[S] Small and cheap spectrometer calibration source?

[harerod]

Since this is my first post to this section, I should introduce myself. I am a freelance electronics engineer. My work mostly revolves around microcontroller applications for various fields, e.g. medical and lab diagnostics. A couple of years ago, forum member branadic introduced me to EEVBLOG. He did this whilst showing great patience in discussing the basics of metrology via email and telephone to a person whose normal work considers 0.1% accuracy as the ACME of precision.


Last year I worked with Hamamatsu sensors and came across their C12880MA. I liked the device and decided to treat myself by adding a small spectrometer to my lab. I could have bought their eval board, but rather chose to roll my own, using an AD7980 ADC plus STM32F4. The design works quite nicely, at the moment I am writing the Windows software.


Today's question is about a cheap calibration source for the Hamamatsu C12880MA spectrometer module. Although the spectral range of this sensor is given with 340..850nm, even in the visible range the sensitivity varies by about 30%.
My current approach is to use white LEDs and compare their spectrum to the sensor output. I expect quite a bit of variation between devices, though.

I see several ways to tackle this issue:
- could anybody point me towards a small and cheap calibration source?
- would anybody with a calibrated spectrometer be so kind as to run an analysis on a couple of LEDs provided by me? (C512A-WNN / SPMWHT541MP5WATMS0, both available from Digikey)

As far as accuracy is concerned - I have no idea what is possible and at which cost. Anything better than 10% should be great, I guess.

Thanks in advance!

Edit210430: Link to the sensor: https://www.hamamatsu.com/eu/en/product/type/C12880MA/index.html
Edit210510: It seems that I haven't specified my problem clear enough for readers not familiar with the C12880MA: each module comes with spectral calibration data. However, the sensitivity data is only given as a "typical" curve in the datasheet. So I am looking for a fairly simple broad emission source to calibrate the spectral sensitivity. The spectral resolution of the sensor is only 15..12nm. After a look at the actual datasheet, my request might not appear that insane after all.

[Harfner]

Not sure whether it is relevant, but in my experience, white LEDs (the blue LED with yellow phosphor variant) show very much variance between individual parts of the same type. It may be even that the relative intensity between 'yellow peak' and 'blue peak' is completely different between individual parts.
For intensity calibration, why take a halogen bulb? There is a temperature marked on the box, which is quite accurate. At least that was suggested to me by Ocean Insight for calibration of one of their spectrometers.

[dietert1]

In 2016 i also made an effort to calibrate our spectrometer from aseq-instruments. They made a 500 to 1100 nm spectrometer for us, useful for checking pulse oximetry probes (LEDs at 660 nm and 880 nm)
In the end i found a standard neon indicator bulb useful for wavelength calibration and i cross-checked with an Oriel calibration lamp we have with a 6060 supply. I found that the original calibration had a scale error and was off up to 15 nm. Or the instrument was hit during shipment, who knows.
For intensity calibration i used daylight spectra that i captured with a small hole camera in front of the fiber. I remember that i had to repeat those measurements several days until i had a reasonably blue sky with similar to expected results. I found skylight spectra in publications on the web. The CCD sensitivity curve that i received from the seller of the spectrometer was helpful, too (published by Thorlabs).

Regards, Dieter

[jpanhalt]

Since you like to "roll your own," have you considered using chemical actinometry to calibrate intensity?

It's been a long time for me, but that is what I did in the late 60's as needed for my graduate research.  I used ferrioxalate as the detector and phenanthroline as the indicator.  There are lots of options, here's a short review: https://www.nature.com/articles/s41598-018-23735-2

loc. cit.
A widely accepted actinometer is the Hatchard-Parker actinometer (ferrioxalate) which is characterized by a quantum yield higher than 0.96. It involves the reduction of K3Fe(C2O4)3·3H2O in the presence of light, and the resulting Fe2+ ions are quantified by the complexation with 1,10-phenanthroline.

 

Wikipedia ("actinometry") also has some references.

[JohnG]

Perhaps these may provoke some thinking along these lines. See attachment and link:


https://iopscience.iop.org/article/10.1088/0143-0807/31/4/022 (sorry it's behind a paywall)

Cheers,
John

[richnormand]

Wavelength:
I use the 632.8nm line from a HeNe laser. Old gas HeNe lasers are easy to get.

For the low end I use a small Neon bulb as it has many lines that are well defined in spectroscopy catalogs. You can also get Neon lamp with specific isotope selection. At the short end I use a fluorescent light from the ceiling! The blue and UV mercury lines shine through the wall phospor.
Since this is a gas discharge the lines have a well defined wavelength compared to LEDs.

Finally I have several calibration hollow cathode lamps made for that purpose. I got them used from ebay for a low cost. They might be "past due" for lab use but they work quite well for me with very narrow lines.
I use a HeNe power supply to drive them with a current limiting resistor in series to about half the lamp rated current.

Sensitivity:
As far as intensity is concerned a calibration tungsten lamp used in integrating spheres is my source in my DIY system. I drive it with a small transformer, a variac to do a slow intensity ramp and a meter for current and voltage across it.
It usually comes with a good calibration sheet. A bit pricey at the onset but you get what you pay for. For a given black body emission at a given lamp power you then map the deviation vs wavelength.
They last a very long time if you take care of them. LEDs are likely all over the place for intensity output and color balance even in the same bin/lot.

[harerod]

dietert1, thank you for highlighting the destinction between wavelength and spectral response calibration.

The C12880MA only yields 12..15nm spectral resolution. Each module comes with wavelength (NOT spectral response) calibration data.

The datasheet has a graph showing the typical spectral response. Using that should already reduce the relative error below 10%.

So what I am looking for is a wide spectrum calibration source, whose spectral intensity is known within 10%. I think that a simply way to accomplish that is to measure the spectral intensity of a white LED under known conditions (supply, temperature).
Edit: Spectrum of the SPMWHT541MP5WATMS0 as seen by the sensor. This is the raw ADC data, not compensated for spectral sensitivity.

[richnormand]

I have done lots of photometric and radiometric measurements (including colorimetry) on LEDs and "white LEDs" excited by a blue one.
At the end of the day I would trust a good old black body, naked filament, bulb to measure rough spectral sensitivity response in the visible of the device better.

[harerod]

I agree that the "white LED" might not be the best of light sources. I did play with Tungsten filaments (a.k.a. car headlights, since they are about the only incandescend bulbs left at my place) and they give a smoother spectrum. I like the white LED, because it takes so much less power. I might try and have a look at RGB-LEDs next.

Here are a screenshot of the transmission spectrum of a color filter and a picture of the test setup.

[artag]

What do the Oriel lamps require if you don't have the official power supply ?
I have a 6035 Hg lamp.

[richnormand]

What do the Oriel lamps require if you don't have the official power supply ?
I have a 6035 Hg lamp.

this site seems to imply 10mA:
https://4gte.com/products/oriel-6035-mercury-argon-lamp-10ma-pencil-style-calibration-lamp/
You just need a voltage sufficient to strike the discharge and then current limit. For spectroscopy calibration I usually go for about half the rated current hoping they will last longer.
Once the plasma is established the needed voltage is much lower thus it is better to have a strike pulse to start it.

[richnormand]

I agree that the "white LED" might not be the best of light sources. I did play with Tungsten filaments (a.k.a. car headlights, since they are about the only incandescend bulbs left at my place) and they give a smoother spectrum. I like the white LED, because it takes so much less power. I might try and have a look at RGB-LEDs next.

Here are a screenshot of the transmission spectrum of a color filter and a picture of the test setup.

 Have you considered a simple brake light bulb?  Look at the black body peak as you ramp up and physics will give you a good estimate of what the curve should look like for the whole spectrum intensity.

[harerod]

...
 the black body peak as you ramp up and physics will give you a good estimate of what the curve should look like for the whole spectrum intensity.

richnormand, since I am not the brightest bulb in the box, thank you for finally spelling out "black body peak". I will do some homework and report back.

We are living in a time where the only incandescent bulbs in my lab are from the spare set of my first car, bought for 16,90 Deutsche Mark in the early mid 1990s. Remember when high and low beam were the same 12V 60/55W bulb? I experimented with that bulb, because I could hook up 5Amp Keysight E3600A-100 clips to it. The conducted heat at 9V/4A melted the plastics anyway.

[dietert1]

How stable is this method if you have a bad/unknown intensity calibration causing misalignment of the peak? Is this iterative?

Regards, Dieter

[harerod]

How stable is this method if you have a bad/unknown intensity calibration causing misalignment of the peak?...

dietert1, thank you for all your input. Your question brings us back to the very specific purpose of the original post: The search for a cheap calibrated reference.

I plan to prepare different bulbs / LEDs and then continue to try and find a kind soul who is willing to run that through a calibrated spectrometer.
Edit: added 5W bulb at half and full power

[richnormand]

Here are a few links that might be useful for you.

Spectral irradiance of various Quartz Tungsten Halogen Lamps.

https://www.newport.com/medias/sys_master/images/images/hfb/hdf/8797196451870/Light-Sources.pdf

https://www.newport.com/p/6315

http://apps.usd.edu/coglab/schieber/pdf/handbook.pdf

Note that the 6315 is quite similar to a tungsten bulb (only cost $100 if you want a repeatable source)
The handbook  shows the expected curve/intensity of solar radiation and in section 5 shows using the blackbody peak spectral position to get the temperature.
Deviations from the curve will give you a measurement of the spectral sensitivity of you instrument for detectivity vs wavelength.
It's crude but it works well for a first approximation of the setup.

Once you have measured the power input to the lamp and its lumen output (meters on ebay for about $25) you should be close.

[dietert1]

Thanks for the links. The last one (handbook) explains using a calibrated photo diode
as transfer standard, but then you need to use a monochromator or a set of monochromatic sources.
They write:

An overall uncertainty to absolute of 10% or less is considered very good for
radiometry equipment, and is usually only achievable by certified metrology
labs. An uncertainty of 1% is considered state of the art, and can only be
achieved by NIST itself.

Regards, Dieter

[dietert1]

When i look a harerods measurements, they seem to confirm the 400, 500 and 600 nm bumps of the typical sensitivity curve. What happens if you use the typical curve to correct your measurements and then compare them to the expected spectra? That exercise starts with digitizing the typical curve and making an EXCEL sheet to apply that curve to measured data.
Those recommended Newport "quartz tungsten halogen" bulbs start at US $ 32 for the 50 W version (6332). Don't know how different they are from a H4 car bulb. That Newport bulb has an average lifetime of 50 h..

Regards, Dieter

[DaJMasta]

You probably want to approach the calibration in two different steps, this is what commercial spectrometers typically are setup to do.  You want a reference with multiple pure wavelengths being emitted - the standard for near visible wavelengths seems to be mercury argon lamps - and then you want a reference for intensity, which is almost invariably is something like a tungsten halogen which has been characterized.  The second lamp is usually only required for absolute irradiance and gives you intensity calibration over the range (vs. a known spectral output), but then also quantifies your counts output.

The more practical and easier to do on the cheap one would be the wavelength calibration, and in that case the goal should probably be several tight, known peaks to generate a calibration function from.  Something like an LED can be close, but will not be that consistent and will have a large FWHM.  A laser diode should be tighter, but still can be somewhat variable, and an actual gas based laser should be the most stable in terms of single wavelength sources.  Because you want lots of spectral lines, getting one of those lamps filled with something that emits several seems like the simplest choice, and since the spectral lines are what they are because of the physics of the gas used, if you can source a lamp from another source but using the the same gas, it should still be viable as a reference - at least as good as several laser diodes.

Another approach could be to borrow or buy a working one (there are some spectrometers that aren't too expensive), characterize a few sources, and then compare the output of yours and manipulate it to match.  Still going to work best with many narrow spectral lines and is subject to drift of the unit tested again, but is probably plenty good enough for a proof-of-concept calibration.

[harerod]

This thread gave me a clearer picture about the requirements for spectral and sensitivity calibration of spectrometers. I want to thank everybody for their valuable input.

When I had the first results from my software, I contacted Hamamatsu sales Germany last week. I asked them for a digital version of the spectral response curve from the datasheet, as seen here: https://www.eevblog.com/forum/metrology/small-and-cheap-spectrometer-calibration-source/msg3560894/#msg3560894

They politely asked me whether I would mind waiting a few days, due to golden week. My response was that I know enough about Japanese work life as not to begrudge them a couple of days off.

Yesterday I was told that for each device not only a spectral calibration is performed, but that the sensitivity information is also recorded. I was also cautioned that their calibration source is not NIST calibrated. Would I be so kind as to send them the serial number?

This morning I got an Excel sheet, labeled 相対感度 / relative sensitivity for my unit.
I must say, working with a partner like that is a rare delight.

I will now incorporate this information into my software and move forward from there.

[mjs]

I've designed some stuff with Hamamatsu components and never been let down by them. Once I received an EOL notice about photomultiplier socket we used - 5 years ahead and with a suggested replacement part number..

[richnormand]

Been a long user of Hamamatsu for pmts and photonics modules. Once I needed some schematics for an older cooled ccd camera and was told the original engineers retired years ago but would it be OK if they passed my email to him.
Sure enough, about a week later, he got in touch and we (mostly he) solved the issue.

Customer service like that is pretty hard to find now.

[nimish]

Been a long user of Hamamatsu for pmts and photonics modules. Once I needed some schematics for an older cooled ccd camera and was told the original engineers retired years ago but would it be OK if they passed my email to him.
Sure enough, about a week later, he got in touch and we (mostly he) solved the issue.

Customer service like that is pretty hard to find now.

Did you ever find a good setup for calibration?

[richnormand]

@nimish
Oriel tungsten bulb for a calibrated output at a set voltage/current in my integrating sphere and the spectrometer looking in.
An alternative is  a halogen 12V from the auto store.
https://phys342.rterrien.sites.carleton.edu/final-projects/blackbody-radiation-of-a-lightbulb/
Not quite as precise but the expected output and spectrum is usable and I use a styrofoam cooler case as the integrating sphere.
It compares well to the two real sheres (1m and 200cm in the lab) for stability.

For wavelength calibration I use a simple Ne2 neon indicator bulb. Many spectral lines from neon that are well identified in spectroscopy books.
https://www.researchgate.net/figure/Neon-emission-spectra-solid-line-from-dust-free-dc-positive-column-thin-line-from_fig1_323359177
https://www.researchgate.net/publication/245400165_Atomic_Emission_Spectra_of_Xenon_Krypton_and_Neon_Spectra_from_200_to_900_nm_by_Sealed_Inductively_Coupled_PlasmaAtomic_Emission_Spectroscopy
For the blue end a simple fluorescent tube will have several lines that are easy to identify. Again that works well since the spectral lines positions and wavelength are a property of the gas used.
https://commons.wikimedia.org/wiki/File:Fluorescent_lighting_spectrum_peaks_labelled.gif
The fancy cold cathode laps I have have smaller linewidths but you need meter size spectrometers to resolve that.

Although I have more sophisticated resources  I rarely use them  as my setup is working well.

[FFY00]

Have you considered a DRS-50-1 bulb? They can be had for ~25$ on ebay.

http://lampes-et-tubes.info/dlsp/dl250.php?l=e

I personally have an Ocean Optics HG-1 mercury argon source.