Customer Handoff & Final Project Documentation
Table of Contents
Team Vision for Final Demo and Handoff
During the final phase of MSD II, the team planned to thoroughly characterize the performance of the signal receiving PCB in the integrated system, develop the data collection code to include 8-channel functionality and modulator calibration, integrate the laser with the custom controller, develop a user guide, organize documentation into a single customer handoff folder, prepare a functional demo for ImagineRIT, finish the paper, and assess final system performance vs. engineering requirements.
All of these goals were successfully accomplished. During PCB testing, the team discovered that its performance was inferior to that of the high-end amplifier and processing system used previously. Its performance was investigated in detail, allowing the team make recommendations for future system revisions. All other components worked as expected, and the team was prepared to hand off all materials to the customer by the end of this phase.
The System Architecture document can be viewed here.
Test Results Summary
The Test Plan Document can be viewed here
Alpha Wave Signal Recovery Testing
The Alpha Wave Signal Recovery Testing Document can be viewed here.
- The purpose of this test was to assess the system’s ability to resolve an alpha wave signal from raw EEG data. This test would serve as an initial proof-of-concept to show that optical sensors have the capability to detect EEG signals.
- A raw alpha wave signal was collected from a test subject. This data was filtered with a bandpass filter and a pass band of 7-15 Hz to check for the presence of alpha waves. The raw signal was then loaded onto a signal generator and fed through the optical system. Both the Femto amplifier with PowerLab system and the custom amplifier and microprocessor were used for testing, as the custom amplifier had demonstrated inferior signal recovery in the past.
- Results are displayed in Figure 1 below. Figure 1a shows the raw data. At 12 seconds, a large artifact is visible as the subject’s eyes close. When the data is filtered in Figure 1b, alpha wave activity is clearly visible, with the largest peaks occurring around 80 uVPP (representative of a very large alpha wave signal), and smaller peaks occurring around 50 uvPP (a more typical alpha amplitude). When this signal is passed through the optical system in conjunction with the Femto amplifier and PowerLab system, the alpha activity is preserved as shown in Figure 1c. Finally, when the optical signal is instead processed by the custom amplifier and microprocessor, the signal is less well resolved, and the noise baseline is higher. It is important to note that these results are not entirely physiologically relevant, as alpha waves are collected as a differential measurement between two electrodes, whereas this alpha wave signal was just passed into a single electro-optic modulator for detection.
Figure 1: Alpha wave collection and filtering (a-b) and subsequent recovery by the optical system (c-d).
- Large amplitude alpha waves are well-resolved by the optical system when used with the Femto amplifier and PowerLab. Further work needs to be done to detect these signals with the custom PCB.
Custom Laser Controller Testing
The Custom Laser Controller Testing document can be viewed here.
- The purpose of this test was to integrate the laser into the laser mount and custom temperature and controller so that its performance could be compared to the high-end, tunable laser controller used in previous testing.
- Prior to connecting the laser, the PCB was tested thoroughly to ensure that the temperature control component of the system was functioning properly and to verify that the maximum current limit of 145 mA was never exceeded on the laser. Each of the pins were checked thoroughly to ensure that the custom Dsub 9-pin cable was built correctly. After this system was fully checked, the laser was integrated.
- The laser pins were cut appropriately, and the laser was soldered by hand onto the laser mount. A cooling fan was also integrated into the temperature controller for maximum temperature stability. The fan and the controller itself were powered by a 12 V power supply. The power output of the laser was monitored by a ThorLabs power meter for stability. The full integration is shown in Figure 1 below. The laser power output was compared to data obtained from the high-end Agilent controller previously.
- The team was seeing considerable laser drift over time at high power with the Agilent laser controller. This drift is shown in orange in Figure 2 below. After the laser was integrated into the custom controller PCB, this downward power drift stabilized, as shown in blue in Figure 2.
Figure 2: Laser power vs. time in the high-end Agilent laser controller (orange) compared with that in the custom controller PCB (blue).
- The laser is expected to perform well in the custom controller PCB, thereby eliminating the need for the expensive Agilent tunable laser controller.
Custom Signal Receiving PCB Testing
The Custom Signal Receiving PCB Testing document can be viewed here.
- The purpose of this test was to assess the performance of the custom signal receiving PCB in the full integrated system and to compare its minimum resolvable voltage to that of the Femto amplifier and PowerLab system.
- The best-performing modulator (JDSU 10020480) was used for this test, as the purpose of the test was to assess the system’s ability to handle very small input signals and this modulator is able to detect small, EEG-relevant voltage changes. Minimum resolvable voltage was recorded using the standard protocol for a variety of systems, including the Femto amplifier with PowerLab system, the entirety of the custom signal receiving board, and combinations of the two. Minimum resolvable voltage was recorded both with and without the splitter. With the splitter, each channel received approximately 1.6 mW of power. Without the splitter, 4.5 mW of power could be directed into a single photodiode before the DC photodiode output hit the rails of the amplifier.
- Test results for each system are summarized in Table 1 below:
|System||Min. Resolvable Voltage (uVPP)|
|Femto amplifier and PowerLab, + splitter||50|
|Femto amplifier and PowerLab, - splitter||25|
|Custom amplifier and microprocessor, + splitter||375|
|Custom amplifier and microprocessor, - splitter||150|
|Custom amplifier, no filter, and PowerLab, + splitter||150|
|Custom amplifier, custom filter, and PowerLab, + splitter||100|
These minimum resolvable voltage specifications are slightly worse than they were prior to polarization controller removal, suggesting that polarization maintaining fibers are imperfect at maintaining the optimal polarization in the system. Additionally, minimum resolvable voltage was a factor of two worse when integrated with the splitter. These results suggested that a higher power laser may be necessary to obtain sufficient sensitivity on an eight-channel system unless the custom designed electro-optic modulators can achieve large improvements in resolution.
Finally, the performance of the custom signal receiving PCB was inferior to that of the Femto amplifier and PowerLab processing system. Improvements to minimum resolvable voltage were made when the signal was routed through the amplifier and filter of the custom board but then processed by the PowerLab at a 1000 Hz sampling frequency. This suggested that both the low-pass filter on the PCB as well as the sampling frequency of the microprocessor should be improved on the custom board in order to achieve the desired resolution.
The sampling frequency hypothesis was tested by lowering the sampling frequency to 200 Hz in LabChart, which is close to that of the custom PCB. Figure 1 below shows noise followed by a 10 Hz sine wave sampled at 200 Hz, while Figure 2 shows the same sampled at 1000 Hz. The 1000 Hz signal was better resolved than the 200 Hz one, suggesting that increasing the sampling frequency would indeed benefit the system.
- Several system improvements may need to be made to the system to improve resolution, including an increase in laser power, a higher-order low pass filter, and a higher sampling frequency capability in the microprocessor of the custom PCB.
EEG Cap Comfort Testing
The EEG Cap Comfort Testing document can be viewed here.
ENGINEERING REQUIREMENT: Percent of users who confirm comfort
Marginal Value: > 80% users Ideal Value: > 95% users Performance: ~80% users
- 70% of users reported the design of the cap as ‘comfortable’ or ‘very comfortable’
- 60% of users reported the chin-strap of the cap as ‘comfortable’ or ‘very comfortable’
- 10% of users reported the chin-strap of the cap as ‘uncomfortable’
- 60% of users reported the application of the cap as ‘comfortable’ or ‘very comfortable’
- 100% of users reported the cap to be ‘snugly fit’
- 90% of users reported feeling comfortable wearing the cap for 1 - 2+ hours
- 80% of users reported the overall comfort of the cap to be ‘comfortable’ or ‘very comfortable’
- Place cap onto subject’s head
- Clip chin-strap into place
- Adjust cap size as necessary
- Ask subject to complete comfort survey
The Calibration Test Document can be viewed here.
The Calibration Test Data can be viewed here.
- A test was performed to validate the effectiveness of the calibration routines created for data collection with this system.
- May 15, 2016
- Plug the laser into the optical channel splitter
- Plug two of the split channels into the inputs of two distinct modulators
- Plug the modulator output channels into channels 2 and 4 of the PCB
- Power the laser using the custom controller circuitry
- Using two separate bias controllers, set the bias voltage for each modulator so that it runs at its optimal operating point. Apply the bias for 15 minutes to help counteract bias drift.
- Apply a 10Hz, 0.025V sine wave into each of the two modulators and collect roughly 30 seconds of data. If only one sine wave generator is available this will need to be done twice.
- Disable the sine wave input and run the DC calibration software
- Repeat step 6
- Disable the sine wave input and run the AC calibration software
- Repeat step 6
- The signals collected during this procedure are available in the Calibration Data directory. Before calibration each channel reads its signal as being centered around a different voltage value and as having a different amplitude. After DC calibration each signal is centered near zero, though continuous bias drift introduces small errors into this baseline. After AC calibration each signal remains centered near zero, but now share similar amplitudes (each roughly 0.004 units).
Final Project Documentation
The P16071 Technical Paper can be viewed here.
Imagine RIT Poster
System User Guide
The System User Guide Document can be viewed here.
Schematics, Code, and CAD Files
The Schematics can be viewed here.
The Code can be viewed here.
The CAD files can be viewed here.
Test Plan Documents
The Test Plan Document can be viewed here.
The Test Plan for MSD II Document can be viewed here.
Problem Tracking and Project Risks
The Problem Tracking Document can be viewed here.
The Next Team's Project Risks Document can be viewed here.
Performance vs Requirements
The Performance vs Requirements document can be viewed here.
MSD 2 Self-Critique
The MSD 2 Self-Critique document can be viewed here.
The Gate Review presentation can be viewed here.
Recommendations for Future Work
The Recommendations for Future Work document can be viewed here.
- Modulator design
- The next modulator will needed to be patterned on a lithium niobate wafer using the standard semiconductor manufacturing tools (most of which are available at RIT)
- The designers of Photrode recommend the following paper for detailed design, fabrication, and testing procedures: http://ece562web.groups.et.byu.net/notes/lnbo3_modulator.pdf
- Further research is needed to understand the modulator performance differences observed between 10020417 and 10024180 modulators. The test standards sheets sent from JDSU indicate no major differences between the two. EOSpace specializes in low-noise modulators, so it may be worth consulting them.
- Fiber-modulator interfacing is not normally done at RIT, and therefore may need to be contracted out to a separate company to achieve low insertion loss. Low insertion loss will be extremely important in future designs to achieve the desired resolution.
- Polarization controllers are undesirable because they make the system more susceptible to motion, but they may need to be reimplemented to achieve the desired performance.
- The two paths of the waveguide’s interferometer will need to be tuned precisely based on the laser’s wavelength to ensure that they recombine at the desired point on the bias curve. The team found that maximum SNR was achieved near (but not at) the minimum point on the intensity curve, rather than at the half-intensity point. If fabrication tolerances are a concern, it may be possible to pattern large numbers of interferometers on a single wafer and down-select based on measured path lengths, as RIT’s measurement tools likely have superior resolution to their patterning tools.
- Signal receiving PCB design
- Increase the voltage rails of the amplifier and filter. This would have the benefit of being able to use a higher laser power, which could possibly allow the system to resolve a smaller EEG signal. To increase these rails, new op-amps would have to be chosen, since the op-amps used only run off of 5 V maximum. Also, the ADC would have to be reconfigured to expect in the new, higher voltage, or possibly also be replaced with a different part to compensate.
- Filter could be improved to have a steeper cutoff after 100 Hz, in order to better attenuate immediate noise frequencies. This would require the filter to be redesigned as a higher order filter, or for additional filters of the same type to be added in cascade. It should be noted that this improvement is mostly speculation, so prototyping should definitely be done before a full implementation to prove that the change is needed.
- Red wire on SPI-UART schematic could be tied to the digital voltage plane in the PCB, or a Reset button/switch could be implemented if desired, if a further iteration is done.
- Use a different microcontroller entirely, in order to have access to a higher clock speed. This would increase the sampling rate that could be used, which in testing, appears to improve the performance of the system.
- Use a microcontroller that has an implemented USB protocol, and possibly that includes an ADC that would be comparable to the 8-channel ADC used.
- Red wire seen on UART-USB schematic should be tied to ground in the PCB, if a further iteration is done.
- Include a screw-in header for the power input wires, so the wires have a cleaner connection method. Also, a power switch could be implemented between the power wires, and the jack, to implement a cleaner method of power cycling the board.
- An aluminum shielded box should be used to enclose the system, as it would provide a stable attaching surface for all external connectors, and would help block out noise from external sources.
- Signal Capture Code
- Allow the calibration process to skip unused channels.
- Create an installation script which automates dependency downloading and compiling.
- Switch data visualization and saving to python (pipe the input data) in order to perform live filtering and remove disk IO as a potential limiting factor on frame rate. This would also help in removing the unix dependencies.
- Allow for user-specified filters (expand example functionality from the AC calibration file)
- Higher laser power may be needed for an 8 channel system to ensure that each modulator receives sufficient input power. Minimum resolvable voltage was cut in half when the input laser power was increased from 1.6 mW (the maximum power from each channel of the 1x8 splitter) to 4.5 mW.
- It is important to understand that there may be diminishing returns in increasing laser power, as the team recorded increases in laser noise as power was increased.
- A splitter with a lower insertion loss would also improve power transmission through the system. The current splitter’s insertion loss is 10-11 dB, which is fairly high.