M.Eng. Project List: 2018-19 Academic Year

DateFaculty MemberProject NameProject Summary
08/23/18Adie, StevenMultimodal ultrasound and optical coherence tomography imagingThe aim of this project is to develop an ultrasound imaging setup that takes advantage of existing 10 MHz ultrasound transducers and driver electronics in the group, to perform co-registered ultrasound and optical coherence tomography (OCT) imaging. This ultrasound hardware is currently being used for applying acoustic radiation force (ARF) for ‘palpation’ in optical coherence elastography (OCE), but could additionally be utilized for high-resolution ultrasound imaging. Our ARF-OCE experimental setup is already designed to achieve overlap of focused ultrasound and OCT beams, and our ultrasound pulse-receiver needs to be synchronized with our LabVIEW-based OCT acquisition software. The MEng student/s will work on the demodulation and digitization of ultrasound (‘echo’) receive signals, and synchronize these resulting ultrasound ‘A-scans’ so that they are co-registered with each OCT A-scan. Successful completion of the project will enable co-registered ultrasound imaging (providing deep tissue imaging, at ultrasonic resolution) and OCT (high optical resolution, with OCT imaging depths). The resulting multimodal imaging capability will provide significant flexibility for in vivo cancer imaging studies.
08/23/18Adie, StevenGPU-based processing for high-speed computed optical coherence tomographyThis MEng project involves the implementation and optimization of computed imaging algorithms for real-time reconstruction and display of images acquired with a spectral domain optical coherence tomography (SD-OCT) system. In particular, this project will implement high-speed processing for computational adaptive optics (CAO), which is a computational image formation method that can correct defocus and optical aberrations in OCT datasets. The CAO correction needs to be optimized in real-time for each plane in the reconstructed volume. These optimized real-time volumetric reconstructions are needed for high-throughput cell imaging studies.
09/11/18Apsel, AlyssaPrototyping Scalable Synchronization for Peer-to-Peer NetworksThe overall goal of the project is to build and demonstrate low-power communication of large-scale wireless networks for IoT applications. P2P nodes cannot rely on strong master-to-slave asymmetries for constant transceiver operation since all peers have limited energy resources. Instead, P2P IoT nodes must maintain average low-power operation by duty-cycling power hungry RF circuitry, which relies on scalable synchronization. Despite a range of commercial low-power radios, mesh networking for low-power IoT remains a challenge.
08/23/18Batten, ChristopherImplementation and Evaluation of Parallel Applications on Sharing ArchitecturesThe problem of underutilized hardware resources has been a long-standing challenge for computer architects who are tasked with efficient management at the software-hardware boundary. A well-known approach to address this problem is to collect underutilized resources (caches, networks, compute) into a small pool of highly utilized shared resources that are accessible to many workers at once. Our research group is developing novel sharing-based architectures. The goal of this MEng design project is to implement, verify, and evaluate various parallel applications executing on these new sharing-based architectures. The student will have access to research-grade hardware simulation platforms for evaluating the detailed performance of these applications. Near the end of the project, the student will have the opportunity to experiment with mapping parallel microbenchmarks to a recently fabricated test chip which includes a simple sharing-based architecture.
08/23/18Batten, ChristopherDesign and Evaluation of Various On-Chip Interconnection NetworksToday's embedded, network, graphics, and server processors already contain tens to hundreds of cores integrated onto a single chip, and this number will surely increase with future scaling. Onchip interconnection networks (OCNs), which connect cores on the same chip to each other and to main memory, are becoming a critical design component in terms of performance, power consumption, and programmer productivity. The following paper provides a good introduction to on-chip interconnection networks: http://dx.doi.org/10.1145/1183401.1183430 The goal of this MEng design project is to design, implement, verify, and evaluate various onchip interconnection networks at the register-transfer level suitable for mapping to an ASIC. The student will have access to a state-of-the-art commercial standard-cell-based toolflow and the PDK for an advanced technology node for evaluating the area, energy, and timing of these OCNs. Near the end of the project, the student will compose one or more OCNs with processors and a memory system to conduct a more complete evaluation.
08/23/18Batten, ChristopherPython-Based Hardware Design FrameworkHardware design is becoming increasingly complex due to growing design and verification costs. Previous work has argued that the best way to amortize these costs is to spend engineering resources on constructing reusable hardware generators rather than on specific hardware implementations. Hardware generators include highly parameterized models across various levels of abstraction that can generate many specific hardware implementations enabling a more rapid and productive design space exploration. Our research group is developing a novel Pythonbased framework, called PyMTL, that will enable computer architects to rapidly design, implement, test, and evaluate new research ideas. A recent research paper on the PyMTL framework is available here: http://www.csl.cornell.edu/~cbatten/pdfs/lockhart-pymtl-micro2014.pdf This MEng design project has two overall goals: (1) to develop a set of PyMTL tools that extend this framework in various ways; and (2) to implement an in-order superscalar processor model which will serve as a useful case study for the PyMTL framework.
08/23/18Batten, ChristopherDesign and Evaluation of a RISC-V Vector UnitRISC-V is a new open instruction set architecture that is growing in popularity across both industry and academia. RISC-V was explicitly designed to enable specialized extensions for various application domains. For example, the RISC-V vector extensions are designed to accelerate high-performance numerical computing. More information about these extensions was presented at a RISC-V workshop last spring: https://goo.gl/iPxxej https://www.youtube.com/watch?v=S4fxBZD79gc https://goo.gl/KNbM1u https://www.youtube.com/watch?v=ESu9NI3h1Y4 The goal of this MEng design project is to design, implement, verify, and evaluate a RISC-V vector unit at the register-transfer level suitable for mapping to an ASIC. The student will have access to a state-of-the-art commercial standard-cell-based toolflow and the PDK for an advanced technology node for evaluating the area, energy, and timing of the vector unit. Near the end of the project, the student will compose the vector unit with a scalar processor and memory system to conduct a more complete evaluation.
08/17/18Bitar, EilyanDeep Learning for Prediction and Control of Complex SystemsThis project will investigate how recent advances in “deep learning” might be employed to enable the effective identification, prediction, and control of complex stochastic dynamic systems, including autonomous robots and power networks. In particular, this project will explore the design and test the performance of model predictive controllers built on input-convex deep neural network representations of the underlying physical systems being controlled.
08/13/18Bojanczyk, AdamSpace-Time Adaptive Processing on NVIDIA GPUsSpace-time adaptive processing (STAP) is a signal processing technique for detecting airborne moving targets in the presence of clutter. STAP algorithms are composed of dense matrix operations. NVIDIA GPUs are well suited for dense matrix computations. This project will explore the feasibility of using GPU accelerators for speeding-up STAP computations.
08/13/18Bojanczyk, AdamTime-Difference of Arrival (TDOA) localization methods.Signal source localization plays an important role in array signal processing with applications in communication, sonar and radar systems. One approach to transmitter localization is to measure TDOA of the signal at spatially distributed receivers. The accuracy of the transmitter position estimate is sensitive to any uncertainty in the TDOA measurements. Recently Convolutional Neural Networks and Machine Learning techniques has been applied to enhance the accuracy of the estimate. The goal of this project is to apply these enhancements to the case of a moving transmitter.
08/14/18Delimitrou, ChristinaCreating representative end-to-end microservices applicationsMicroservices are a new paradigm in cloud computing services. Instead of the traditional monolithic application design, where all functionality is included in a single binary, cloud applications under the microservices programming model consist of 100s or 1000s of very small services, each responsible for a small fraction of the overall application functionality. The benefit of microservices is that they are much simpler to develop and debug. The disadvantage is that they complicate scheduling and resource management substantially since the cluster scheduler needs to consider a complex graph of dependencies between microservices. The goal of this project is to design a few representative end-to-end applications based on microservices that we can use to explore new scheduling and resource management mechanisms.
08/14/18Delimitrou, ChristinaTask scheduling for a swarm of heterogeneous dronesSwarms of IoT devices, specifically UAVs, can be used in many scenarios, including disaster recovery, weather monitoring, and space exploration. The actual edge devices have strict battery constraints, and limited compute and memory capabilities. Therefore, when the swarm of UAVs must execute a more demanding computation, it often leverages a backend cloud. The goal of this project is twofold. First, students will have to configure a local small swarm of programmable drones to move in a coordinated fashion, recognize images in their environment, and avoid obstacles. Second, once this application is set up, students will benchmark the difference in performance and battery-life for the following three setups: (i) when all computation and communication happens between the drones themselves with no centralized coordination system, (ii) when a local server at Cornell is used to coordinate the drones’ actions, and (iii) a hybrid scenario where part of the computation happens on the edge devices, and part on the backend cloud.
08/14/18Delimitrou, ChristinaScalable simulation of end-to-end cloud microservicesMicroservices are a new paradigm in cloud computing services. Instead of the traditional monolithic application design, where all functionality is included in a single binary, cloud applications under the microservices programming model consist of 100s or 1000s of very small services, each responsible for a small fraction of the overall application functionality. The benefit of microservices is that they are much simpler to develop and debug. The disadvantage is that they complicate scheduling and resource management substantially since the cluster scheduler needs to consider a complex graph of dependencies between microservices. The goal of this project is to leverage an existing queueing network simulator to capture dependencies between microservices in an accurate, and scalable way. Students will then use the simulator the evaluate scheduling policies, including autoscaling, and rate limiting and explore whether they can satisfy quality-of-service when the system is highly utilized.
08/14/18Delimitrou, ChristinaChallenges and opportunities of Serverless ComputeIn a conventional cloud computing environment, users deploy applications on the number and type of servers they deem necessary. A new programming model has recently gained traction, called serverless compute, where users only need to worry about writing application code while cloud providers manage deployment across servers and charge users only for the compute power they use [1]. Instead of thinking about applications as collections of servers, developers define applications as a set of handler functions which can be triggered by various events and have access to a common datastore. Examples of this new model include Amazon Lambda and Google Cloud Functions. Serverless benefits a wide range of emerging applications, including microservices, and IoT workloads. This project examined the challenges and opportunities of serverless computing for cloud and IoT services. Students will use an open-source serverless platform, Fission, and a few representative applications and compare performance, scalability, and cost versus conventional cloud systems. [1] Serverless Computation with OpenLambda https://www.usenix.org/system/files/conference/hotcloud16/hotcloud16_hendrickson.pdf
08/14/18Delimitrou, ChristinaLeveraging ML to automatically detect performance issues in cloud applicationsCloud applications are shifting from large monolithic applications that encompass all functionality, to complex graphs of hundreds or thousands of microservices, each corresponding to a small fraction of the end-to-end service. A major challenge with microservices is detecting where performance issues initiate and which microservice is the culprit of a quality-of-service violation. Distributed tracing systems, like Zipkin or Dapper, can help in this direction by collecting end-to-end request traces. The goal of this project is to apply ML and data mining techniques on distributed traces to automatically detect performance issues, and the microservices that initiated them.
08/13/18Doerschuk, PeterDeep Learning to select particles in electron microscopy imagesproblems for single-particle cryo electron microscopy, e.g., Zheng, Wang, Doerschuk, \Three-dimensional reconstruction of the statistics of hetero- geneous objects from a collection of one projection image of each object", Journal of the Optical Society of America Series A 29(5):959-970, May 20121 and Ma, Gong, Aubert, Turker, Kao, Doerschuk, Wiesner, \Surfac- tant micelle self-assembly directed highly symmetric ultrasmall inorganic cages", Nature 558:577{580, 28 June 20182. The applications are both to biological particles, especially viruses, and to nanotechnology particles. The electron microscope provides large images, e.g., 4000 4000 pixels, which show many instances of the particle of interest. The rst stage of the computational pipeline is to extract small subimages, e.g., 200 200 pixels, that contain individual particles. This problem is called \particle picking". This task is complicated by the low signal-to-noise ratio of the images and the complicated \non-particle" content of the images. The task is simplied by the fact that the dierent instances of a particle are of the same size. The goal of the project is to use modern Machine Learning techniques, especially Deep Learning techniques, to solve the particle-picking prob- lem. Conceptually, there are three qualitative steps that are necessary: recognize each object in an image, classify each object that is recognized, and draw a bounding box around each object. There was a workshop3 in April 2018 that considered Deep Learning for a variety of problems in biological single-particle cryo electron microscopy, including particle pick- ing. There is also a github page on all sorts of cryo electron microscopy software, including particle picking software4. Please note that the list of particle picking software on the github page is proably incomplete be- cause many of the large packages (e.g., Relion, EMAN2) include particle picking. I think we can improve on what was described at the workshop and what is listed on the github page!
08/13/18Doerschuk, PeterCombining generative models from Deep Learning with maximum likelihood estimators: Better answers from poorer dataGenerative Adversarial Networks (GAN) in Deep Learning have been quite successful in providing Deep Neural Networks that accurately represent particular image domains (e.g., human faces1). When solving an inverse problem, such as a tomography problem, if you have a good description of the image domain that contains the answer then you can compute a xed-quality approximation to the answer using poorer qual- ity data. In a tomography problem, the data could be poorer quality in many senses including lower signal-to-noise ratio, larger angular sam- pling interval, or limited range of angles. Some of these data issues are very important in applications. For instance, in medical X-ray computed tomography, using a lower dose of X-rays is desirable but causes lower signal-to-noise ratio which is undesirable. The purpose of this project is to test an idea for combining Generative Adversarial Networks with Maximum Likelihood statistical estimators in a tomography problem.
08/30/18Hammer, D. + Samsel, P.Determining the processing uniformity of a plasma processing system used for treating surfacesComplete the development of a diagnostic device that measures plasma properties and can be moved throughout the plasma volume; use that device to determine the uniformity of the processing plasma by collecting and analyzing experimental data that can be used to obtain plasma density and temperature; correlate that uniformity with the uniformity of the surface treatment. Important note: This project does not require a working knowledge of plasma physics. It is fundamentally a project to turn a single point measuring instrument into an automated multipoint instrument (probe on the end of an arm that can move in known steps in 2D or 3D).
07/12/18Hysell, DavidRadar Imaging Based on Compressive SensingOur research group uses radars with multiple, spaced antennas to construct 3D images of plasma waves and other phenomena in the Earth’s upper atmosphere. Imaging work performed in the past used methods based on optimal beamforming and on Bayesian inversion. A new imaging paradigm based on compressive sensing has recently emerged. The method has been demonstrated on synthetic data using a simple programming toolbox. This project involves developing robust imaging software and testing it with real radar data.
07/24/18Jena, DebdeepComputational Design of Quantum Semiconductor DevicesAdvances in the computational capacity and ab initio density functional theory has now made it possible to design and understand semiconductor materials and devices from the 1st principles. This goal of this project is the computational design of semiconductor materials and junctions for high efficiency and high frequency RF filters and transistors using nitride piezoelectric materials.
08/29/18Kan, EdwinMicrowave imaging based on RFID networkThe overall goal of the project is to use the commercial RFID system to perform microwave imaging in a specific capture volume, where the reader and tag antennas can be regarded as the observation points. As the RFID tag is very low cost (< $0.1), many tags can be profusely and redundantly deployed to provide the necessary spatial diversity in any imaging methods. From the multiple-input-multiple-output (MIMO) network point of view, the scatterer shape and location can be derived from the collected line-of-sight and multi-path signal profiles. The design project will start from configuring and programming the commercial RFID readers and tags to build a MIMO network, and use the extracted matrices of magnitudes and phases for still imaging. Applications derived from this technology include Internet of Things (IoT), smart buildings, body imaging, and covert object recognition.
08/16/18Kress-Gazit, HadasCreate Skills for Baxter RobotIn this project you will program skills for a Baxter robot, and potentially other robots available in the Autonomous Systems Lab, using the Robot Operating System (ROS). The skills you create will range from simple to complex, such as picking, placing, pushing, and other manipulation tasks. These skills will be used in the creation of symbols that allow the robot to determine when it can apply those skills. We are looking for a set of skills that can be easily and reliably run on a Baxter robot. The code should be able to be used in the future with reasonable ease and be well documented.
08/16/18Kress-Gazit, HadasIntegrate Depth Sensor with the iRobot CreateIn MAE 4180/5180 “Autonomous Mobile Robots” labs are performed with the iRobot Create. This project is to integrate a new sensor (an RGBD camera) with the robot, provide beacon and range information over wifi and integrate with the robot and a MATLAB interface used to control the robot. This project is intended for a team of two students.
08/21/18Lal, AmitSAW-Bulk gyroscope with controller circuit and graphene readoutGyroscopes are sensors that measures rotation. Consumer gyroscopes uses a suspended mass-spring system to measure the rotation. They are in almost every smartphone these days, but most gyroscopes suffer from low shock and vibration tolerance. On the other hand, surface acoustic wave (SAW) and bulk gyroscope has exceptional shock tolerance suitable for specific applications. In SonicMEMS laboratory, we design and develop a SAW and bulk gyroscope on lithium niobate (LiNbO3) piezoelectric substrates with interdigital transducer (IDT) electrodes or graphene piezoresistive pickups. Because gyroscope control and readout can dramatically improve gyroscope stability and noise, a well-design controller circuit is crucial.
08/21/18Lal, AmitDifferential equation solver using a spectral method in an Intel FPGAA differential equation is a function that involves derivatives which has applications in heat conduction, plasma, fluid dynamics, and quantum mechanics. Despite their wide range of application, only the simplest form can be solved explicitly and a numerical method should be applied to find the approximate solution in a computer. Many different methods such as the finite difference method and spectral method were developed to numerically solve the differential equation. The pseudo-spectral method is one of them that uses FFT and inverse FFT to solve the diffusion equation and the Vlasov equation. SonicMEMS lab is developing a spectral method on a Stratix 10 FPGA to improve the speed and accuracy of solving these equations.
07/30/18Land, B. + Dehim, D.Virtual Reality Biofeedback for MindfulnessYou will be working in a cross-functional team with students from Cornell and other universities to build a virtual reality mindfulness meditation platform for K12 education. The VR headset incorporates biosensors for various software applications including biofeedback visualization in the virtual environment and adapting the guiding mindfulness lesson audio in real-time based on user’s emotional responses. Your responsibilities will encompass the electronic assembly of the biosensors and circuitry in the headset. We are looking for 1-3 students.
07/13/18Land, B. + Hao, H.A Close-loop Brain-Heart Machine Interface SystemWe propose an affective Brain-Heart Computer Interface (BHCI) system for monitoring stress and detecting changes in emotion with the complimentary information from brain and heart. A BHCI paradigm includes three components: (i) Signal acquisition module. We use a portable commercial 4-sensor electroencephalograph (EEG) headband with sampling rate at 220Hz*. We also want to develop a portable electrocardiograph (ECG) sensing system that sends signals to a computer with sampling rate ~250Hz and which is easy to use in an office setting. The ECG must be small, wireless, and require minimal electrode preparation. (ii) Real-time signal processing module. As part of an MEng project last year, students designed and built a ECG monitor for this project. This year the project will include the EEG and feedback components, plus testing. The report is available to students who sign up for the project.
08/16/18Land, B. + Johnson, B.Convert Spike Hound from Matlab to PythonSpike Hound is a data acquisition, Stimulus Generation, Instrument Control, and real time analysis package designed for use in physiology research and education where continuous multi-channel data is acquired and discrete events are analyzed. The software allows for connection to a broad range of popular data acquisition hardware from high end, calibrated data acquisition boards to the sound card in your computer. Spike Hound allows for signal visualization, real-time filter applications, audio output to a monitor, and data logging with associated meta-data.
08/16/18Land, B. + Skovira, J.High Speed Nanomachines for Semiconductor MetrologyThe demand for low powered, high performance electronic devices has triggered the aggressive scaling down of transistors into the 10 nm node and beyond technologies. However, problems remain. Besides the complex lithographic challenges that are eminent at these nodes, device testing struggles continue to persist. At these extreme nodes, conventional optical techniques for failure analysis are ineffective due to wavelength limitations. Xallent will extend device failure isolation and analysis into the 10 nm node and beyond by developing high speed and cost effective nanomachines for semiconductor metrology. Xallent is seeking an interdisciplinary MEng team to work on a nanomachine project.
08/13/18Land, B. + Skovira, J.Home security using Raspberry PiI would like to replace the security system in my house (DLINK) with a system that does not need to go out to the cloud for information sharing and control. There are several components. A minimal system would consist of PIR motion sensors, and mains-capable load control, and perhaps a camera. All the separate pieces have size and packaging requirements. Control will be via server on the devices themselves. Security notifications will be by email or SMS. All communication will be by WiFi to a commodity wireless router.
07/12/18Land, B. + Skovira, J.ECE Promotional Display TechnologyECE often takes part in events talking to the public. We would like a set of demonstrations which show off ECE concepts and techniques and which are packaged to be robust and reusable.
07/12/18Land, B. + Skovira, J.PIC and Pi: PIC32 and Raspberry Pi interfaceThe PIC32 microcontroller is a powerful, 32-bit cpu with many peripherals and available libraries, but no real operating system. The Raspberry Pi runs Linux, with all of the cool stuff available to a full Linux distribution, but does not do fast-deadline realtime very well. The combination should have the best of both worlds.
08/30/18Land, B. + Tilsen, S.Computational modeling of speech motor controlUsing electromagnetic articulography the Cornell Phonetics Lab collects real-time data which represents the movements of vocal organs (i.e. the lips, tongue, and jaw) during speech. These data are complex because speech movements are often conducted simultaneously or with a substantial degree of overlap. Current theories of speech motor control decompose speech into basic motor plans called "articulatory gestures", which are modeled with critically damped, transiently forced harmonic oscillators. The aim of this project is to develop a generative, computational model of speech motor control which produces realistic movement trajectories from a low-dimensional gestural model.
07/12/18Land, B., Bass A., Tararsky, R.Tracking a small fish and understanding how it makes soundDanionella dracula are a species of miniature fish closely related to zebrafish, a genetic model organism in behavioral neuroscience. Adults measure less than 20mm in length. Male D. dracula exhibit dimorphic morphology and behaviors from females, where males form territories and defend the territories from other fish using a robust, stereotyped aggressive display, featuring a series of postural movements with the jaws and fins and producing a buzz-like sound. We are looking to collaborate with researchers interested in producing a machine-learning based behavioral assay that automatically tracks the aggressive posturing and sound production of a male fish, recording instances of sound production and the position of body parts during the display. This assay would then be used to quantitatively compare the aggressive displays of male fish under different hormonal treatments/genetic modifications to study the effect of certain neuromodulators on the complete aggressive display.
07/23/18Land, B., Skovira, J., Yoon, Y.Sociometric Badge replacement App for cellphoneA researcher in Design and Environmental Analysis at Cornell is currently using custom devices to measure individual interactions between people working in groups. The devices are relatively expensive and hard to repair, so the researcher wants to replace them with a custom app running on a cheap commodity cellphone. Ideally, the App should replace the social sensing devices (i.e., Sociometric Badges) with small smartphones to detect different types of social interaction. The current social sensing device has built-in Bluetooth, microphone, and infrared sensors used to capture speaking speed, tone, volume, physical movement, and face-to-face interaction time (frequency/duration). While the social sensing device detects speech data, it does NOT record language or conversation, but only the energy and frequency of verbal interaction for privacy protection. All data is recorded for later dump to a spreadsheet with timestamps of events. The cellphone sensors should be able to easily handle everything, except detecting face-to-face interaction via IR transceivers. It may be possible to use openCV running on the camera to detect a facing device (not necessarily the actual face), or it might be necessary to figure out how to optically communicate using the cellphone screen and camera.
07/12/18Land, BruceDE1-SoC Board, Cyclone5 FPGA OpenCLWe are specifically interested in teaching how to use the FPGA as an accelerator for the ARM9 processors using openCL, and how to measure power/performance tradeoffs.
07/12/18Land, BruceSolar panel electrical simulatorThe goal is to make a power supply with output impedance characteristics equivalent to a photovoltaic solar array. Power level should be around 100 watts. There should be controls to set the simulated absolute temperature of the array, the solar intensity, and the number of cells in series (open circuit voltage). There may or may not be an embedded controller to run the system.
08/14/18Martínez, JoséA prototyping effort for in-storage computing on solid-state drivesSolid-state drives (SSDs) use semiconductor technology instead of traditional hard drives’ magnetic plates to in order to store data. SSDs have become pervasive over the past few years; chances are, your smartphone has many gigabytes worth of SSD storage. SSDs need intelligence in order not to break down prematurely due to wear. Storage cells Degrade relatively quickly with every new piece of data stored in them. As a result, SSDs Incorporate a Flash Translation Layer (FTL) to intelligently move the data around, in a manner which is transparent to the user, so as to distribute wearout uniformly across storage cells. This intelligent unit provides an opportunity for SSDs to carry out computation. The goal of this project is to explore opportunities to offload computation from the main processor to the FTL within SSDs. These intelligent operations may include data preprocessing and filtering, or other locally achievable computation that may not need to travel to the processor and back. Not only may this free up the processor to do other work in parallel, it may also accelerate data processing by bringing computation closer to it. The challenges to realizing this are many, including not only isolating and migrating work to the FTL, but also doing so in a manner that is programmer-friendly as well as secure.
08/23/18Molnar, AlyoshaDesign and testing of a printed circuit board (PCB) for a wireless backscatter RFID-style communication system.Students will design a PCB for a near-field backscatter RFID-style system to allow communication with a custom integrated circuit that will be mounted on a honeybee. Students will select and order components, and students will assemble and test the PCB. The PCB must meet sizing and performance specifications set by the needs of the application.
08/23/18Molnar, AlyoshaDesign and simulation of a camera and associated signal processing circuitryStudents will design a camera in Cadence and write code to convert real images into Cadence stimulation waveforms. Time permitting, students will also design memory and image processing circuits.
08/17/18Molnar, AlyoshaSensing and Suppression of Non-Linear Affects in Software Defined TransmittersAny RF transmit chain will have a certain amount of non-linearity associated with it, these non-linearities cause distortion to the desired signal as well as a number of spurious emissions. In many designs the spurs are handled through the use of integrated filter, but in a software defined transmitter such a filter would limit the tuning range. However if the non-linear effects can be adequately characterized then additional signals can be injected into the transmitter to null the spurs at the output.
08/17/18Molnar, AlyoshaGaN Mixers For Software Defined TransmittersAs the RF spectrum becomes increasingly fragmented and scarce the need to be frequency agile becomes vastly more important. This need for flexibility has lead to the rise of software defined radios (SDRs) unfortunately, to date, many of the topologies used to get frequency flexibility struggle to handle large output power. In this project novel circuit topologies will be used in conjunction with commercially available GaN devices to achieve both flexibility and output power.
08/17/18Molnar, AlyoshaNeural Electrode Characterization and ModelingNeural electrodes have been used for both recording biological signals as well as providing stimulation signals used in neuromodulators. It has become more desirable to use the same electrodes for both stimulation and recording to develop a truly closed-loop neural prosthesis. However, it has been nearly impossible to stimulate and record from the same electrode simultaneously due to the magnitude difference in signals as well as the nonlinearities inherent in neural electrodes. By characterizing and implementing a circuit model of the nonlinearities, it can become possible to cancel the stimulation artifact and develop electrode front-ends that will enable true ‘neural duplexing.’
08/17/18Molnar, AlyoshaCOTS Neurostimulator for ‘Neural Duplexing’ ApplicationsDesigning a true closed-loop neural prosthesis becomes more desirable as neural interface technology progresses. However, the signal magnitude differences between the stimulation signals and the desired neural signals make it difficult to record biological responses elicited by a given stimulation signal. By altering the neural electrode interface architecture to resemble the architecture more commonly seen in full duplex radios the stimulation artifact can be suppressed while the neural signal is amplified, thus allowing the capability for true closed-loop neuromodulation.
09/13/18Petersen, K. + Zhang, C.Fine-grained Activity Recognition through Reconstructing 3D Body Posture using wearable sensorsHuman activities are complicated, which usually involves rich body movements. Distinguishing different activities from body movements can be very challenging without knowing the complete body posture of the person. In this project, you will reconstruct 3D upper-body postures using wearable sensing data. Then use the derived body posture info to recognize fine-grained activities (e.g., eating, drinking, exercising).
07/26/18Petersen, KirstinDesign of a Mobile Robot to Support Collaborative Human-Robot SwarmsIn a collaboration between the Verifiable Robotics Lab, the Human-Robot Collaboration and Companionship Lab, and the Collective Embodied Intelligence Lab at Cornell, we are interested in how robots can support swarms of people under duress, e.g. during a disaster situation. This challenge necessitates both novel reactive controllers and electro-mechanical designs that enable safe, intuitive natural interactions with untrained people, while remaining low cost, durable, and low maintenance to permit mass deployment and hour-long operations. To this end, we are inspired by work in urban search and rescue robotics, soft robotics, and swarm robotics.
07/26/18Petersen, KirstinAutomated Image Analysis of Honeycomb (continued)Honeycomb panels are widely utilized for their high strength to weight ratio, in everything from impact-safe packaging to naval, automotive, aircraft, and aerospace structures. For purposes of easy design and fabrication traditional honeycomb consist of regular hexagonal cells. These panels are often used to support non-uniform pressure profiles and are connected via heavy or intricate mechanical joints. In nature, however, honey bees intersperse hexagonal cells of different sizes to adapt to changing colony needs and practical environmental constraints in a continuous manner. The behavior of bees in these boundary regions remain largely unstudied. It seems likely that, through evolution, bees have developed near optimal design strategies to produce comb that is structurally sound, use minimum material, facilitate the needed storage capacity for food and brood, and is robust to minor variations in fabrication. In an effort to optimize the design of traditional industrial honeycomb panels, our lab is developing automated methods to study the composition of natural comb.
08/13/18Sabuncu, M. + Kuceyeski, A.Diagnosis of psychological disorders using multi-modal neuroimaging dataWe will use machine-learning techniques (neural nets, SVM, random forests, elastic net) to classify ~1500 subjects into several appropriate psychological diagnoses based on multi-modal neuroimaging data. We will use a publicly available database of subjects ranging in age from 5-21 years old that contains various MR images, EEG data, demographics and neuropsychological/behavioral assessments. The subjects are either normal or have one of several DSM-5 psychological diagnoses, including Attention Deficit/Hyperactivity disorder, anxiety, learning disorders, autism, conduct disorders, communication disorders and depression disorders. The multi-modal neuroimaging biomarkers we will use in this project include temporal patterns of neuronal activation extracted from resting-state and task-based functional MRI (fMRI) and spectral data from resting or task-based EEG data. The overall goal of this project is to develop an algorithm that clinicians could use to automatically diagnose subjects based on their multi-modal neuroimaging profile. Two secondary goals are to 1) identify specific patterns of activation (either within fMRI and/or EEG) that are indicative of a particular diagnosis, which could provide insight into the biological basis of that disorder and 2) identify which modality is more predictive of these disorders in general, which would assist clinicians in reducing the number of tests implemented during the diagnostic process.
07/12/18Sabuncu, MertNeural network based motion correction for multi-photon microscopy imagesCalcium channel imaging produces high-resolution videos of neurons firing in an animal model, such as a zebrafish. One challenge in analyzing these data is that the animal moves during the experiment and the 3D frames need to be spatially aligned with each other (rigid body image registration). This problem is conventionally solved by solving an optimization problem, which can be rather slow. In this project, we will build on recent deep learning strategies to develop a fast registration tool.
08/21/18Shoaran, MahsaStudying the EEG signatures of tactile stimulation for brain-computer interfaces based on sensory stimulationA brain-computer interface (BCI) provides a non-muscular channel for interaction with the external environment. By measuring the EEG signals from sensory cortex and extracting relevant biomarkers, a subject’s somatosensory attention can be decoded by state-of-the-art machine learning algorithms. Features related to the sensory cortex activity play an important role in the tactile BCI system. The goal of this project is to develop a BCI system based on sensory stimulation. To better facilitate the decoding of subjective somatosensory attention, the effect of sensory stimulation will be thoroughly investigated, with respect to different body parts (hand, foot), stimulation depths (tactile, proprioceptive), waveform parameters (amplitude, frequency, frequency or amplitude modulation), and its EEG dynamics across spatial-spectral-temporal space.
08/16/18Shoaran, MahsaDesign of a cost-efficient machine learning hardware for real-time neural data classificationThe application of machine learning techniques has been continuously growing, and one of the most important and fastest growing areas is medicine and healthcare. For example, ASIC implementation of machine learning models would be beneficial for real-time applications such as closed-loop neuromodulation and motor intention decoding. To meet the tight power budget in portable or implantable medical devices, it is necessary to embed ML into integrated circuits rather than power-hungry FPGA-based microprocessors. The goal of this project is to explore and develop hardware-friendly integrated classifiers for on-chip diagnostic applications such as neural data classification and movement decoding. The students will start with an architecture recently developed in the lab and will explore various hardware optimization techniques to build a scalable and low-complexity machine learning hardware for a wide range of neural data classification problems.
08/16/18Shoaran, MahsaMotor decoding from electrophysiological recordings for reliable brain-machine interfacesA brain-machine interface (BMI) decodes neural activity into useful control signals for guiding robotic limbs, computer cursors, or other assistive devices. In its most basic form, such a system might involve learning a basic mapping of how neural signals relate to cursor velocity and then “closing the loop” to enable direct neural control of cursor velocity. Such systems have shown promise; however, improving performance and robustness remains a challenge. The goal of this project is to improve aspects of real-time neural decoding algorithms based on state-of-the art machine learning principles. Example topics would include the following: optimization of calibration/training procedures used for algorithm parameterization; robust and efficient algorithms for non-linear neural signals; or, algorithms that adapt to recording and/or neural non-stationarities.
07/12/18Shvets, GennadyDesign of ultra-thin metalenses using all-dielectric gradient metasurfacesMetamaterials and metasurfaces revolutionized the field of optics, resulting in novel capabilities such as compact ultra-efficient diffraction gratings, sensors, and lenses. All-dielectric metasurfaces are 2D arrays of precisely engineered complex nano-antennas that exhibit unusual functionalities such as asymmetric scattering, resonances, and chirality. More recently, our group started focusing on using all-dielectric and all-semiconductor metasurfaces for making ultra-thin lenses with high numerical aperture. The problem is rather complex because a very large individually designed dielectric antenna resonators must be properly placed to create an effective zone plate. Some of the designs cannot be obtained on a desktop computer because of the extreme computational requirements for memory and processing speed. Therefore, such designs must be made on a supercomputer cluster.
07/12/18Shvets, GennadyIntegration of a microfluidic system for cells delivery with a photonic biosensor based on a plasmonic metasurfaceThe goal of this project is to advance the concept developed in our group for conducting optical biosensing of live cells in a flow. This enables us to differentiate between normal and cancerous cells, as well as to detect the effects of drugs and other perturbations (e.g., thermal shocks, reduced nutritional content, etc.) using all-optical spectroscopic techniques. The unique aspect of this platform is that it enables spectroscopic characterization of live cells in a flow by integrating microfluidics, plasmonic biosensors, and AC electrodes that attract the cells to the sensor using dielectrophoresis. The scope of the project will be tailored for either one student, or a 2-person team.
08/13/18Skovira, J. + Chang, C.Autonomous Control of UAV to Measure Solar-Induced Fluorescence and Reflectance of CropsWorking with researchers in the Cornell School of Integrative Plant Sciences, Soil and Crop Sciences Section (SIPS-SCS), develop autonomous flight control for drone based measurements. Drone flights will be used to access the development of specific crops at a variety of Cornell locations. Autonomous control for the drone originates with a Raspberry Pi that will direct the flight path by interfacing with the drone controller. At a set of prescribed ‘stops’, the system will interface with spectrometers to perform crop measurements. Also required will be an on-screen operator interface and on-board logging, both designed to inter-operate with existing Agriculture system requirements.
08/13/18Skovira, J. + Chang, C.Tower-based Spectrometer Control SystemWorking with researchers in Cornell College of Agriculture and Life Sciences (CALS), School of Integrative Plant Science, Soil and Crop Sciences Section (SIPS-SCS), develop a system to control spectrometer operation for the study of crop growth. During summer, 2018, an initial prototype system was developed. The prototype is an autonomous motor control system used to align a fiber optic between sky- and vegetation-facing measurement positions, under control of a central data recording system. The system is based on a DC motor controlled with an embedded Arduino with motor shield and logging shield. This MEng project is designed to implement the follow-on requirements of the project for better system interoperability. The current prototype was deployed outdoors on July 3, 2018 and has been operating during daylight hours, collecting ~1500 sample readings a day.
08/21/18Skovira, J. + Fletcher, D.Enhanced Animal Simulator - Evaluating chest compressionsResearchers at the Cornell Veterinary School, under the direction of Dr. Daniel Fletcher, have developed a physical simulator to model canine patients. The open-source simulator architecture includes manager software written in C++ running on a Linux PC, a user interface and simulator patient monitor written using Javascript and HTML 5, and a hardware interface based on a Beaglebone Black microcontroller and several custom PCBs. This animal simulator is used in clinical classes for training veterinary students as well as veterinary technicians and veterinarians. Although the current system has a wide range of functions, including palpable pulses, chest movements, and heart and lung sounds, additional functions would enhance the student experience. The simulators are used extensively for CPR training, an essential skill for veterinarians and veterinary technicians. The most important aspect of CPR is delivery of chest compressions to provide blood flow to the tissues, and the rate and depth must be correct in order for the compressions to generate adequate blood flow. The current system uses an accelerometer to detect chest compressions, but this project would involve developing algorithms and incorporating additional hardware to allow the system to estimate chest compression rate and depth in real time and to provide that feedback to the trainee.
08/21/18Skovira, J. + Fletcher, D.Enhanced Animal Simulator - Modeling an animal’s eyesResearchers at the Cornell Veterinary School, under the direction of Dr. Daniel Fletcher, have developed a physical simulator to model canine patients. The open-source simulator architecture includes manager software written in C++ running on a Linux PC, a user interface and simulator patient monitor written using Javascript and HTML 5, and a hardware interface based on a Beaglebone Black microcontroller and several custom PCBs. This animal simulator is used in clinical classes for training veterinary students as well as veterinary technicians and veterinarians. Although the current system has a wide range of functions, including palpable pulses, chest movements, and heart and lung sounds, additional functions would enhance the student experience. One addition that could supply critical response feedback would be a simulation of an animal’s eyes. Simulating a range of responses, including eye blinks, eye movements, and pupil dilation and constriction in response to light and other external stimuli, would add additional realism to the simulation and give critical physiologic feedback that will allow learners to diagnose a number of neurologic and other systemic diseases.
08/21/18Skovira, J. + Fletcher, D.Enhanced Animal Simulator - User interface designResearchers at the Cornell Veterinary School, under the direction of Dr. Daniel Fletcher, have developed a physical simulator to model canine patients. The open-source simulator architecture includes manager software written in C++ running on a Linux PC, a user interface and simulator patient monitor written using Javascript and HTML 5, and a hardware interface based on a Beaglebone Black microcontroller and several custom PCBs. This animal simulator is used in clinical classes for training veterinary students as well as veterinary technicians and veterinarians. Although the current system has a wide range of functions, including palpable pulses, chest movements, and heart and lung sounds, additional functions would enhance the student experience. The simulators are used extensively for CPR training, an essential skill for veterinarians and veterinary technicians. The current system is controlled using a rudimentary user interface implemented using Javascript and a web page hosted by the simulation manager computer. A more complete user interface with an intuitive, modern design would lower the barrier to entry for new users. In addition, the simulator can be pre-programmed using an XML file that must be written manually. A visual scenario designer using drag and drop components would make the system more accessible.
08/21/18Skovira, J. + Fletcher, D.Enhanced Animal Simulator - Modeling an animal’s pulsesResearchers at the Cornell Veterinary School, under the direction of Dr. Daniel Fletcher, have developed a physical simulator to model canine patients. The open-source simulator architecture includes manager software written in C++ running on a Linux PC, a user interface and simulator patient monitor written using Javascript and HTML 5, and a hardware interface based on a Beaglebone Black microcontroller and several custom PCBs. This animal simulator is used in clinical classes for training veterinary students as well as veterinary technicians and veterinarians. Although the current system has a wide range of functions, including palpable pulses, chest movements, and heart and lung sounds, additional functions would enhance the student experience. One addition that could supply critical response feedback would be a more realistic simulation of an animal’s pulses. The current approach to pulse simulation is functional, but not realistic, and offers a limited ability to model pulse abnormalities commonly noted in clinical patients and essential for diagnosis of life-threatening diseases.
08/21/18Skovira, J. + Fletcher, D.Enhanced Animal Simulator - Video debriefing softwareResearchers at the Cornell Veterinary School, under the direction of Dr. Daniel Fletcher, have developed a physical simulator to model canine patients. The open-source simulator architecture includes manager software written in C++ running on a Linux PC, a user interface and simulator patient monitor written using Javascript and HTML 5, and a hardware interface based on a Beaglebone Black microcontroller and several custom PCBs. This animal simulator is used in clinical classes for training veterinary students as well as veterinary technicians and veterinarians. Although the current system has a wide range of functions, including palpable pulses, chest movements, and heart and lung sounds, additional functions would enhance the student experience. The system currently uses a rudimentary video recording system based on Open Broadcater software to collect video from webcams and a simulated patient monitor during simulations. This video is then synchronized with a log that allows the facilitator to play specific parts of video during the post-scenario debriefing session to identify specific actions taken by the team that warrant discussion. A more robust video collection system would improve the teaching quality of the system.
08/21/18Skovira, J. + George, J.Surface IdentificationAvailability and collection of data from sensors have become simpler today, however the challenge to understand the data available is a challenge. Kionix, Inc has a large array of accelerometers for detecting vibrations. This project proposes to explore the use of accelerometers in determining the surface traversed upon. Examples of surfaces to identify are: • Car - Asphalt, gravel etc. • Vacuum Cleaners - Tiled, carpeted or cemented floor • Construction Machinery - Mud, sand, asphalt etc. A recommended approach, given the possible subtle variations – type of vehicle and differences in similar surfaces, would be utilizing machine learning on the vibration data to determine the nature of the surface. The student team will work with researchers from Kionix, Inc during this project.
08/28/18Skovira, J. + Protasenko, V.Construction of deep-UV light disinfection unit based on commercial mobile I-Robot 2 Roomba platformThe U.S. hospital system welcomes new approaches in reduction of rate of “hospital acquired infection” (HAI). On average, ~700,000 Americans every year stay additional 17 days in hospital rooms due to HAI. Unfortunately, bacteria and viruses causing HAI are developing a resistance to chemicals used for room cleaning. New approaches in handling HAI are required. Exposing hospital rooms to deep-UV (DUV) light is one of possible solutions. DUV light fuses together the strands of DNA and therefore suppresses microbe’s growth. Currently, toll and rigid structures of mercury lamps are used for DUV treatment of hospital rooms. Unfortunately, such treatment leaves intact “shadow areas” where DUV light of bulky stations cannot reach. Mobile and autonomous platforms combined with deep-UV light source should be able to reach such areas (under beds, tables, sinks, chairs, …) for DUV light treatment. Project goal is to construct a workable mobile DUV unit based on I-Robot 2 Roomba platform. In order to improve DUV treatment, panels of DUV light emitting diodes will be accommodated into unit. The diffusive light of DUV panels should improve deactivation of bio-organism “hidden” in roughness and micro-scratches of materials found in hospital rooms (stainless steel, wood panels, plastic, and fabrics). So, we foresee few benefits from combination of I-Robot 2 Roomba platform with panel of DUV lights (with possible addition of near-UV light as well).
08/28/18Skovira, J. + Schneider, D.Cornell Cup Robotics + Da Vinci Labs Robotic SystemThe Cornell Cup Robotics team has been repeatedly recognized by the White House in 2015 & 2016, top industry leaders such as the CEO and CTO of Intel, numerous top colleges, and the international Making Community. Last year the team created a new low cost educational robotic platform to compete with Lego Mindstorms, MakeBlock, &Vex – and now their robot is being licensed for production by Da Vinci Labs! This year the team has opportunities to be even more creative, and invent new features & experiences with a potential reach of hundreds of thousands if not millions of users. With recognition of this work’sp potential impact from former Obama White House officials, this is an excellent opportunity for both people new & experienced in robotics to make a real-world difference. The team will be travelling & showcasing their work at NASA Kennedy Space Center! Please attend 1 of our info sessions (Rh 531, 5:00PM, 8/27 & 8/29). All Majors & Years welcome. MEng Project & Course Credit Available.
08/28/18Skovira, J. + Schneider, D.Create Cornell Cup Robotics’ R2-D2-like Robot and Get Your Droid into a Real Star Wars FilmThe Cornell Cup Robotics team has been repeatedly recognized by the White House in 2015 & 2016, top industry leaders such as the CEO & CTO of Intel, numerous other top colleges, & the international Making Community. This year we will be updating our latest version of an R2-D2 lab assistant to be capable of autonomous navigation of our lab as well as various human-droid interactions. Vision, sensor fusion, AI, robotic arm design, controls, power, communications, are just a few technical development areas. ARM & its ecosystem of partners have also offered enthusiastic support In addition to being showcased at NASA Kennedy Space Center, this year in a Cornell Eng. supported effort, we’ll be running a campaign to get our R2 a cameo in a real Star Wars film! For more info on the project & the exciting attention from Hollywood the project has already earned, please attend 1 of our info sessions (Rhodes 531, 5:00PM, 8/27 & 8/29). All Majors&Years welcome. MEngProject &Course Credit Available
08/13/18Skovira, J. + Ulinski, M.Wireless Power Transfer for Truck Lighting SystemsLarge tractor-trailers have demanding lighting requirements under harsh road conditions. Reliability must be VERY good to avoid fines and provide maximum safety. A critical point of failure is the connection between the wiring internal to the trailer and the light fixture on the outside of the trailer body. These connections tend to fail by corrosion caused by water entering the light fixture where wires penetrate the fixture. What is needed is a wireless power transfer to the lights. Truck-lite has worked with an inductor manufacturer to create mockup components of the coupled system. Further work has been done with this supplier to design the inductive coupling components following manufacturing process limitations. The next step is to develop the power DC-AC inverter which drives the mockup for testing, and the mechanical interface/latching mechanism/manufacturing processes for the designed inductive coupling components.
08/14/18Skovira, J., Winkler, D., Gabrielson, R.Digital Turnstile Tag (DTT) Ground Station DevelopmentThis project will develop of a ground station capable of receiving logged geolocation data from a tag long after it is deployed, dramatically increasing the scope of studies using geolocation by eliminating the need to recapture the tagged birds. Ground stations located at concentration points in migratory routes (“digital turnstiles”) will be able to collect logged data from a large number of tagged birds as they pass by on their regular migration. The conservation of migratory bird populations requires detailed knowledge of individual migration routes and timing. For birds too small to carry tags containing GPS receivers, tags that perform solar geolocation logging are being used instead. Solar geolocation involves recording light level and time data throughout the migration period so that times of sunrise and sunset can be reconstructed. By analyzing data on the rate of change of light level at twilight, the individual’s location can be estimated on a daily basis.
08/14/18Skovira, J., Winkler, D., Gabrielson, R.Digital Turnstile Tag (DTT) DevelopmentThe conservation of migratory bird populations requires detailed knowledge of individual migration routes and timing. For birds too small to carry tags containing GPS receivers, tags that perform solar geolocation logging are being used instead. Solar geolocation involves recording light level and time data throughout the migration period so that times of sunrise and sunset can be reconstructed. By analyzing data on the rate of change of light level at twilight, the individual’s location can be estimated on a daily basis. This project will continue the development of a geolocation tag capable of transmitting its logged data to a ground station long after it is deployed, dramatically increasing the scope of studies using geolocation by eliminating the need to recapture the tagged birds. Ground stations located at concentration points in migratory routes (“digital turnstiles”) will be able to collect logged data from a large number of tagged birds as they pass by on their regular migration.
08/14/18Skovira, J., Winkler, D., Gabrielson, R.Insect Identification TagBiologists and ecologists are interested in monitoring and identifying populations of insects automatically at remote locations. In certain genera of flying insects, species and sex can be identified by the frequency of their wing beats. Development has begun on a small portable instrument with a miniature microphone and a CC1310 wireless microcontroller that samples the acoustical spectrum of its environment over the range of relevant frequencies. The microcontroller must be programmed to compare the detected spectra with “template” spectra of species of interest and identify the most likely match(es). The microcontroller must time stamp and log the results and/or transmit them to a remote data server.
08/31/18Skovira, JoeIntruder Security LightingIntrusion events, including shooting at schools and populated locations, present problems with protecting those in danger. This project proposes an easy-to-use system to temporarily distract an assailant, using non-lethal methods. We propose a high intensity flashing light directed at the intruder in order to distract and temporarily blind an assailant. Two initial prototypes are proposed: a hand-held version that can be deployed by any staff member. It should be designed to track a face in the visual field and focus a bright, flashing light face-on. The second prototype should use methods from the initial work, expanding to a hallway-mounted version with multiple security light elements designed with similar face-tracking and lighting elements. Additional security elements are possible (and encouraged).
08/31/18Skovira, JoeAutonomous Drip Irrigation SystemIn conjunction with the Dilmun Hill Barn Project and ST Micro, this project aims to continue work on an autonomous drip irrigation system. Existing drip irrigation systems must be manually monitored. The Dilmun Hill project proposes low-cost technological solutions for smaller farms with restricted budgets. One helpful system would enable farmers to autonomously control a farm-spanning drip irrigation system to both economize and target water usage and to help with system breakdowns. A previous MEng project (AY 2017-2018) began work on a system using ST Micro devices and a long range, low power wireless radio network for intercommunication, in this case, a SigFox network. Additional work is needed to develop a fully deployable system.
08/14/18Skovira, JoeCo-ProcessorsIn the world today, we are surrounded by an overload of signals and data. There’s a major challenge in identification and extraction of desired signals from this complex surrounding. Kionix’s latest accelerometer has built-in user configurable filters that can be used to extract the signal of interest from such complex environments. This in turn can be combined with digital engines in the accelerometer such as the wake-up function, which is suitable for a variety of applications. Unique advantages of these co-processors include higher performance while reducing the power consumption and offloading the computation from the microcontroller onto the sensor itself. This project addresses real world applications of filters and accelerometers: • Environments with multiple signals and a desired signal is to be extracted. • Detection of abnormal/irregular signal variations in an environment.
08/13/18Skovira, JoeIntelligent Truck-trailer lighting systemA manufacturer of wiring harnesses for truck trailers proposes a project to replace a standard, 7-wire lighting harness with ‘multiplexed’ connection. Reducing the number of wires in the wiring harness will reduce cost, installation effort, weight, and improve connection reliability. Note that this change involves adapting the current system to include intelligent controllers at the lighting destination points in order to decode commands for a specific device from the multiplexed bus. This project will build on a previous MEng project that developed a low-cost, Raspberry Pi and Pic-12 based solution.
08/13/18Skovira, JoeSensor Mesh NetworkInexpensive sensor cost suggests that sensor mesh networks will be become more prevalent. This project addresses networks of sensors, including accelerometers, and how such a network might be designed: • How do sensors within the network communicate? • Added sensors should be able to join the network automatically • One, two, and three-dimensional networks are possible. • Demonstrate a sensor network in a real-world application The example sensor network for this project will be a network of accelerometers designed into a ‘blanket’ to determine and communicate the underlying shape covered by the blanket. The student team will work with researchers from Kionix, Inc during this project.
08/13/18Skovira, JoeSensors for Wildlife MonitoringKionix, Inc has a large array of sensor modules for detecting changes in external environments. This project proposed to explore the use of sensor devices for use in wildlife monitoring. Exploring a variety of possible sensors: • How could sensors be applied to detection of behavior in wildlife monitoring? • How would sensors be integrated into wildlife tags? • How would sensors be powered in the wild? • How would sensors communicate collected data? One example would be to use an accelerometer to record animal activity in a natural environment. The student team will work with researchers from Kionix, Inc during this project.
08/13/18Studer, ChristophMassive MIMO Acoustic CommunicationMassive MIMO will be a core technology of fifth-generation (5G) wireless systems and enables unprecedented data rates and spectral efficiency compared to existing communication technologies. The goal of this project is to build an acoustic communication system consisting of 32 (or more) loudspeakers acting as transmitters and receivers in order to perform massive MIMO communication over the air.
08/13/18Studer, ChristophAcoustic Localization and Tracking via Machine LearningLocalization and tracking of people moving in space finds widespread use in practice. The goal of this project is to develop a prototype system for localizing and tracking people using a passive array of microphones. In order to enable a fully passive operation (i.e., no signals are transmitted, as it is the case for echolocation), the project will build upon recent results from Prof. Studer’s research group in dimensionality reduction and deep learning [A]. [A] C. Studer et al., “Channel Charting: Locating Users within the Radio Environment using Channel State Information,” arXiv preprint:1807.05247, 2018; https://arxiv.org/abs/1807.05247
08/13/18Studer, ChristophDesign of an Acoustic Wireless Communication SystemAcoustic communication is widely used for underwater communication, but can also be used to demonstrate the working principles of modern wireless communication systems. The goal of this project is to develop a hardware prototype of an acoustic communication system that operates over-the-air using a large number transducers (i.e., loudspeakers that simultaneously act as microphones). The prototype system can be used to demonstrate massive MIMO data transmission, full-duplex communication, as well as recent user localization techniques that rely on deep neural networks and dimensionality reduction.
08/27/18Suh, EdSecure Hardware Acceleration for Efficient BlackchainThe goal of this project is to develop a secure hardware accelerator that performs core parts of blockchain (crypto currency) protocols so that the security can be ensured in hardware while improving efficiency.
08/27/18Suh, EdTagged RISC-V Architecture on an FPGAThe goal of this project is to develop a fully-functional RISC-V computing system on an FPGA. The system will be based on the Berkeley Rocket Chip processor, and extended with tagging features to improve security.
08/17/18Tang, KevinJoint Routing and WAN Optimization for Elastic TrafficMuch of the world's data and computation is moving to the cloud. With the proliferation of devices such as smart phones and big data applications, the traffic over our network has kept on growing fast and becoming more dynamic. In this picture, networking becomes even more crucial as it is networks that connect servers to form datacenters and link datacenters and users together. To meet the new requirements in this new era, networks must be orders of magnitude better in terms of flexibility and performance. And that requires much finer-grained network management, so that it can react fast, move load accurately and serve different applications accordingly. The overall goal of my research is to realize such a network management. In this project, particularly, we want to focus a critical technology challenge that is key to realize the benefit of such network management, i.e., enabling joint transport layer and network layer optimization. This cross-layer optimization has been studied abstractly as a mathematical problems from various angles for a long time. However, it has been considered a bad idea from a systems point of view since it breaks the Internet layering structure and is hard to realize among ISPs with standard routers. Now with enterprises such as IBM, Google, Microsoft, Amazon etc. operating their inter-datacenter WANs with increasingly usage of SDN based switches, it becomes clear that such a cross layer optimization is not just desirable, but more importantly, potentially feasible. This project will focus on joint routing and WAN optimization for TCP flows to maximize their throughput.
08/13/18Tiwari, SandipQuantum neural computationNeural networks have been a successful means to simulate and analyze many body problems. Quantum computation is another path to answering similar questions. This project aims to evaluate the efficacy of both of these approaches towards solving manageable Hamiltonian problems---Ising models, for example---and develop an understanding of the origin of success as well as pit falls by comparing them and by developing suitable algorithms for improving them.
08/13/18Tiwari, SandipEntropic changes in neural networks’ information mechanicsThe variety of neural networks and their approaches perform pattern matching quite efficiently. A useful measure of evaluating this efficiency and confidence in the claims of matching is Fisher entropy. And networks attempt to obtain the accuracy through the principles of maximum likelihood, entropy, and others, each with its own domains of successes and pitfalls. This project will bring a statistical view through information mechanics centric evaluation using the Fisher and other measures and various constraints placed on a variety of neural network approaches operating on the same problems. The end goal being a reasonable understanding of how information content extraction maps to the different principles.
08/13/18Tiwari, SandipRecurrent and other neural networks in understanding and creating musicPattern matching in music has been traditionally through recurrent neural networks. It has been employed for melody and harmony. But, it is also possible to make such models in networks such as those employing deep learning. This project will use TensorFlow to explore music, to be able to make recursive networks supervised, and to compare, at least, the difference between deep and recursive learning of music.
08/13/18Tiwari, SandipStochastic circuitsSuperparamagnetic junctions are an efficient source of programmable random generation. Stochastic circuits exploiting randomness are an approach to energy-efficient computation, cryptography, and fast inexact non-Turing pattern extraction. This task will employ Verilog based model of random telegraph behavior of superparamagnetic junctions, together with digital and analog circuits for independent random replication and transformations to evaluate and show efficient circuits in encoding, in secure communications, and in pattern matching.
07/31/18Wagner, AaronCompression using Generative ModelsGenerative models are machine learning algorithms that produce realistic data when fed with noise. Recently generative models such as Variational Autoencoders (VAEs) and Generative Adversarial Networks (GANs) have been shown to achieve better compression for images, compared to traditional compression techniques. This project will consist of implementing generative models to compress different data formats such as images, videos, audio and hopefully improve upon state of the art compression performance. Another aspect of this project is to improve quality of samples generated by generative models, which is an active area of research. This part will consist of implementing modifications to current generative models, in particular VAEs and running various experiments to test theoretical hypotheses.
07/30/18Wicker, SteveBlockchain-based tamper-proof logging and tracking in ad-hoc networks for emergency response and digital agriculture applicationsThe goal of this project is to develop EMR and digital agriculture applications for tamper-proof log using blockchains. The blockchain must tolerate intermittent loss of connectivity to other members of the blockchain network as well as to the rest of the Internet. The ad-hoc nature of the network means that IP addresses will not be stable. Blocks will have to be propagated around the network using protocols that account for unstable IP addresses, such as gossip-style protocols.
07/26/18You, FenqiMachine learning and spatial analysis for agricultural big data analyticsApplying machine learning algorithms to remote sensing data for estimation of crop yields across geographic regions; Delivering spatial analysis for crop yield estimation at scales.
07/26/18You, FenqiQuantum computing for computational optimization and applicationsDeveloping quantum computing frameworks for solving large-scale optimization problems for research, engineering and energy applications, by leveraging IBM’s (industrial partner’s) quantum processors and facilities.
08/13/18Zhang, ZhiruCo-Design for Digital IntelligenceThe following projects are being offered in Dr. Zhang’s research group, where the main theme is building intelligent and highly efficient digital computing systems through co-design of algorithm, programming language, and hardware. This project consists of Three Topics, which will require Two (2) Students per Topic.