Graham already built a linear array spectrometer[noparse]:)[/noparse] After one of Grahams recent notes, my curiosity got the best of me and I googled him on the web... Amazing. Is there anything you haven't built? We have similar interests but I have never even heard of an EDN... and you are building one already!!!
I'm also very appreciative of Phils comments.
At this point, learning is an important consideration. I have as much to learn on the analog side as on the digital side. I know that with analog we are possibly going to run into some limits until the nextProp... I wouldn't care. I would take it as far as I could and move over to the rotating mirror.
I think I need higher DPI.
I'll look around...
Thanks
Rich
Rich
Comes down to two things, the NA of your optical system and the resolution of the camera. If you work out the size of the focal spot expected from your objective lens then multiply this by your magnification that spot should be sampled by 4 pixels of the camera in a 2D array so make that two pixels for a linear array. This is basically akin to sampling theorem. NA is much like f number its the sin on the maximum angle the light from the objective makes with the sample. The spot size is (1.22lamda)/NA.
Sensitivity:
It depends if you mean low light levels or low bit depths. The key for dealing with low bit depths is large pixels because the signal to noise ratio depends on the size of the quantum well or rather the number of photons it takes to "fill" the pixel. The more the better. I think SNR is propotional to the sqrt(N) where N is the number of photons. You can look at the camera specs if you want to find out quantum efficiency and the variation in sensitivity with wavelength.
One quetion, why not use a 2D array rather than a linear array and mirror?
You say you will have a 35mm image size but there is no reason why you cannot add magnification to your mirror system is there?
Graham
I actually had a version of my last note that answered your question... but the whole thing got a little too wordy.
I do have a plan to use a 2D array... two of them, in fact. But right now, it would make more sense to implement that with something other than a Prop.
I have longer range intentions, that would re-use this particular set up for mirror calibration purposes. But it is true that a very inexpensive ocular spectrometer would be an interesting adjunct in research ophthalmology. And I think that goal might be attainable with the prop. AND I would learn a great deal in the process.
Thanks again
Rich
ps... before I became ill with my little cancer... which I no longer am... I was building a double slit set up. You wouldn't have happened to do that already would you?
Also, with regard to 2D. At this point that would be the Propcam... about which Phil is pretty tight lipped right now. So, I don't want to ask him any questions. And I wouldn't want to put him off by starting work on another sensor. But I eventually will need complete control of a 2D sensor... and for example with the Propcam, even though the source code would be right there, in order to actually know what it is doing, I would have to know more than might be made available. And I would want that sort of information in the public domain so that other people could duplicate my work... exactly the opposite instinct of most business people.
In theory, I know that whatever I wish to do... I will eventually be able to do it with the Prop or with the nextProp. But in terms of practical realities, for 2D sensors... as you know the world is full of inexpensive, flexible, well documented and cheap examples. The only reason that I hesitate to go to one of those is that right now is that I have immersed myself in this architecture and I don't want to diffuse myself.
Again,
You do ask the best questions, which is a distinquishing characteristic of all true scientists.
Thanks,
Rich
Check out the arrays from Micron and Omnivision, you will probably find something you like but don't expect "full control" just what is designed in.
Graham
How important is spectroscopy?
In the early 90's, I had a conversation with a widely respected Russian scientist who told me that Russia was using spectroscopy to measure the oxygen content of the leafs of trees in large forrested areas. I have no idea how old the data was at that point.
Russian scientists were shocked to find large areas of forrests, which looked perfectly normal, but in fact were no longer producing normal amounts of oxygen, which meant that they were no longer consuming normal amounts of CO2. We didn't get into exactly what had caused the change, but I got the sense that it wasn't urban polution, such as carbon monoxide. So, if global warming is a fact... and if CO2 is related to that fact, exactly what happened to those Russian forrests and possibly others around the world might be a key to understanding the problem and fixing it.