Week of 02/26/2018 - Progress Report 12

Tamara Jovanovic

Topics covered: System stability measurements

Materials used: Optical Setup of the Non-Invasive system, lab equipment, Menlo Systems C-fiber Laser, Yokogawa Spectrometer

This week’s goal was simple, yet time consuming. The idea was to acquire maximum stability of the optical system. I’ve talked about stability in previous progress reports and various methods of stability were discussed. Keeping maximum possible power through the output collimator and also getting a good spectral data at the same time is the ultimate goal.

After some discussion with our advisor, Dr. Asghari, we came to several conclusions about how to fix stability.

First, our optical system is highly sensitive. We’ve had trouble getting rid of noise on the absorption spectra of glucose. Because of this, we decided to eliminate the problem from the start. Though our equipment is high quality and shouldn’t pick up noise from other light sources in the room, just to be sure, we started using a cardboard box over our optical system, just to make sure. From now on, all the measurements that are being taken are going to be with a box over the system.

After some adjustments to the alignment and with using the box, we have seen some good changes. The idea for stability is to get all this data about discrepancies and baseline tests in free space and through water so that we can calculate the differences when taking glucose data. That is why all the data for this week is taken in free space, by solely adjusting some other parameters on the spectrometer. The parameters adjusted were the following:

• Resolution

• Sensitivity mode

• Average of waveforms taken

In the testing, two components would be constant at all times, and one would be changed. This was repeated until all combinations of these 3 parameters were interchanged. Therefore, there was a lot of data collected, as all combinations of these parameters would result in 33 combinations, or 27 spreadsheets.

The summary of all the combinations of setting of the spectrometer are shown in the table below. All measurements were taken in free space.

7 trials of each combinations were taken. The general shape of all the trials in free space looked like this:

Regarding resolution, the three ranges tested were 2nm, 1nm and 0.5 nm. Regarding average waveforms, we did 1, 5 and 15. And regarding sensitivity mode, we did High 3, High 1 and Normal. Since 15 average waveforms at highest sensitivity took about 20 minutes per trace, this was a really long data collecting process. However, after all the data was sent to the guys for analysis, I am hoping we can have some benchmark stability calculations for discrepancies in future glucose measurements.

If necessary, other resolutions will be used. However, the generation of those traces would take a really long time, I don’t think it would be a necessary step for our project. At 0.02 nm, each trace would take roughly 4 hours to generate, and we need 7 per each combination. However, these are the first measurements.

Reference level was kept constant at -25 dBm. The idea is to take the average of all trials and get the standard deviation. Hopefully, this results will give us some conclusive results and so we’ll be able to analyze glucose data more precisely in the coming weeks.

For next week’s plans, we’ll consult our advisor on whether we need to take higher resolution or change any other parameters in free space, and then we’ll move on to taking data through distilled water and moving forward, glucose water.

Jonathan De Rouen

Topics Covered: Pre Processing data, Worked on calibration of Cuvette, Research on more stable devices

Materials Used: Matlab, Rotational Stage, Spectrometer (Yokogawa AQ6370D)

This week I first began with doing some initial analysis of our system by using some measurements Tamara gathered and plotting them into MatLab. Without adjusting or moving our system we gathered data for just the free space input, the glass cuvette with water, and the Quartz Cuvette with water. From this data which we traced 7 times each on the spectrometer I calculated the standard deviation each plot had over the entire spectrum we were analysing from which was from 1500 - 1670 nm on the Spectrometer. Below you’ll see the initial plotted data and the standard deviation that each point had in order to find a trend.

From seeing these results I became worried about the calibration of the rotational stage due to such large standard deviations from each point. Although since this was initial data and there were obvious big traces where calculating standard deviation would greatly increase. I decided to first take a look at the rotational stage and try to align the system even better with the Power meter and the rotational stage. By adjusting the rotational stage with the quartz cuvette and water in the system and trying to maximize the power I was able to get a different shaped curve. However when aligned the maximum power I could read was about 30 nW when we had a .7 mW power in free space. However I was not concerned greatly because the curve at that point was a distinct figure that was decently aligned. I also used this opportunity to test the box system to try to lessen any interference from the lights or other sources of light from the surroundings. In the next set of trials I focused on the freespace, and the quartz cuvette with water, and ignored the data for glass and Water, due to the different refractive indices, I believed that when one was aligned to there maximum the other form of cuvette would be out of alignment. So repeating the trial again the results concluded that our box had negligible effect on our system.

However this analysis of our possible problem was not significant for now due to us being uncertain if we were using the most accurate combination of settings from our spectrometer. In order to better find if we were using these optimal settings we took 27 different sets of 7 trials each, at varying settings. I provided some initial code for analysis working with our first 9 data plots. From these results it seemed that the best setting at the highest resolution for our system was keeping it at a medium sensing mode and at our middle averaging sample rate, which was 5 samples per trace. Below is the maximum standard deviations of our plots in vector form left to right increasing averages of waveforms at 1, 5, and 15 samples per trace, and every 3 points increasing the sensing mode at High1, High3, and Normal, all while keeping the resolution at 2 nm.

So according to the format provided above the best case for these 9 trials was point 2 followed by point 9 or a 5 average sample waveforms at High1 sensing, or normal sensing at 15 samples. However Ezequiel has the final results of this analysis.

Another focus for this week was researching other methods of maintaining stability. Through my research I found many optical system cuvette holders like this one from Ocean optics and another from Thorlabs.

However due to the price of this products being approximately $500 dollars and more new, it is highly unlikely that we can use this cuvette holders. However I did find some cuvette holders for sale on Ebay for less than $100 dollars. A 90 degree Fluorescence Cuvette holder from Ocean Optics with SMA 905 connections for $80 dollars and $70 dollars each before shipping and both used. However I am not sure how much I can trust these products since they are designed for a different spectrometer after all, but it should not matter if the SMA connectors will work.

This type of device would improve our sample collection because instead of worrying about our alignment through free space, the fibers would be directly touching the cuvette assuring the light was passing through. However they are also SMA connectors or in multimode, which is not something we have been using, since we use the FC single mode fibers.

Next week Progress:

1. Talk to our advisor about our calibration settings but move on and confirm using quartz cuvette and water if this is the best setting for analyzing our system.

2. Continue to analyze data and hopefully get some glucose data

Ezequiel Partida

Topics covered: System stability measurements, Matlab Standard Deviation Analysis of Stability Data, Process Data Capability of Optical System Analysis

Materials used: Stability Data, Matlab, Research

This week, we were tasked with gathering data to test the stability of our optical system with different settings on the Spectrometer. Through free space transmission, we took 7 trials for each setting combination of resolution (2nm, 1nm, 0.5nm), sensitivity (Normal, High1, High3), and average samples (1, 5, 15). After helping Tamara to take 7 trials for each combination of settings (27 total spreadsheets of data), the files were saved and used in Matlab for analysis.

In Matlab, I created a function that takes in the name of a .CSV file and performs standard deviation analysis of all the data to test that setting’s stability. The function also plots the average of the 7 trials over wavelength, and additionally plots the subtraction of each trial with reference to the average. In a separate figure, the standard deviation for each point in the data is plotted to note the magnitude of standard deviations across the entire wavelength range being analyzed. An example of the function’s outputs are shown below. (The function is ran with a file containing trials of data for Spectrometer settings at High1 sensitivity, 0.5nm resolution, and 15 average samples).

Once this function was debugged, it was used in a script that allowed me to compare all of the files we gathered from the system. The script I wrote went into a directory containing all 27 excel files and plugged them into the StdA function to retrieve the standard deviation value from each corresponding file. Then, each result was placed into a corresponding structure containing the filename and the standard deviation value. These were then all compared to find the corresponding filename yielding the least amount of standard deviation. The script is shown below for reference.

Another functionality I added to the function was for the user to choose whether to look at the data containing the entire spectrum or only for the part of the spectrum containing dBm values 20 dBm from the peak (20 dBm is 100 times lower than the peak). Also, I added an input to determine whether to look for the maximum standard deviation point or for the average of all the standard deviation points. Therefore, after doing all this, I ran my script 4 times to look for the best settings on the spectrometer per our data acquisition this week.

1. Full Spectrum, Maximum Standard Deviation

​2. Full Spectrum, Average Standard Deviation

3. 20 dBm Range Spectrum, Maximum Standard Deviation

4. 20 dBm Range Spectrum, Average Standard Deviation

As it can be seen from the above tests, it seems as if the settings that yield the best results with respect to standard deviation for 7 trials of data is High1 sensitivity, 2nm resolution and 5 average samples. However, as it can be noted in the plots in the above tests, when looking at the spectrum from only the top 20 dBm range, which provides information of power which is significant only, the maximum standard deviation per point whether it is average of maximum is around 0.05 dBm. This is significantly minimal with respect to the double digit dBm values. For example, for a point of -30 dBm, the maximum standard deviation would only be 0.16%. This is basically insignificant. However, in statistical analysis, these numbers can be considered in terms of the capability of our system. For a system running multiple trials for a certain data, capability processes can be performed to analyze the capability and reliability of the system. In the medical and automotive industry, it is widely accepted that a capability, Cp, index of around 2 is good for a system. Cp can be calculated as Cp = Tol / NT, where NT is 6 times the standard deviation. The 6 signifies that ideally, we want our system to yield data that can grow up to 6 standard deviations without touching desired limits, giving our system precise stability. For our data this week, a maximum standard deviation of around 0.05 dBm and an ideal Cp of 2 gives us a tolerance width of around 0.6 dBm. This means that for each trace, we are potentially giving a set tolerance of +/- 0.3 dBm per point. I believe this makes our system pretty consistent given that this calculation is based on a maximum. If this calculation was made for our best settings, as shown above, we could set a tolerance of +/- 0.01 dBm, which is amazing!

Plans for next week (after spring break):

1. Use our best settings to test with distilled water.

2. Verify that our stability is consistent with distilled water and glucose solutions.

3. Gather data for 20 dBm range only, ignoring lower power values.

4. Discuss statistical analyses in order to move forward with glucose data and system stability.