Putting a Spectrometer on a Cell Phone

Will your next cell phone include a spectrometer? In this interview, Alexander Scheeline of SpectroClick, Inc., talks about what is involved in creating such a device, and his company’s work toward that end. This interview is part of Spectroscopy’s 2014 interview series with the winners of award that are presented at the SciX conference. Scheeline, along with Thu Anh Bui, of SpectroClick and Vietnam National University, in Hanoi, won a FACSS Innovation Award at the 2013 SciX conference last fall.

Your talk at the 2013 SciX conference discussed how to use a complementary metal–oxide–semiconductor (CMOS) camera — the kind of camera that is used in cell phones — to make a spectrometer. Why has a cell-phone based spectrometer not been commercialized to date?
I'm not going to speculate on why other people haven't sold cell phone spectrometers (1). We stumbled into inexpensive spectrometry when I taught instrumental analysis to students in Hanoi, Vietnam. They had had no opportunity to use any instrument more sophisticated than a balance. I devised a build-your-own spectrometer kit, thinking the students would use digital cameras as detectors until my teaching assistant, Kathy Kelley, took one look at my Windows software and asked, "Why isn't this going on an iPhone? There's an app for that!" That was early in 2009.

What are the key challenges in making such a device?
What makes a cell phone spectrometer (CPS) or webcam spectrometer or digital camera spectrometer hard [to develop] is that these devices are not designed for doing science; they're designed to take snapshots. How can wavelength be calibrated? How do we deal with the 8-bit dynamic range of each color in the typical cell- or webcam pixel? How do we cope with the intentional nonlinearity of the images? People like intense, saturated colors. Science demands repeatable, preferably linear response. Unlike with charged-coupled devices (CCDs), any attempt to use commercial, off-the-shelf cameras requires that we use whatever the consumer already has. We can write software, but we can't optimize hardware.

What types of spectrometry are possible in such a device?
At this point, we're targeting absorption, reflectance, surface plasmon resonance, and fluorescence. The stray light levels, resolution, and throughput are such— at least the way we're thinking about it— that I doubt Raman spectroscopy is feasible.

How powerful a CMOS camera is needed to make a spectrometer?
There's a sweet spot. With too few pixels, one can't get adequate resolution. With too many pixels, data processing is glacial. Anything less than 1 megapixel is going to sacrifice resolution and dynamic range. I look at the 41-megapixel camera in the Nokia 1020 cell phone, and I shudder at the amount of data in each image. Fortunately, our algorithms are fast — data-processing time scales in proportion to the number of pixels plus the number of observed orders.

What other parts are needed to make a spectrometer using a CMOS camera? Could you actually fit everything you need into a cell phone? If not, how small could the spectrometer you envision be?
Any spectrometer needs a light source, collimation optics, a sample cell, and dispersion device in addition to a detector. I doubt that anyone will build all those components into a cell phone any time soon, though there are examples of people doing reflectance and fluorescence with just the flash on the phone and the camera. We've developed a package with a light source, collimation, sample cell, and dispersion that fits comfortably in the palm of your hand, and we will be exhibiting it at Pittcon this year. It uses a USB camera and Windows software. It's about 9-cm square and 4-cm high. If we didn't have the built-in camera, obviously we could go smaller. I'm aiming for something the size of a computer mouse. Considering that billions of people have cameras in their phones, you can imagine where this is headed.

How would one make such a device?
You can get a pretty good idea from what we published in the Journal of the Analytical Sciences Digital Library in 2009 (2) and in a “Focal Point” article in Applied Spectroscopy in 2010 (3). All the components are there, but the free-form component positioning suffers from high stray light, inadequate dynamic range, and poor mechanical stability, which prevents repeatable, accurate wavelength calibration. The breakthrough that got us launched was Thu Anh Bui's brainstorm to generate hundreds of orders simultaneously. Our patent-pending grating solution (4) uses stacked double-axis transmission gratings, giving rise to what we've dubbed a SpectroBurst. The multiplicity of orders fills the array detector and gives a whole range of throughput and resolution trade-offs. By averaging orders that are neither saturated nor too weak to see, we hope to gain precision. By having some orders saturated and others not, we get dynamic range that is the product of the dynamic range of the camera and the grating and optics throughput. When we do fluorescence, we can use the low-dispersion orders for low-intensity measurements, and the higher-dispersion orders for stronger signals. I know of no other system that, in a single measurement, can allow such a trade-off after the data are already in hand. This flexibility is why we think some early adopters may want to try our prototype.

Figure 1: A 5-megapixel “SpectroBurst” produced by a grating solution that uses stacked double-axis transmission gratings.

What stage are you at in developing a prototype device?
The device we will be showing at Pittcon is dubbed the “AAH-200” and is our third design. The first design (the basis for what we discussed last fall at SciX) used a 3-megapixel Mightex BCE-C030-U camera. Our more recent designs employ a 5-megapixel device. The former was easier to program, control, and characterize because the interface was less consumer-like and more science-like. We are still dealing with automating calibration. Until the calibration issues are resolved, measuring application performance isn't possible. Prior to SciX, we ran one absorption working curve, extracting data from the images manually. The Beer's law plot was terribly nonlinear for reasons we think we now understand. Wavelength calibration and cuvette alignment stability were both critical issues.

Is special software needed to help interpret the spectra obtained using the device?
Absolutely yes. SpectroBursts present data-handling challenges, on which we have made significant headway. Once getting a spectrum is automated, the rest of the data handling will have a lot of similarity to what goes on in conventional systems. But getting to a clean, well-characterized spectrum? Aye, there's the rub (as Hamlet said)!

What applications could such a device be useful for?
This instrument should work for any visible spectrometry measurement for which resolution need not be better than 1–2 nm and for which there is sufficient intensity. We foresee applications in environmental monitoring, point-of-care medicine, home medical diagnostics, agriculture, homeland security, defense, and education. We think we can get better results than some of the first commercial inexpensive single-beam spectrometers, with better resolution and broader dynamic range, at a lower cost than many if not most competitive products. But that assumes we make it through every startup's early hazard, the "valley of death," where costs and risks are front-loaded and income is back-loaded.

Could such a device be cheap enough to enjoy mass use?
We're banking on it. We have found ways to squeeze cost out of most of the components and are working on dropping cost on the last few items on which we haven't previously focused. One of the harsh lessons this former academic is still learning is that the overhead of business can be more expensive than the product sold. Lawyers, accountants, landlords, and exhibit organizers all cost money, and eventually the consumer pays for it all.

What are your next steps in this work?
We are going to crack the calibration problem. After that, we'll find some alpha users, clean up the computer interface, harden the code, and go to market. We have all sorts of interesting ideas on building training materials into the software, using analytical instrument markup language (AnIML), the, together with QR and bar codes and the web to serve method instructions to users at the point of need, to flip the classroom so students buy their own instruments instead of having colleges or even high schools provide them, and combining electrochemistry and separations with spectrometry. But before any of that can happen, we need to turn SpectroBursts into spectra. By the time your readers see this, we'll be there.

References

• S.X. Wang and X.J. Zhou, U.S. Patent 7,420,663 (2008).

• Scheeline and K. Kelley, J. Analyt. Sci. Digital Libr. Entry 10059, 11/30/09. Reprinted in m-Science: Sensing, Computing, and Dissemination, E. Cannesa and M. Zennaro, Eds. (The Abdus Salam International Centre for Theoretical Physics, November 2010).

• Scheeline, Appl. Spectrosc. 64(9), 256A­­–268A (2010).

•  “Energy Dispersion Device,” T.A. Bui and A. Scheeline, U.S. Patent application 13/596,242, 20130093936 A1 filed 28 August 2012, published 18 April 2013.