I have an interest in developing Super Resolution on the OpenFlexure system. fairsim.org (https://www.fairsim.org/) seems like a possible method, looking for some other’s thoughts. I am thinking of using my 10x / 0.25 objective. The Raleigh Limit assuming a green LED (550nm) illumination is 1.34um. So I should be able to get to a resolution of about 1.34/2 or 0.67um, assuming a 2x improvement in resolution. I have a 40x/0.60 objective that gives 0.559um, so I can check the resolution with the 40x objective.The advantage of using the 10x objective is a much wider field of view. I have a number of tools for measuring the resolution, sharp edge, various targets, 10um beads, 1umbeads and a few diffraction gratings. I can see the limits of resolution when measuring the lines/mm on the gratings and when measuring the various beads. But is the fairsim method the best way to improve the resolution or is there a better way?
Did you attempt to use more than three rotations? Just a personal opinion - it feels like the centre will always have more information - the further away will have less due to the angular/pixel grid alignment and gaps appearing between the rotations.
It should be able to calculate the best method of obtaining information through sub-sampling techniques and plot vs accuracy of information extracted vs time to acquire. I’m not sure if that’s been done before but it should be easy enough.
I made an attempt at this a few years back, as a masters project using the UC2 structured illumination implementation together with (I think) the FairSIM plugin. It never worked quite as well as we hoped, but I suspect that’s down in no small part to only having got it working a couple of weeks before said student finished, and not having carried the project on further.
If you come up with a nice way of coupling the structured illumination into an OFM, I see no reason it shouldn’t work - though it will definitely be easier with an infinite-conjugates microscope objective.
@rooksoup designed a plate that allows you to bolt an OpenFlexure body onto four posts of the sort typically found in optics labs. That might make it easy to prototype something, if you have access to an optics lab. The design is currently in a branch on GitLab, I should be able to dig it out if it would be a help.
In theory the infinite & backfocus focusing could use a Bahtinov mask.
With a setup using infinite conjugate - could you not use a mini 7 or 9 position filter wheel? Five filters taken up by single slits* rotated for the fairsim, a mini bahtinov and naturally an open position for full visual imaging?
I can see the accuracy (thermal expansion/warping) demanding a more stable filter plate material but it could be prototyped in plastic. From experience it would need calibration on selecting each position, even with registration notching of the wheel.
* I say ‘single slit’ but a configuration to prevent diffraction.
Probably a little more complex but it would make a neat, compact solution.
When we did it, we were projecting two spots into the back aperture, which needed to be coherent with each other. Doing that using just an aperture would be pretty lossy, though if you’ve enough spare power it could perhaps work. As you say, it might be tricky to get precise enough positioning of a filter wheel.
The UC2 design used a DMD as a programmable diffraction grating, which is ~5% efficient - not amazing, but quite convenient because there’s no mechanical repositioning between images.
Either way, a nice implementation of a filter wheel for the microscope would be really useful. Even better if it’s mechanically precise enough to automatically swap filter cubes for fluorescence. That’s something quite a few people have wanted to do for ages, but it’s enough engineering that nobody has taken it to the point where it “just works” for everyone.
You could do a filter wheel that has basic registration but then has pin and and servo V block to provide accurate registration lock as an operation extension, it could be made into a 3D lock that also ensures a Z access registration too. That I suspect would be accurate enough.
Most of the telescope filterwheels have a Hall effect sensor to detect filter 1 as not to contaminate any light path. It would be easy enough then to have a servo to move it with a wire spring to register in the basic form, and a an optional servo driven V-lock block for an advanced form to replace the wire spring registration. Usually the startup sequence does a full rotation to defect filter 1 and optionally count a full resolution with filter positions to calibarate. The V-lock works to centralise the filter position pin in the V so in reality the rotation doesn’t need to be super precise and the repeat registration is then pretty decent and it means the servos don’t have to have power which helps prevent fluctuations from being amplified by the radius of the wheel.
The wheel can then be designed for cubes, screw in filters or bare filters.
I feel we’re probably diverting the original thread - I’ll see if there’s a filter wheel thread.
My thought was to first just do it manually, the 3 angles, but after further reading, the 3 translations are small so a manual adjustment doesn’t seem feasable. So in the original SIM application, was there an optical way to shift the image?
I assume the idea of using a filterwheel to get each position would involve 9 positions or 3 manual and 3 automated?
I was thinking the 5 rotations by slit orientation where each filter position has a single slit orientation.
The filter wheel would go on the laser side. However the same design could be used in theory as the same mounting is used. That would give you the rotations, but you’d need more - including a mechanism for phase. That is quite complex due to needing to support phases of the laser wave length. There’d need to be a replacement optics component and parts under the stage to house any mirroring. I think given the complexity the housing would extend out the front of the scope. The fiterwheel could be mounted vertically on the laser input part of that.
So - same modular component used, but need a fairsim implementation in a box that then uses the pi HQ camera or another camera. The more optics the more expense. The stage would be free to move (which then allows the green pixels to be used in the sensor to best effect although it add complexity compensating, so a mono camera would be better overall). The biggest concern I would be the wavelength phasing and plastic stability for repeatability.
I’ve consulted this optical path: Video-rate multi-color structured illumination microscopy with simultaneous real-time reconstruction | Nature Communications
Humbly I’m not an optics expert so would welcome further input.
I am still trying to simplify the process. I did read about using a DMD (Digital Micromirror Device) to generate the rotation and shifts. They used a repurposed DLP300schip + DLPC1438 controller. I don’t like the idea of reporposing, but the electronics would be a challenge. The DMD is fairly inexpensive. OS-SIM | openUC2 Docs