Week of 03/12/2018 - Progress Report 13

Tamara Jovanovic

Topics covered: System stability measurements with DI water and a 100 mg/dL glucose solution

Materials used: Optical Setup of the Non-Invasive system, lab equipment, Menlo Systems C-fiber Laser, Yokogawa Spectrometer, Quartz cuvette, distilled water, 100 mg glucose solution

This week, I focused on getting data in the range specific for what we are going to be doing from now on. In last week’s calculations, after comparison and analysis, we came to the conclusion that taking data in free space is best at the following setting:

This is definitely not the predicted outcome. By using common sense, we predicted best parameters for taking data would be the High 3 sensitivity and with 15 average waveforms instead of 5. This data gave us a lot to think about, and so Ezequiel compared all the data, and the table above still was the best combination of parameters. So, we decided to double check everything, and start taking data through distilled water and a 100 mg/dL glucose solution.

After some consultations with Dr. Li and research about the glass cuvette, we started taking measurements in the glass cuvette through distilled water. However, this is when we ran into the problem of significant power loss. According to research, the power loss through glass of water at 1550 nm should be only -3dB. But we got a lot more than that. So we decided to take a step back and start realigning our system through water, while trying to maximize the power and minimize the power loss.

I then compared the glass and quartz cuvette one more time, with DI water inside. The glass cuvette definitely gives a smoother curve than the quartz cuvette, but unfortunately there is more power loss through glass than there is through the quartz. So, we decided to dig deeper into research and hopefully figure out soon why our system is giving us that much power loss through a glass cuvette with water.

In the meantime, I switched back to the quartz cuvette and started realigning the system. The first step for this is looking at the power output from the output collimator through DI water in the quartz cuvette. After our advisor, dr. Asghari gave us some pointers and suggestions about alignment, we were able to get ~200 nW of power through water. If I am being a stickler, this is not ideal, as in the perfect world, we would want to have much more power through just water. However, since that was the best that we could do this week, we decided to take measurements at this power and record and compare in the future, in case we are able to get more power at the output.

The tests were conducted at the following parameters, outlined in the table below, with a box over it covering up the system and eliminating all the outer world disturbances:

As you can see, the only resolution used is 2 nm, because that is the highest resolution that our spectrometer can write. This is one of the best settings, so we are keeping it constant.

The solutions tested were clear DI water at the 4 different settings, and then the 100 mg/dL glucose water solution, all in the quartz cuvette. As usual, 7 traces were taken of each combination of parameters.

The sensitivity was varied between High 3 and High 1. All the results would’ve been taken at High 3 but because of the results that we got last week, we decided to test High 1 as well and compare those results accordingly.

The average waveforms taken were varied between 5 and 15, with getting all combinations of them with High 1 and High 3 sensitivities. It is really great that our spectrometer can take that many average waveforms per trace, and so we want to utilize that feature to our advantage and get the most accurate results. That is why this week we didn’t use 1 average waveform per trace, but only 5 and 15.

As mentioned earlier, a box was used to cover up the whole system and eliminate all the influences of the outside world, i.e. other UV light, maybe some unwanted reflections, dust, dirt etc. Hopefully, when taken into analysis in Matlab, this will be helpful.

The alignment of this week’s data acquisition was taken a little bit differently. After we got the maximum power we could (~200 nW, as mentioned previously), I then took a look at its waveform at that power and got something looking like this:

These are actual absorption lines of DI water through the quartz cuvette. The fact that the ripples start right from the beginning indicates that the process of alignment was done correctly. From this point on, the collimators weren’t touches, just the rotation knob on the rotational stage. The cuvette was slightly turned to one and then the other side in attempt to smooth out this curve and get rid of as much ripples as possible. The waveform that was the result of this process is as follows:

The fact that the adjustment of the knob on the rotational stage gave me this kind of a smooth curve is great and beyond great and better than expected and anticipated. Hopefully, after the results of the spreadsheets generated by this testing are plugged into the Matlab script, we will be able to better see the discrepancies and hopefully create a tolerance percentage which will be used in all future measurements.

I am not sure if you will be able to tell, but the set up of our system now looked as shown in the picture below. If you look carefully at the angle of the cuvette and the rotational scale on the cuvette stand, you can see that the cuvette is not quite 100% straight pointed at the collimator which intakes the laser signal, it is slightly turned.

By just eyeballing the resulting absorption spectra, I can say that they look pretty good and consistent throughout the 7 trials taken. Though, as we’ve seen before, this is not the indicator that it is actually like that. I will have a more clear vision of the results once these results are plugged into Matlab and analyzed in there. The error will be plotted versus the average of the 7 trials of each of the 8 specific measurement set-ups and we will be able to see the error as well as where the power loss occurs and at which point is it maximal and at which it is minimal. THis will hopefully be a good base for analysis in the coming days.

The dynamic range observed on the spectrometer’s screen was approximately 30 dB’s. Also, as mentioned before, this does not provide me with the ability to actually see where the waveforms are and what the discrepancies are. However, just by looking at it, all the measurements taken at the sensitivity High 3, which is the highest one that we have, seemed smooth and good, even just the top part of the absorption spectrum that I was looking at. However, for measurements through the 100mg/dL glucose solution, and at High 1 resolution, I noticed that at the beginning of my dynamic range, the spectra seemed like it was noisy, or rippled or something. Maybe this is the indicator that the High 1 setting of sensitivity is not a good setting. But that is just an observation and I will have more insight into what’s happening once the results of the measurements are plotted in Matlab.

Next week’s progress:

• Keep taking glucose and DI water data

• Adjust alignment for more power through just DI water

• Print new cuvette stand to maximize stability

• Research more about absorption of water and solutions through quartz at 1550nm

Jonathan De Rouen

Topics covered: Water Absorption, Power retention, Enhance data collection, enhance alignment

Materials used: Cuvette, NIR detector card, research

This week I worked on improving our setup by moving our location on the optics table. I also worked with our advisor, Dr Asghari, to try and improve our power retention of our overall system. Instead of using different collimators they are now both the same type of collimator. Due to this we were able to retain approximately 2.5 mW of power in the free space. I also tried to understand why we were having so much loss in power when we included water in a cuvette in our system. When we added a cuvette in the system our power dropped to the nanometers which meant we had a very high absorption due to water. This was confirmed by Dr. Asghari by using the laser through a beaker of water to the NIR detector card. Some research we found displayed the absorption of water to be approximately 3dB loss which made how there was no transmission of light back to be very confusing. This assumption came from this plot from the research from this website. http://www1.lsbu.ac.uk/water/water_vibrational_spectrum.html

However this was definitely not the case. Further research after realizing this data was wrong led us to find an article by OSA publishing from 1951 about the Near infrared absorption spectrum of water. This article took research from multiple scientist studying the absorption at different wavelengths. The results of there data is shown below.

There data shows that the wavelength we are analyzing water at is close to the maximum absorption wavelength at that given light. This data seems to correlate with how much loss in power with our system. I have reached out to Dr. Li and asked for some further explanation about this phenomenon as well. However we can still continue getting data.

I also assisted Ezequiel with writing a program to set and gather data for us.

Next week

• Gather more data

• Talk to Dr. Li about waters absorption at NIR light

• Determine to test if a different wavelength is a viable option for our system.

Ezequiel Partida

Topics Covered: Matlab Code to Automate and Facilitate Data Collection, Optical System Alignment Maximization, Data Standard Deviation Analysis Continuation

Materials Used: Optical System, Matlab

This week, a lot was done in terms of progress having to do with data collection. 1 First of all, for Jonathan and I, saving data for future analysis was made more efficient and faster with a code we worked on to automate the entire process. 2 Secondly, using this process, we were able to more accurately determine which setting on the Spectrometer yields less error. 3 Finally, more changes were made to our optical setup to maximize power retention through distilled water in order to once again minimize error. These 3 accomplishments are mentioned more thoroughly below.

1. Matlab AutoCollect Function

Before this week, in order to collect multiple trials of spectral data and then analyze it in Matlab, we had to use the 7 traces on the Spectrometer and control them manually. However, it is much more efficient to use an automated script that can 1) Do this on its own, and 2) take more than 7 trials dependent on the user needs. Therefore, Jonathan and I quickly worked together to write a function that does this. The function connects to the OSA remotely, sets up any desired settings, loops a specified amount of times to take a data trace, and outputs all traces to a .csv file. To use the function, one has to simply call it, defining its 2 inputs – Filename and number of iterations. The function in Matlab looks as shown below.

One major way that we used this function was to determine what setting on the OSA is the best in terms of error. Previously, we manually ran through only 7 traces and did standard deviation analysis. Now, we could run for as many traces as we wanted to in order to get a more accurate number.

2. Best Spectrometer Settings

Using the AutoCollect function for a total of 20 trials each at Free Space output, we captured the following:

Based on the table above, it can be clearly seen that using the OSA at HIGH1 sensitivity yields better and more error-free spectra. For worst case scenario across all wavelength standard deviation values (Max. Standard Deviation), using HIGH1 was almost 20 times better. For an average of standard deviation values across the entire spectrum, HIGH1 was around 10 times better. Additional details to note about this experiment is that 20 trials seems like quite a lot to determine good amount of variability and error. Also, both tests seen above were performed with a box over the optical setup to prevent any external light from affecting the results. The code was also automated to take the data, so there were no interruptions or human error. Finally, the Matlab functions used to perform this analysis have inputs that allow the user to determine whether to look at a full spectrum, or only within a 20 dBm range. For this experiment, the option of only a 20 dBm range was used since there is significantly less noise in that region of the spectra.

3. Optical Setup Optimization

Late in the week, Jonathan, Tamara and I worked hard with Dr. Asghari in order to align our system even further in order to retain the most power possible through distilled water and generate spectra that are not noisy. For example, early in the week, we tried to analyze glucose data using our concluded best OSA settings. However, even through distilled water, we got spectra that looked as shown below.

Due to the ripples in the spectra and the overall noise contributing to high standard deviation values of around 0.5 dBm, analysis of glucose concentrations would be nearly impossible to accomplish. Thankfully, our system was able to be aligned so we could take preliminary data with around 200 nW of power through distilled water. Applying our Matlab standard deviation functions to each experiment ran, the results below were gathered.

Based on the data from the table above, it is seemingly inconclusive which setting on the spectrometer yields the least amount of error for the amount of power we retained in this experiment. However, it can be useful to note that the test run that generated the smallest MEAN standard deviation was for HIGH1 sensitivity, at 15 average samples and distilled water at 0.0075 dBm. The next two smallest values were also for HIGH1 sensitivity, so that could be a pointer that HIGH1 sensitivity yields more consistently low error. However, if we look at the highest peak of standard deviation across an entire spectrum, the smallest standard deviation of 0.02 dBm was for HIGH3, 5 average samples, through 100 mg/dl glucose solution. Nonetheless, all of the standard deviation values acquired with our Matlab function were very low, which is a good sign that our system is becoming more stable. Additionally, only 7 trials were ran per test, whereas if we had used our AutoCollect function, we could run tens to hundreds of trials in order to more accurately determine lower error.

However, for this week I believe we made great progress. Also, for reference, the values in the table above were outputted in the following form (Below is the test with the lowest MEAN standard deviation at HIGH1, 15 AVG, Water Solution).

Also for reference, I made a quick plot to plot the difference between distilled water and glucose concentration to see if there were any discrete changes. Below is what I found.

At first glance, this makes me hopeful because the difference sees a peak of 0.2 dBm difference, which is significant enough compared to around 0.004 dBm standard deviation. Hopefully this week we can continue experimenting with data collection to make more concrete conclusions.

Plans for next week:

1. Automate data collection to yield lower error spectra for new power output.

2. Experiment with different glucose concentrations.

3. Go back to RMSE and cross correlation analysis to make conclusions.