Cheapest possible "toy" spectrometer? $323

[andy]

I came up with this design for what I imagine would be the cheapest possible raman design that I can think of.

These were my ideas:

• Use a 90 degree scattering to avoid dichroic lens
• 1" lens to get as much light as possible [potentially even a larger cuvette path length]?
• Use plano lens to focus light even though it doesn't perform well
• Use a Longpass Colored Glass Filters at a higher then normal Cut-on wavelength
• Use USB Spectrometer from Thunder Optics which is very cheap (though I think this design would require a fiber spectrometer unless the usb input can be placed in the output beam path).
• Use a random laser from ebay

Interested to hear if this design would work.

I got the idea from the attached poster but I used a different setup

[Luc]

It's a question of tradeoff as always.

But here are a few things to consider:

• non back-scattering setups require more alignments per sample. with solid it might be ok if you have a lot of subsurface scattering but with liquid it's going to be really painful. My first liquid cuvette was actually a 90° one! I quickly changed to backscattering.
• you can work without a dichroic. Some older setups use a plane window with a dot mirror at the center. The laser reflects on the dot but then the Raman signal passes around it (the dot part is then obscured).
• to be tested what's the actual effect of single PCX lens. I expect you'll get attenuation of some wavelengths but maybe we can live with it
• although some people show Raman spectrometer built from ILX511, TCD1304 or equivalent sensors, they are super noisy.
• all my previous comments about ebay lasers they can work but it's difficult to scale

So I think it's possible to make it much cheaper, but to be checked at which performance/robustness tradeoff.

[andy]

Yeah with this design/concept the idea would be for the tradeoffs to always be in favor of cost and see how cheap you could go and still see something. Currently on the market there is not a sub-$1000 raman system.

The 90 degree would produce rays in all directions, so is it because you would have to collimate the light as shown in that presentation? The paper mentioned below also suggested backscattering for alignment reasons.

Currently the lowest cost spectrometer I could find was that $150 webcam based one which is based on a CMOS but is very low resolution and might not even have a high enough SNR to see something. Some other people have suggested using smart phone cameras. I am not sure that would work unless you had a powerful enough laser but that might be unsafe in the environments this would be used in.

Emmanuel, Neethu, et al. "Fabricating a low-cost Raman spectrometer to introduce students to spectroscopy basics and applied instrument design." (2021): 2109-2116.

[alexose]

Something I've been wondering about: How much can software help with a noisy sensor? Could we run thousands of readings through an denoising algorithm and get a stable result?

[andy]

Something I've been wondering about: How much can software help with a noisy sensor? Could we run thousands of readings through an denoising algorithm and get a stable result?

I don't think so. This would work if the noise was predicable (think headphones that hear fan noise which has a specific pattern, the headphones can pick up on that pattern and subtract it). But dark noise is truly random so can not be dealt with in this way. I attached a page from this book:

McCreery R. L. (2000). Raman spectroscopy for chemical analysis. John Wiley & Sons.

[alexose]

Thanks for providing this, Andy. I found some pretty informative videos regarding Raman noise that the Bruker corporation put out:

https://www.youtube.com/watch?v=c1xS_oC1arM
and
https://www.youtube.com/watch?v=IQMr-xflXlI

(In fact, this whole video series is excellent. Looking forward to watching all of them)

I suppose I still don't understand why it isn't possible to average tons of spectra together. If the noise is truly random, it should be possible to increase the collection time and get a stronger SnR. The trade off would just end up being speed for clarity. There would be a point where it's not worth trying, however... nobody wants to run Raman samples overnight. (Or do they?)

[andy]

You can definitely add up spectra to decrease the noise. The collection time does increase the SNR up until the saturation point of the sensor or you can average several readings together.

In Luc's software, he also adds up pixel columns whereas in in CCD spectrometers they use a lens to push the photons all into one line.

Also you can reduce noise through cooling the sensor

[alexose]

I would love to get the ball rolling on one of these builds.

The first step seems to be settling on a suitable laser and sensor. I'm still not sure how to pick a good low-cost laser, but presumably someone has done this research already?

Regarding the sensor I'd be very tempted to continue on the path that Esben Rossel built and use the TCD1304. They are super cheap, well suited to DIY devices, and have plenty of videos explaining how to use them (https://www.youtube.com/watch?v=xcpMxeOgUcM and https://www.youtube.com/watch?v=6aPIwcFicFU)

Let's not rule out active cooling if it means we can improve performance. Peltier chips are not expensive, nor would they add too much design complexity.

[andy]

I've lost a little bit of enthusiasm for this. In my experimentation it turns out Luc's choices mattered more then I thought.

You need a pretty sensitive and high resolution spectrometer. You need a narrowwidth laser, and the lenses have to be aligned just right to capture the light.

Without these ingredients you can see some spectra but its barely visible over the noise floor and the spectra is smoothed out to the point of not really being able to use it for non targeted analysis. Though you could probably measure rough concentrations if the analyte is over 10%.

The way forward would probably be to Luc's suggestion of manufacturing a probe through CNCing then using external tooling to align the lenses with glue to set them. I just don't have the skill or capability to do that. Hubs can do the cncing cheaply but you have to put in a large order.

I have learned through running my store that its expensive to sell these as well, once you account for taxes, refunds, shipping, transaction costs, time to build, maintaining a store, paperwork, working with customers, and so on. That adds 20% or so

Add that up and the original OpenRaman DIY starts to look like a good deal

[alexose]

Thanks for your honesty, Andy-- Even deciding that it's not possible I think is a useful result! I'm willing to accept that we can't do better than Luc's original design, but I'm finding this exercise useful in helping me understand what the tradeoffs are in Raman spectroscopy.

I tracked down Luc's comment regarding eBay lasers:

Cheap e-bay lasers are not an option in that regards because they are a lottery: very few ones will have correct performances but most of them will be crappy or even dangerous to operate

I've also found this to be true. The cheap CO2 laser I use for laser cutting seems very, shall we say, handmade. If it didn't operate inside of a large metal box, I would be too nervous to use it.

However, sources suggest that Luc has perhaps found a quality laser candidate in the $150 range. If it pans out, I think the dream of a super-low-cost OpenRAMAN spec lives on.

The way forward would probably be to Luc's suggestion of manufacturing a probe through CNCing then using external tooling to align the lenses with glue to set them. I just don't have the skill or capability to do that. Hubs can do the cncing cheaply but you have to put in a large order.

I have a CNC mill and would be happy to make parts as needed. I'm limited to aluminum, but it could work. I'd also be happy to experiment with designing tooling to align the lenses. This seems like a potential task for SLA 3D printing, but I admit I don't understand the problem space well enough to offer solutions just yet. Perhaps sometime we can brainstorm ideas on how to perfect the alignment process.

This would just leave the final piece, which finding the correct sensor. Wisdom of the masses indicates that the TCD1304 will work, but perhaps we can do better. Can we design an SMA input for this sensor? Perhaps this already exists on AliExpress... https://www.aliexpress.us/item/22518327 ... 4itemAdapt

It's not clear to me how to compare a chip like the TCD1304 to a purpose-built spectrometer like the C13054MA mentioned in our earlier email thread. (https://www.hamamatsu.com/content/dam/h ... c1233e.pdf). Maybe apples to oranges?

At the end of the day, it's a question of how many Raman-scattered photons reach our sensor, and how likely it is that those photos interact with the CCD itself. There are some software tricks that we can use to improve the results, but physics is king in this application.