2021-2022 Capstone Project: Various RF, Analog/Mixed-Signal, and Machine Intelligence Projects

This year’s Capstone project can be conducted solely through on-line means due to on-going COVID-19 related concerns plus the fact that many students are not on campus this Fall. Students should have high speed internet enabling them to remote desktop onto campus computers.

Three specific project areas are available, each appropriate for 1-2 students. All students under my supervision will meet together through zoom calls and there will be an overall team dynamic with individual responsibilities.

Project area #1: RF and Analog IC design

Specific RF and analog IC design projects that may be of interest:

• Wideband tunable bandpass filter based on active inductors. The team will design and implement a high-Q bandpass filter that uses a "simulated" or "virtual" inductor comprised of a gyrator. This type of circuit is useful for software defined radio applications.
• Low-power SAR ADC using non-binary weighted capacitor arrays. This is a type of ADC that can achieve high resolution at low power. This project involves mixed-signal circuit design and would be good for students interested in a career designing ICs for biomedical applications.
• Other RF and analog IC design project suggestions will certainly be considered if I think I could supervise the project to a successful completion.

Project area #2: Digital IC design

• Machine Intelligence accelerator. This project will create an IC to accelerate the training of neural networks. Specifically, the implementation of CRAM (Computational RAM) will be a focus point for the project. This type of memory can operate on its own contents, without needing to send data over a bus where it is processed elsewhere and then returned to the memory.
• FPGA-based Cryptocurrency miner. This project will develop Verilog code for mining bitcoin, with the goal of deploying the code to an FGPA and mining coins.
• Other Digital IC design project suggestions will certainly be considered if I think I could supervise the project to a successful completion.

Project area #3: Machine Intelligence for Silicon Photonics

Students in this sub-project will be co-supervised by Dr. Ye.

The goal of this project is to use artificial neural networks (ANNs) to accelerate the design of silicon photonics components. A silicon photonics structure will be parameterized and multiple versions of the structure will be simulated using a commericial finite difference time domain (FDTD) simulator (Lumerical). The performance of the parameterized structure will be evaluated so that for each set of input parameters there is a corresponding set of simulated output parameters. An artificial neural network will be trained using machine learning techniques and the simulated data. The resulting ANN will be used to accelerate the design flow of similar parameterized silicon photonics structures by avoiding many of the FDTD simulations. Part of the project will be to determine the energy savings and increased computational efficiency and speed of the ANN-based design flow compared to the conventional FDTD-based design methodology.

This project is for students who are interested in machine learning applications, photonics and silicon photonics, and neural network applications. Students should have some knowledge of Python (the project will use PyTorch). Knowledge of finite element simulations and Lumerical would be an advantage. 

2020-2021 Capstone Project: Collaborative IC Design Using Augmented Reality

This year’s Capstone project will be conducted solely through on-line means due to on-going COVID-19 related physical distancing, unless the situation with COVID-19 changes so that in-person meetings are possible later.

While remote desktop environments and screen sharing while video conferencing over Zoom or Microsoft Teams will enable us to collaborate, a more fluid collaborative environment should be possible using XR technologies (XR is either augmented reality or virtual reality).

A part of the project will be the use of XR communications to facilitate group meetings, design collaboration, and design reviews. If there are students in the group primarily interested in augmented reality or virtual reality technologies, development/use of XR technologies (hardware and/or software) can be the focus of the project for those students.

This Capstone project will therefore discover the following for small engineering teams, academic research groups, and lab groups of students seeking a more fluid on-line collaboration and educational environment:

·      Appropriate types of end-user hardware (e.g. AR, VR, form factor);

·      Requirements for networking infrastructure (e.g. bandwidth, latency);

·      Available software versus what is required for streamlining the user experience;

·      Server-side requirements.

I found a YouTube video that shows an AR interface similar to what I was thinking about when drafting this project description:

 

Otherwise, the project involves analog, digital, or mixed-signal integrated circuit design. Cadence tools will be used (Analog Design Environment, Composer, Virtuoso, etc.) along with Synopsys tools (e.g. Design compiler). 

Specific IC design projects that may be of interest:

• RFIC direct down conversion receiver using non-coherent local oscillators. This type of receiver avoids DC offsets found in "traditional" direct conversion receivers by using a spread spectrum LO. Students will design the various receiver building blocks including LNA, mixers, LO generator, AGC, phase detectors, and filters.
• Wideband tunable bandpass filter based on active inductors. The team will design and implement a high-Q bandpass filter that uses a "simulated" or "virtual" inductor comprised of a gyrator. This type of circuit is useful for software defined radio applications.
• Low-power SAR ADC using non-binary weighted capacitor arrays. This is a type of ADC that can achieve high resolution at low power. This project involves mixed-signal circuit design and would be good for students interested in a career designing ICs for biomedical applications.
• Machine Intelligence accelerator. This project will create an IC to accelerate the training of neural networks. Specifically, the implementation of CRAM (Computational RAM) will be a focus point for the project. This type of memory can operate on its own contents, without needing to send data over a bus where it is processed elsewhere and then returned to the memory.

Other IC design project suggestions will certainly be considered if I think I could supervise the project to a successful completion.

Capstone Project for 2018-2019

Ok folks, I finally found some time to post some project ideas for 2018-2019. I'd be happy supervising any one of the projects mentioned below.

It'd be best if I were approached by a team of 4-6 people that form a cohesive project team. 

Even better would be if the team members have some complementary skills. For example, if some team members were good at (or interested in learning) embedded coding while others wanted to focus on analog circuit design, and still others wanted to approach the project from the system level, then that usually works out the best.

The suggested projects are:

Category #1: IC Design Related Projects

These projects involve intensive analog, digital, or mixed-signal integrated circuit design. Cadence tools will be used (Analog Design Environment, Composer, Virtuoso, etc.) along with Synopsys tools (e.g. Design compiler).

This category of project is directed at students who are interested in possibly conducting an MASc in the area of integrated circuits, and then continuing on into an industry career in this area, or continuing on into a PhD conducting research into integrated circuits.

Specific IC design projects that may be of interest:

• RFIC direct down conversion receiver using non-coherent local oscillators. This type of receiver avoids DC offsets found in "traditional" direct conversion receivers by using a spread spectrum LO. Students will design the various receiver building blocks including LNA, mixers, LO generator, AGC, phase detectors, and filters.
• Wideband tunable bandpass filter based on active inductors. The team will design and implement a high-Q bandpass filter that uses a "simulated" or "virtual" inductor comprised of a gyrator. This type of circuit is useful for software defined radio applications.
• Low-power SAR ADC using non-binary weighted capacitor arrays. This is a type of ADC that can achieve high resolution at low power. This project involves mixed-signal circuit design and would be good for students interested in a career designing ICs for biomedical applications.
• Machine Intelligence accelerator. This project will create an IC to accelerate the training of neural networks.

Other IC design project suggestions will certainly be considered if I think I could supervise the project to a successful completion.

Category #2: Embedded Systems and Wearables

These projects utilize off-the-shelf components to build "cool stuff". Generally some custom mixed-signal circuitry is designed and implemented and used for sensing and control within the target application. An embedded controller (micro controller, Raspberry Pi, FPGA, etc.) is used to collect and process data, and either act upon the data or give the user of the system actionable information.

Here are some concept projects I've been mulling over. If you have an interesting project in mind along these lines I'm certainly willing to hear more about it as well.

• nIRS-based imaging. Two years ago I ran a project for early onset Alzheimer's detection using nIRS. We were building the sensor system and processing system that would inject near infrared light at two wavelengths into the skull and measure the returned signal for each wavelength. From these values we could track the oxygenated blood flow in the illuminated areas. This type of information can be useful for potentially detecting early onset Alzheimer's symptoms. The group got fairly far along with the project and did very well in the departmental competition. There was still work left to do and so I continued on with the project last year.  For last year's project we started mostly from scratch and rebuilt the imaging system using near-infrared light, a different processing structure, and a different architecture for the system which was an improvement on the previous year's work. The target application for last year's group was cognitive load evaluation. In other words as cognitive load increases the blood flow to the prefrontal cortex changes and can be tracked. This group also did extremely well and came close to completing the project but not quite. So this year I'm very interested in seeing this project through to final completion which would involve successfully imaging biological structures, for example circulatory systems, using the infrared light. This is a challenging project involving analog and mixed-signal circuit design, sensor interfacing, micro controller programming, and signal processing.
• Open-source heart rate monitor.  When I go to the gym I use the Wahoo TickrX heart rate monitor which pairs with my phone and collects my heart rate as I work out. From that it can calculate calories burned and a few other metrics. I would love to develop an open source heart rate monitor that is actually useful in the gym environment. This will involve acquisition of the ECG signal, signal processing, sensor interfacing, and probably some machine intelligence and neural networks. I well know the limitations of my current gear and I have some ideas on how to improve the mechanism so we could discuss those ideas.
• Electromyographical measurement system.  The system will measure muscle activity electronically. I won't say much more about it here except to mention that it will be a fascinating project but very challenging.  As with the other projects listed above this one is multidisciplinary and requires analog and mixed-signal circuit design, sensor interfacing, embedded programming, and signal processing.
• Confocal imaging system. I have a high-powered microscope that I would like to retrofit and turn into a confocal imaging system. This will involve some mechanical design and 3D printing, a control system to sweep the focus of the lensing system, and some imaging hardware and signal processing that will take the image sequences and build the 3D model. 
• Other ideas:  I've quite a few potential projects floating around so I didn't list them all here but could discuss them with students interested in fairly complex multidisciplinary projects. 

( Please excuse typos in the above text – I was using voice dictation.)

Capstone Project for 2012-2013

Project: Prosthetic hand controller with gesture recognition and enhanced coordination.

The objective of this project is to research, design, and implement a below-the-elbow prosthetic hand and control it with an EMG-based gesture recognizer. The EMG gesture recognizer will have extremely low latency, meaning that the user will experience no discernible delay between muscle activation and prosthesis motion.

The basic type of prosthetic limb I am imagining is demonstrated in the following videos:

In the first video below, we see a new user training her device. While it is not explained in the video, I can see some EMG electrodes mounted on the sleeve that she is wearing. These electrodes pick up the faint EMG signal, and then some processing is performed resulting in the waveforms shown on the screen during the video. Depending on which muscles she activates, she can get the hand to open or close or move to various positions.

In the next video below we see a more advanced hand that displays more "human-like" motion. I'm only guessing here, but my instinct is that the motion is pre-programmed to look more realistic. In other words, the user initiates a movement and then the on-board controllers perform the movement in a human-like motion pattern. I think this is a better approach.

I love the sophistication shown in the next video. In fact when I first viewed the video, I was initially confused over whether the person portrayed was simply wearing some sort of glove over his real hand. There are couple of screenshots showing the electromyographic signal on a computer screen. If you read the comments it is explained that these weak signals are in fact interpreted by the hardware so that the motion appears quite lifelike.

Project Breakdown:

The previous videos should give some idea of the target application in the sphere of EMG-controlled prosthetic hands. A tremendous amount of work was performed last year regarding the EMG subsystem. In fact, that work is still on-going due to the wonderful Capstone Award that the group received last year. I'd like to use last year's MuscleMate project results as a basis for this year's project. Since last year's project is in the process of being commercialized, we would have to "license" their current technology for inclusion in this year's project. An alternative is to use the final project reports from last year and built upon those directly. Either way should work well.

A quick summary of the tasks required are as follows:

• Implement an appropriate EMG biopotential sensor along with the requisite signal processing chain up to the ADC stage.
• Implement a prototype prosthetic hand. The person(s) doing this should have some mechanical design and fabrication experience, perhaps through a robotics club or similar.
• Implement the primary micro controller, including the gesture recognition algorithms and signal processing algorithms.
• Implement the motor drivers and one or more micro controllers to control the various motors.
• Implement the power management circuitry.

That is a lot of work and I believe it would take a bare minimum of three completely focused and brilliant students to accomplish the above in the time frame available. More reasonably, four or five students would form a great team, as long as they have some inclination and ability in the required areas.

What follows is a better description of each of the aforementioned tasks.

Subproject: Biopotential Sensor and Amplifier Design 

I've been running projects for the past few years that involve acquisition of biosignals, in particular ECG and EMG signals. When you interface your body to a computer, and for the first time see the signals that you produce within yourself materialize on the oscilloscope, well, let's just say that you experience a unique “oh wow” feeling. 

Unfortunately, collecting biopotential signals is not a simple as contacting a wire to your skin and connecting the other end of the wire to the acquisition unit. The biopotential signals are extremely small, on the order of micro volts to millivolts and so they must be amplified to be usable. Frustrating the acquisition of such signals is the presence of noise such as 60 Hz power line hum that permeates our typical indoor environments. Of no less importance and equally frustrating, when the sensor material makes contact with human skin there exists a chemical reaction at the interface, and this chemical reaction in and of itself can produce a highly variable unwanted additional signal component which may be orders of magnitude larger than the desired biopotential signal. 

The figure below contains a simplified circuit model representative of the electrode–electrolyte interface that exists between the sensor and the skin surface. The series resistance, labeled as Rs, models the impedance of the skin. Rs is highly variable amongst individuals, and even for the same person will change depending for example upon the sensor location upon the body or the skin preparation. The parallel combination of resistance and capacitance labeled Rd and Cd model the path for leakage current between the skin and the electrode, and the charge buildup between electrode–electrolyte interface, respectively. 

Simplified model of the electrode to skin interface. You can read more here.

The other component shown in the figure above models the half cell potential that arises from the chemical reaction that occurs at the interface between the sensor and the skin. Since the half cell potential arises from a chemical reaction, it is not surprising that the potential voltage that is realized is a function of the materials involved in the chemical reaction, the temperature and pressure at which chemical reaction occurs, and the point in time at which we monitor the chemical reaction. In other words, the half cell potential which is modeled as a battery in the above figure can be a highly variable source of frustration as it represents a signal quantity over which we may have limited control.

If you do a Google search for the term “half cell potential” you will quickly find tables that catalog half cell potential for various materials, such as the table shown below. Generally we want to minimize the half cell potential so that the maximum gain can be applied to our desired signal. If, for example, the half cell potential is 0.5 V, and our desired signal is on the order of the micro volt, then it becomes difficult to directly amplify the desired signal without saturating all of our analog signal processing stages. Some form of filtering or offset adjustment is required before the desired signal can be amplified. Alternative techniques exist, such as those that employ drive circuitry, but these techniques are not always applicable in every situation. 

Some half-cell potentials. A more comprehensive list is here.

During the course of last year’s project (“MuscleMate”), we experimented with a variety of sensor configurations. We tried to fashion our own so-called dry electrodes using zinc coated staples, with all of the analog differential gain circuitry co-located with the sensor. Last year’s students fabricated the sensors and tested them, but in the end we had to abandon their usage because of the large variable unwanted half cell potential. One student eventually tracked down a problem: the so-called zinc plated staples we were using were virtually devoid of zinc by the time they were deployed. 

Differential EMG electrode with integrated amplifier. Created by the 2012 Capstone MuscleMate students.

In order to get the project moving along, we resorted to the tried-and-true Ag-AgCl conventional wet electrodes such as those shown below. These particular electrodes employ a solid hydrogel saturated with AgCl enveloping a silver coated electrode. We wanted to avoid using these types of electrodes because they are only supposed to be single use (although we could use them multiple times). 

Conventional solid gel (hydrogel) Ag/AgCl electrodes.

So what's the subproject exactly?:

This year I am looking for a student who can help me solve some of the aforementioned problems associated with acquiring biopotential signals from the surface of human skin. I have some ideas about using analog circuits to correct some of the problems discussed above, so you can expect to do some circuit design and implementation. You can expect to do some experimentation in the lab, as well as simulations using Cadence tools, Agilent ADS, and perhaps ANSYS. You should be creative and able to synthesize prototypes using available materials. For example, several years ago a student who was working with me took a pair of earbud style headphones, removed the speakers, and inserted snaps that she had obtained from a local craft shop that conveniently mated with the style of conventional biopotential sensors we were using at that time to measure ECG signals. It worked like a charm.

The ultimate objective of the project is to research, design, and implement biopotential sensor-amplifier combinations that intelligently track and accommodate inherent half cell potentials. The primary application area of interest is electromyography, but EEG and ECG signals are also of interest and will be considered, time allowing. We will also want to design the sensors so that external interference pickup is minimized. For example power line hum pickup should be reduced as much as practical.

To give you an idea about the circuit complexity, below is a photo of last year's analog signal processing chain (taken from Mark Klibanov's final project report).

How this subproject fits into the grand scheme of things:

This subproject is a part of, and integral to, the prosthetic hand project. You will combine the sensors produced in this project with the hardware designed by the other students.

Subproject: Prototype Prosthetic Hand 

This subproject would be undertaken by the entire group of students, and lead by one or two students who are handy with fabrication and have some robot design experience. The hand design will need to be determined as early as possible in the course of the project. In case you're thinking "Impossible, a group of final year students can't build a prosthetic hand", well, it's been done before. For a simple hand that could be built as a demonstrator, check out the following video.

Come to think of it, it would be pretty cool to create a hand that enables someone to play piano again.

Subproject: Primary Microcontroller 

This subproject involves the design of the main brain for the prosthetic hand. The primary micro controller functions to take the digitized electromyographic signal and interpret it terms of muscle activity to the point that specific gestures can be recognized. For example, it would be useful to recognize a gesture  that would place the prosthetic hand into a position useful for  controlling a computer mouse. Imagine if you could program the arm so that you could do chords when playing guitar. The idea is to build up a repertoire of gestures mapped to specific muscle activation sequences so that the user of the prosthetic device can effect fluid motion of the hand with minimal concentration.

The person responsible for the primary micro controller should have some advanced skills with embedded controller design and programming. The student (Nick Stupich) who worked extensively on the micro controller last year was well-versed in C programming and used Python extensively. You should expect to code your own signal processing routines, such as an FFT routine.  You need to have some knowledge, or be extremely interested in learning about, classification algorithms and machine learning. Last year's group used support vector machines within the muscle activity recognition algorithms.  I expect that this year students will also use a similar classification algorithm.

The primary micro controller functions as the director for the motor control micro controllers.  Once a specific gesture is recognized, the primary micro controller will issue the necessary commands to the secondary controllers in order to  cause the prosthetic hand to move into the desired position.

Last year's group used a FEZ Mini micro controller as shown below. One of the first tasks for this year's group will be to choose an appropriate micro controller for the project. It may even be possible to use a single board computer for the task this year.

FEZ Mini Microcontroller.

Subproject: Secondary Microcontrollers and Motor Drivers 

The secondary micro controllers will take their orders from the primary micro controller. Their function will be to provide smooth motion of the articulated prosthetic hand via a yet to be determined mechanical implementation. At this point I'm imagining that the micro controller or micro controllers in charge of the various motors will implement PID  controllers for each of the motors. I haven't done much research on this subject yet but I suspect that there are commercially available motor control chips that could be put to great advantage in this application.

The student in charge of  this subproject will also be responsible for choosing appropriate  motors for the prosthetic hand. Since the entire project will be run on a shoestring budget, it will be difficult to get the best performance possible but we should at least be able to create a functional prototype if we choose our implementation components carefully.

Subproject: Power Management Circuitry 

Power management is the under appreciated lynchpin of the electronics world. This is a crucial portion of the overall design. No one wants to carry a car battery around in one hand so they can have an electrically powered prosthetic hand.


The student responsible for this portion of the design will implement circuitry that provides power for all of the other components of the prosthetic hand. This would include the micro controllers  as well as the motors used to activate the hand. No less important, the student in charge of the power management circuitry will also perform power budget analysis for the entire device as well as heat dissipation analysis for the entire device.

In all likelihood, the power management circuitry will be responsible for generating multiple voltages both positive and negative from a single positive supply. The student involved should have an appreciation for buck and boost converters, protective devices, and electrical noise reduction.

Qualifications for the Project:

No doubt about it, this will be a difficult project to undertake. It will take an excellent team of students to pull it off. In terms of the dream team I am looking to supervise, here is a list of desired skills and personality traits:

• Versatile and imaginative; 
• Autodidactic tendancies; 
• Interest and competence in chemistry and/or materials science (for the student working on the sensor); 
• Interest and competence in analog circuit design; 
• Possess a "love of tinkering" and appreciation for design and craftsmanship; 
• Interest in research and development;
• Focused and driven to success at difficult problems.

What you will get out of doing this project:

• Experience in analog and possibly mixed-signal circuit design. 
• Experience in bio signal acquisition hardware and software. 
• Increased knowledge in the area of biomedical engineering. 
• Increased knowledge in the areas of materials, electrochemistry, and human interfaces.
• Increased knowledge in the area of analog signal processing.
• Increased knowledge in noise analysis and low noise design.
• Increased knowledge in the area of motor control and digital controllers. 
• An appreciation for end-to-end hardware and software design.

If you are interested in this project, send me an email and we'll discuss it further. I'm looking for a team of three to five students. Last year's group approached me over the summer and the team was assembled and ready to roll before September. If you have a group of colleagues that you believe can pull off this ambitious project, just send me an email and we can all meet to discuss things.





Note: All figures above, except for the YouTube generated pictures, are taken from Mark Klibanov's excellent DOE Capstone 2012 final report.

Still Waiting?

Sorry...I had a bit of a calamity on my hands with my Carleton web page and had to deal with it. Almost done, then to populate this blog with the new goodies.

Out With The Old....

....and in with the new.

Last year's project was so intriguing that it is hard to let it go. Actually, the team is still working on it. We are making progress toward our goals. We're setting up a new web page to document the 2011-2012 MuscleMate  project and how it is evolving. More on that in a later posting.

I'm mulling over a couple of interesting projects for this coming year (2012-2013). So far the world has not ended in 2012, so maybe it is time I posted them here.

Please monitor this page over the next few days, and if something pops up that captures your interest don't be shy to send me an email.

Congratulations!

The 2011-2012 project team was successful on many fronts. The team won the following:

• Department of Electronics Best Presentation Award
• Department of Electronics Best Poster Award
• Carleton University IEEE Student Best Paper Award
• Eastern Ontario Regional IEEE Student Best Paper Award
• Carleton University Capstone Award ($40,000)

This summer the team will continue working on the project with an aim towards commercialization.

After exams are completed, this page will be updated with copies of the student reports, their posters, and a video of their award-winning presentation. For now, a brief summary of the work that was accomplished follows. 

An electromyography hardware acquisition unit was designed using commodity components with a view towards future full-scale ASIC integration. Surface electromyography was accomplished at forearm muscle sites using both dry and Ag/AG-CL differential electrodes. The acquired signal was amplified differentially and then processed using a combination of a bandpass filter and a dual notch filter. The filters were implemented using eighth-order switch capacitor filters with a reference clock frequency of 20 kHz. The bandpass filters retained the salient electromyographic frequencies of 20 Hz to 120 Hz. The notch filters were placed at the problematic 60 Hz and 120 Hz powerline interference frequencies. The processed analog signals were digitized using an eight bit analog-to-digital converter and then further processed for frequency content within a microcontroller. Bluetooth was used to transfer data from the microcontroller to a variety of Bluetooth enabled hardware devices including smart phones and laptops. Gesture recognition software implemented using support vector machines interpreted both the muscle activity intensity and the sequencing of muscle activation thereby providing gesture recognition. The system was designed to accommodate up to six sensors but the number of sensors can be increased by multiplexing inputs. Due to the inherent train-ability of support vector machines, the implemented gesture recognition algorithm may be trained for various applications as well as for different users of the implemented system.