|Product Performed to Expectations:||10|
|Specifications were sufficient to design with:||10|
|Demo Software was of good quality:||8|
|Demo was easy to use:||10|
|Support materials were available:||10|
|The price to performance ratio was good:||10|
|TotalScore:||58 / 60|
Human vision is provides us the ability to see 400-700 nm spectral band. At these wavelengths we usually observe electronic transitions of atoms. In order to get more information about material's molecular structure we have to look at it's vibrational spectra. Since molecules are larger and slower, vibrational spectra lies in the infrared domain. Usually the fundamental oscillation is in the middle infrared domain (>2500 nm), however, it is possible to see "overtones" of these oscillations at higher frequencies, for example at the NIR band (700-2500 nm). Probability of such overtones decreases with higher photon energy, thats why we rarely see them in visual band.
Near infrared spectroscopy is often used for quality monitoring of agricultural products, for example grains, oilseeds, coffee, tea, spices, fruits and etc.. By measuring the transmission or absorption spectrum of the material and comparing it with a known reference, it is possible to determine if there are some changes of quality. Also it is possible to check the composition of the materials by comparing the amplitudes of the peaks in the spectrum.
Principal scheme of the spectrometer is shown below. Spectrum is spread accross diode array by dispersive element (usually the grating) and intensity data is sent to the computer. For VIS-NIR spectrum usually Si and InGaAs detectors are usually used. Their spectral responses are shown below. As you can see Si detector is better suited for visible wavelengths and InGaAs has a bandwidth of ~800 -1700 nm. That's why it is most commonly used for NIR spectroscopy.
Main problems with diode array spectrometers is the noise and price. Noise comes from the fact that the sensor has to be responsive at longer wavelengths. Longer wavelengths means shorter bandgap of the detector, thus it is less imune to thermal noise. This problem is usually dealt with by introducing active cooling to the detector and increasing it's size. Both of these increase the price. For example, Ocean Optics NIRQuest spectrometer operating at 800-1800 nm wavelenghts costs ~20k $.
Texas Instruments DLP NIRscan spectrometer uses DLP micromirror array and a single InGaAs detector instead of the conventional spectrometer design (shown below). DLP is a technology invented and developed by Texas Instruments in 1987's and it is apparently cheaper than using linear InGaAs sensor. This approach provides not only a better SNR but also a possibility to implement custom scanning patterns, spectral filters, variable resolution and etc.
NIRScan Nano is meant to be a mobile spectrometer for portable NIR solutions. It has a sample window where the user can place his sample. Tungsten lamps then illuminate the sample and the reflected light is then coupled into the spectrometer.
One can calculate that DLP micromirror array allows <2nm resolution (1700-900) / ( 854 * 0.8 ) = 1.17 nm. 0.8 multiplication comes from the fact that array is underfilled by 10% on both sides. However, optical setup specifies only ~10 nm resolution.
Well, the box contains mainly the spectrometer itself and some papers. Not even a micro usb cable - but who does not have them lying around everywhere?
The spectrometer is quite small, although not designed to be wearable yet At first I wanted to disassemble it and take a look at the optical setup, but it seems that somebody has done it for me already, here are the links: https://e2e.ti.com/blogs_/b/enlightened/archive/2015/10/15/taking-a-look-inside-the-dlp-nirscan-nano-evaluation-module
Demo software is built on QT C++ and is called NIRscan Nano Gui. Just after I turned it on I realized that my firmware for TIVA mcu and DLP controller are outdated ("TIVA SW version" and DLPC flash version was 1.x.x). I downloaded firmware from TI page and flashed new firmware (2.x.x) on utilities tab. Upgrade was very straightforward and went without any problems, which from my experience is very rare phenomena.
As you can see there are also quite a few temperature sensors (4!) and one humidity sensor. All temperature sensors show more than 30degrees. I can understand detector temperature/Tiva temp and HDC temp (I don't know what it does yet) showing >30deg temperature, but ambient temperature at my office cannot be that high. Also humidity of 10% seems unreasonable. I suspect that the reason of this is a poor placement of sensors, nearby heated parts for example. But I do not need them anyway.
Lets see the scanning window, which is very basic. There is a spectrum window and scan configurations. Spectrum is displayed in 3 ways - raw intensity data, absorbance and reflectance. Intensity tab shows what signal was received by ADC, while reflectance shows ADC value minus IR lamp calibration spectrum. Pictures below show spectra collected when there is no sample in front of the spectrometer. There is some signal in "Intensity" tab, but "reflectance" shows almost no signal, which means that optical alignment of sample illumination unit is well done.
Next, I scanned my skin and the picture below shows it's reflection spectrum. I tried to see some differences between ordinary skin and a mole, however I did not see any. I guess this could not be utilized for skin cancer diagnosis
You can also see some noise at the very ends of the spectrum. This is due to the decreasing sensitivity of InGaAs at these wavelengths (see the curves in the first section).
Scan configuration allows to customize DLP scan patterns. You can choose between "Column" and "Hadamard" scan methods, wavelength range, digital resolution and exposure time for detector.
Column scan is simply what it is - it scans columns of DLP mirrorarray linearly. Hadamard scans DLP in complex pattern and is preffered scan by TI. It multiplexes different wavelengths on the detector and then uses algorithms to reconstruct the spectrum. According to TI it gives ~50% better SNR compared to column scan. I do not really know how all of that works it seems pretty complex.
The fastest time you can achieve with this device is ~1s. I tried reducing wavelength range. Even if I reduce the range to 100nm (instead of 800nm), the scan time stays >1s! I think it is the problem in firmware somewhere, or data transfer rates are slow. My application will require real time monitoring, so this is quite a shortcoming.
I needed to make an adapter for SMA fiber in order to measure laser wavelength. I realized that SMA adapter could be attached on top of the sample illumination module. Collimating lens would only help collecting more light on the slit. Design guidelines recommend using linear fiber, since the slit is not round but linear. However, thats expensive and difficult to align and usually I have more than enough laser power. For test purposes I simply used double sided tape and the result is pretty good. I also had to disconnect tungsten lamp connectors since I do not need illumination.
I had Yb:KGW femtosecond laser by my side, radiating at 1030nm. Since it was set to provide ~0.5W I used scattered light from metal plate and tried registering spectrum with NIRscan Nano. I had to take raw data from the software and analyze it on MATLAB, because main software graphs were clipped. Here are the results:
First impression is pretty good. Curve is very smooth. However, there is a strong background present, which I did not see in the main software. I believe, I should have tried different exposure time, maybe the background would be much lower then. FWHM of the curve is around 15nm, real laser bandwidth is <10nm.
I had OceanOptics NIRQuest to compare it with. It is a large, noisy (due to active cooling) and expensive device, providing 900-2500nm spectrum. I had to direct all the light straight to the fiber, otherwise I had to set a long exposure time and it showed only the noise. The results are below.
As you can see, SNR is much worse than NIRscan nano, even with much more light coupled to it. The curve is very edgy, due to larger digital resolution of the spectrometer. However, FWHM of the curve seems to be smaller - ~10nm. That means that the optical setup of the NIRquest provides better spectral resolution (~5nm) than NIRscan Nano (>10nm specified). For NIRscan nano I used TI recommended "Hadamard" scan pattern, which exposes different wavelengths on the sensor at the same time and uses algorithms to retrieve the spectrum later. I suspect, that it might reduce spectral resolution of the device but maximize SNR. I still have to test this.
According to TI's document http://www.ti.com/lit/pdf/dlpa066 called "Flexible trade-offs In Maximizing SNR and resolution in TI DLP based systems", which shows that spectral resolution of NIRscan nano EVM is <10nm with both "Hadamard" and "Column" scans. My results are different, so I might need to adjust my fiber coupling or scan settings. According to another source http://proceedings.spiedigitallibrary.org/proceeding.aspx?articleid=2505009&resultClick=1, they managed to get ~8nm resolution by reducing the slit from 25um down to 15um. However, I do not dare to experiment with the screws on the slit on my device yet.
I have created my custom software which can perform continuous scan. It is my first ever C++ and QT application, so don't judge The main features are:
Here are some screenshots. Blue spectrum shows current measurement, the red one - loaded reference from file. Background subtraction saves current spectrum as reference and subtracts it from next spectrum (seen in second picture).
Calibration is performed by dividing the spectrum of InGaAs sensitivity curve. 1st picture actually shows calibrated (blue) and uncalibrated (red) spectra. I still have to do some improvements, but the program works so far pretty well .
I have performed an experiment with IR LED connected to Beagleboneblack PWM pin and generated 50Hz - 10kHz square wave with 0.5 duty cycle. The results for COLUMN scan are shown below:
As you can see, there is significant 'chopping' of the spectrum, and the period of chopping increases with frequency until it vanishes. At low frequencies one can notice that part of the IR intensity goes to larger wavelengths also. I do not know what is the reason for intercoupling between different wavelengths, but the chopping of the spectrum can be explained simply by the scanning nature of the spectrometer - since it is not synchronized, sometimes it sees the pulse and sometimes it does not. When going to larger frequencies, number of pulses received per wavelength "averages out" and we see continuous spectrum..
Hadamard scan has shown even worse results at <10000Hz frequencies. At <1000Hz only noise is present, and only at 10kHz i was able to get an almost usable spectrum:
If I compared DC hadamard with DC column scans, I get very similar results:
I am just getting familiar to NIR spectral measurements, since I was usually working with Si based detectors, however first experience with NIRscan nano astonished me. This little device proved to be capable of the same things that 20k $ spectrometer does and even outperform it. Considering it costs ~1000$, that's susprising. The only thing that bugs me is ~1s scan time, limiting it's real time application. Spectral resolution of the optical setup (>10nm) is also limitting DLP capabilities (<2nm), however that is understandable since this is an evaluation module, squeezed in a really tight package and optimized for better SNR, rather than spectral resolution.
Everything from TI side - documentation, forum support, download section, firmware updates, was easy to reach and I can't say any negative word about them
2017-04-06 Attached my custom QT application, first version. Still have to do some adjustments, but I am quite happy with the results. Software tested on TIVA SW version 2.0.1 Also DLPC Flash version 2.0.0