2018 Poster/Demo Session

Undergraduate Projects

Application-Agnostic Software Platform for Autonomous Multi-UAV Missions
Andrew Brooks, Emilio Del Vecchio, Joshua Phipps, Kevin Lin, Pharson Chalermkraivuth, RJ Cunningham; Faculty sponsor: Riccardo Petrolo, Edward Knightly; Faculty mentor: Gary Woods
We seek to build a multi-drone hardware system consisting of the drone itself, a computing unit, and associated sensors, as well as a generalized software platform on top of the hardware to allow clients to program tasks for a team of drones to autonomously complete.

Automated Volleyball Analytics
James Grinage, Connor Heggie, Victor Gonzalez, Sachin Jain, Betty Huang, Rebecca Lee
For club volleyball teams to be nationally ranked, they must place as one of the top two teams in a qualifier tournament and compete in the USA Volleyball Junior National Championships. Based on customer interviews, we have identified a pain point among our beachhead market of club coaches: they need faster access to game statistics to develop winning strategies. Today, at best, coaches can access statistics 12 hours after the game; this is not fast enough to make decisions before the next game held the same day. Often times, the stats are not available even the next day. Without a solution to quickly access reliable and accurate statistics, coaches either resort to intuition or manual statistics to make next-game decisions. As one can expect, decisions made through intuition are inaccurate and tend to be biased. Manual statistics are extremely time consuming and are also inaccurate. This is a large pain point because tournament play determines a club team’s success and if coaches aren’t making data-driven decisions at tournaments, they increase the possibility of losing games. As team Cherrypick, our goal is to remove the need for inaccurate and slow manual statistics and instead provide an automated solution powered by computer vision and machine learning. We have split our project up into a few main components: play segmentation, ball tracking, hit categorization, backend framework, and frontend/UI design. All together, these components will enable volleyball coaches and players to analyze game footage in three steps: record the game, upload the footage, and receive a full statistical breakdown in under an hour.

Big Red: Team DISSECT’s Gum Sized Microcontroller
Ronaldo Sanchez, Chris Chivetta, Ray Simar
Big Red is a very small PCB that is ideal for prototyping. Built off of the TI Launchpad platform, Big Red is intended to be used as a replacement for the TI Launchpads used in ELEC 220 and in senior design projects.

DETECT
Stefan Olesnyckyj, Alex Dzeda, Matthew Disiena, Austin Au-Yeung, Jorge Zepeda, Dr. Gary Woods
Municipal water supplies often employ corrosion inhibitors to counteract the effects of old pipes. Corrosion inhibitors form a passivization layer to ensure that heavy metals don’t contaminate flowing and resting water. Monitoring of corrosion inhibitors is currently tedious and expensive, and we are creating a field portable, reusable device for measurement of corrosion inhibitors. We have currently designed an analog printed circuit board (PCB) that is capable of initiating and taking an accurate measurement on a solution. This measurement is parameterizable to account for a variety of testing situations. Further, we have been able to make an enclosure for our device that reduces high frequency (>1kHz) noise by at least 10dB. From a data perspective, we have been able to qualitatively and quantitatively characterize how important variables affect our measurements, digitally remove noise, and perform simplistic feature extraction to fit a numerical model. In addition, we have obtained qualitative results pertaining to how contaminants in our solution affect our results. In addition, we have written appropriate sample generation, measurement, and safety procedures. Now that we have demonstrated that all of our parts work individually, our goal for the future is to begin testing with the integrated device. The device will be able to fit comfortably in a backpack, can be run over 100 times without degrading, and have experiments be run through a GUI. In addition, we will continue honing our prediction mechanisms and characterizing our samples more thoroughly.

Digital Cure for Epilepsy
Paul Mayer, Qianyun Wang, Joseph D’Amico, Christopher Chee, Anika Zaman, Jeffrey Horowitz, Nitin Tandon, and Behnaam Aazhang
We have developed a support vector machine (SVM) learning algorithm meant to predict and suppress seizures in patients with epilepsy. Trained on patient intracranial electroencephalography (iEEG) data, our algorithm has a high seizure prediction sensitivity while maintaining an acceptably low number of false positives. We have implemented the algorithm onto a small form factor system on module (SoM) chip. The device runs on real-time patient data inputs with optimized computation specifications and power consumption. The final stage of this system will be implemented on an application specific integrated circuit.

Hardware Acceleration of a Software-Defined, LTE Base Station
Chance Tarver, Luke Zhang, Louise Liu, and Joseph R. Cavallaro
Commercial LTE equipment is expensive which often makes performing wireless research difficult. Although we may be able to test new ideas using MATLAB, it is hard to simulate these new ideas end-to-end to see how they behave in a commercial deployment environment. Using a software-defined, open-source, LTE base station project called Open Air Interface (OAI), we can implement the full LTE protocol on a standard computer and can communicate with commercial off-the-shelf (COTS) phones such as the Google Pixel. With everything freely available, this provides an exciting opportunity for academia and industry to innovate at a faster pace. Although a software implementation offers portability, it comes at a price of performance. In this work, we implement the OAI on a CPU with a USRP radio and explore the acceleration of specific tasks such as IFFTs and error correcting codes on GPUs using CUDA.

Inspection Robot for Limited Mobility Environments
Ben Wasserman, Cody Peterson, Irvin SKK, Matthew Hays, William Jones
Our goal is to design and construct an affordable, reliable, flexible use robot that accurately and efficiently performs a variety of visual and sensory inspections in hard to reach, dangerous or otherwise inaccessible locations.

Monitoring Jugular Venous Pressure for At-Home Congestive Heart Failure Care
Nick Calafat, Yida Liu, Akhil Surapaneni, Jack Terrell, Angela Zhang, Dr. Eric Richardson, Dr. Rohan Wagle, Dr. Gary Woods
Congestive heart failure (CHF) occurs when the heart is unable to effectively pump blood throughout the body. This causes fluid buildup in other organs, the most common cause of CHF-related hospitalizations. At home, patients monitor their weight as an approximation of fluid volume, but this method often does not provide early enough warning to avoid emergency hospitalizations. Fluid status can be tracked more directly in hospital by measuring right atrial pressure in the hospital via right heart catheterization or echocardiogram or measuring the jugular venous pressure physical examination. A jugular venous pressure (JVP) monitoring device is proposed for at-home monitoring of right atrial pressure for congestive heart failure patients. A 1D array of 10 MEMS accelerometers are placed over the internal jugular vein (IJV) to detect the height of IJV pulsations. By tracking IJV fluid pressure over time, patients would be able to monitor their fluid retention, adjust their medications accordingly, and manage congestive heart failure from their own homes.

Passive Parity-Time-Symmetry in Semiconductor-Metal Photonic Crystals
Alex Hwang, Chloe Doiron, Gururaj Naik
Physical observables, e.g. position of an atom or frequency of light, are real-valued. Thus, quantum mechanics usually defines physical systems as Hermitian, because Hermiticity implies real observables. However, it was recently shown that non-Hermitian systems can also exhibit real observables if they are parity-time symmetric (PT-symmetric). PT-symmetric systems have been studied intensely because of their unique exceptional point physics and topological properties. So far, all demonstrations of PT-symmetry in optics have been in the infrared, but PT-symmetry would have important and practical applications in the visible. I have engineered a PT-symmetric nanophotonic device in the visible, consisting of two coupled arrays of Si nanocylinders. Because this structure is PT-symmetric, it responds to light in a highly nonlinear fashion. By tuning the amount of loss in our system, we can move between two phases with distinct behavior. In one phase, transmission decreases with added loss, which is normal behavior. But in the PT-broken phase, transmission anomalously increases with added loss. I propose to use such an intriguing nanophotonic PT-system in advanced biochemical sensing.

Prototyping Future Architecture for Wireless Leadless Multisite Pacemakers
Cody Tapscott, June Chen, Chris Chivetta, Yoseph Maguire, Yixin Chen, Ricky Chen
Current pacemakers utilize intravenous leads to sense electrical irregularities within the heart and provide electrical stimulus for correction. The leads are effective, but problematic in a significant number of patients. To spur the creation of pacemakers that forego the use of leads, we are developing a prototype pacing system which will model the main features of a WLM (Wireless, Leadless, and Multisite) pacing unit. Using currently available technologies we have designed a hardware main unit (a “can”) that communicates wirelessly with an array of sensing/stimulation electrodes. This system is completely leadless and therefore addresses one of the main weaknesses of modern pacemakers.

Graduate Projects

A Deep Learning Approach to Automatic Question Generation With Application in Personalized Learning
Jack Wang, Andrew Lan, Weili Nie, Philip Grimaldi, Richard Baraniuk
We introduce QG-Net, a recurrent neural network-based model specifically designed for automatically generating quiz questions from educational content such as textbooks. QG-Net outperforms existing neural network-based and rules-based systems for question generation when evaluated on standard benchmark datasets. It also scales favorably to applications with a large amount of educational content, since its performance improves with more training data. More importantly, we demonstrate that QG-Net, trained on a publicly available, general-purpose dataset and \textit{without} further fine-tuning, is capable of generating high quality questions from textbooks, where the content is significantly different from that of the training data. These results show that QG-Net has the potential to scale up question generation to millions of educational content by significantly reducing the cost of doing so.

A Framework to Computationally Characterize the Depth Limit of Bioluminescent Sources

Ankit Raghuram, Jesse Adams, Fan Ye, Dr. Jacob Robinson, Dr. Ashok Veeraraghavan
Imaging neuronal firing can give scientists insight into how the brain stores information and executes behaviors. Bioluminescence has recently been considered as a modality to image through scattering brain tissue because of its high signal-to-noise ratio as compared to other imaging techniques like fluorescence. I propose a framework to compute the resolution and depth limit of bioluminescence in scattering media by modeling photon trajectories using a Monte Carlo simulation as a means of estimating whether it is possible to resolve bioluminescent sources at common imaging depths for scattering media. Distributions of “signal” and “background” will be computed from Monte Carlo simulations and separation of the distributions using a pixel intensity detection framework will determine what distance is resolvable at certain depths. The proposed framework is performed for both static images and dynamic videos. These results will be corroborated with traditional calcium imaging software. Future work will focus on using this framework to test different parameters of bioluminescence like emission rate and distribution of background sources. Parameters sweeps of these variables will help determine the relationship between these parameters and depth limit (at a specific resolution). This will give scientists designing bioluminescent molecules a goal to work towards to image neuron groups at specific depths.

Absorption localization in plasmonic heterostructures
Seyyed Ali Hosseini Jebeli,Stephan Link,Wei-Shun Chang, Ujjal Batacharjee
Plasmonic nano particles have found many applications in different areas like nano electronics, energy harvesting and conversion and photothermal therapy of malignant tissues due to their strong interaction with light. These structures usually have wide bandwidth and uniform absorption and scattering but using combination of different plasmonic structures the bandwidth and localization of absorption or the scattering pattern could be manipulated. In this work the absorption localization in the target structure is presented both using different materials and also the same material in structures with different sizes. Pt decorated gold nano rods are used to show the heat localization in Pt particles due to their different absorption than gold, also gold nano rods with different sizes are used as coupled dimers to show the effect of coupling on absorption of one of the structures. Significant localization is observed in both cases which shows the great potential of plasmonic heterostructures for manipulating interaction of light and matter.

Aligned Carbon Nanotube Films as a Platform for Hyperbolic Thermal Emitters
Weilu Gao, Chloe F. Doiron, Xinwei Li, Junichiro Kono, and Gururaj V. Naik
Selective thermal emitters emit thermal radiation in a narrow frequency range. High performance selective emitters are exciting optical devices for applications such as chemical sensing and energy conversion. Hyperbolic optical materials are an interesting material platform for selective thermal emitters due to the hyperbolic dispersion, which supports high-momentum (high-k) photons. Additionally, hyperbolic materials have an unbounded photonic density of states (PDOS) enabling super-Planckian emission and strongly enhanced radiative near-field heat transport. To date, hyperbolic thermal emitters have not been demonstrated at high-temperatures due to the lack of a material platform stable at high-temperatures. We demonstrate aligned single-wall carbon nanotube (SWCNT) films that have hyperbolic dispersion in the mid-infrared, behaving as a metal parallel to the nanotube axis and a dielectric in the perpendicular plane. Thermal emission measurements were carried out at 700oC using planar and patterned films. For planar films, we observed polarized thermal emission with a peak near the epsilon-near-zeros point (ENZ). Additionally, by patterning the films we were able to fabricate deep sub-wavelength sized cavities with volumes below λ3/700. This demonstrates that we were able to create hyperbolic cavities with significantly enhanced PDOS.

Automated Multimodal Screening of Fluorescent Biosensors of Membrane Potential
Zhuohe Liu, Yueyang Gou, Sihui Guan, Jihwan Lee, Francois St-Pierre
A quantitative understanding of neuronal computations can be achieved by monitoring membrane potential reported by genetically-encoded voltage indicators (GEVIs). The fluorescent biosensors enable recordings of cellular electrical activity in vivo with subcellular resolution and cell type specificity. However, current indicators are not photostable and bright enough for long-term recording, and their sensitivity and kinetics are not satisfactory for detection of fast voltage dynamics. We report a high-throughput platform to screen mutagenesis libraries of GEVIs by analyzing microscopy videos of HEK293 cells during electric field stimulation. The platform quantitatively ranks candidate GEVIs based on performance scores across multimodal imaging methods. We anticipate that the approach can be extended to the screening of other fluorescent biosensors.

Brain dynamics using Non-Parametric Information Theoretic Tools and Graph Theory, in Human Language
Sudha Yellapantula, Nitin Tandon, Behnaam Aazhang
The human brain is a highly interconnected set of neuronal pathways, and the goal of this project is to understand brain network dynamics from ECoG recordings in an object naming task. Human ECoG data is obtained from cortical electrodes implanted in patients undergoing epileptic surgeries who perform a multitude of language tests. Prior work in understanding dynamics uses power based analysis, or assumes a model of the data. In this work, we use a non-parametric information theoretic tools to infer increases in information flow between all pairs of channels compared to a baseline. Statistical significance was calculated by choosing appropriate null models. We then used graph theoretic tools on graphs that had electrodes as vertices and increase in information flow as edges. By performing community detection using the Louvain algorithm, we were able to identify some of the major stages in which the brain reaches a decision, in the word selection stage, preparing for articulation stage, articulation, and after hearing themselves articulate the word. We can also identify the brain patterns in these activations. We then compare the activations and information flow with the power responses, to test a hypothesis that an increase in high gamma power is related to an increase in information flow to that node. This work paves the way to have a better understanding of the brain activations in a language task.

CameraHRV: Robust measurement of Heart Rate Variability using a Camera
Amruta Pai, Ashok Veeraraghavan, Ashutosh Sabarwal
The inter-beat-interval (time period of the cardiac cycle) changes slightly for every heartbeat; this variation is measured as Heart Rate Variability (HRV). HRV is presumed to occur due to interactions between the parasym-pathetic and sympathetic nervous system. Therefore, it is sometimes used as an indicator of the stress level of an individual. HRV also reveals some clinical information about cardiac health. Currently, HRV is accurately measured using contact devices such as a pulse oximeter. However, recent research in the eld of non-contact imaging Photoplethysmography (iPPG) has made vital sign measurements using just the video recording of any exposed skin (such as a person’s face) possible. The current signal processing methods for extracting HRV using peak detection perform well for contact-based systems but have poor performance for the iPPG signals. The main reason for this poor performance is the fact that current methods are sensitive to large noise sources which are often present in iPPG data. Further, current methods are not robust to motion artifacts that are common in iPPG systems. We developed a new algorithm, CameraHRV, for robustly extracting HRV even in low SNR such as is common with iPPG recordings. CameraHRV combined spatial combination and frequency demodulation to obtain HRV from the instantaneous frequency of the iPPG signal. CameraHRV outperforms other current methods of HRV estimation. Ground truth data was obtained from FDA-approved pulse oximeter for validation.

Cheetah: NVRAM + RDMA enabled low-latency distributed storage
Qingyue Liu, Peter Varman
Fast-changing developments in both storage and networking technologies has forced a rethinking of traditional distributed storage system design. Sub-microsecond latencies of NVRAM storage, fast, lightweight communication protocols based on RoCE (RDMA over Converged Ethernet), and multicore processors provide the basis of building a new generation of scalable distributed storage systems. We propose a new distributed storage system design that aims to exploit the fine-granularity access provided by emerging NVRAM technology with direct remote memory access enabled by RDMA protocols to achieve microsecond-level transaction latencies and high transaction throughputs. Our design includes three components: extensible distributed data placement and migration algorithms, lightweight and consistent data access protocols, and scalable failure detection and recovery mechanisms. Initial experiments on RDMA protocol over IB networks and simulation results of the data access protocols show the consistency and low-latency achieved by the proposed design.

Compressive Hyperspectral Imaging and Machine-Vision
Yibo Xu, Jianbo Chen, Liyang Lu and Kevin F. Kelly
Compressive imaging is a novel imaging technology where images can be recovered from far fewer linear measurements than required by the traditional Nyquist theorem, utilizing the inherent sparsity of the image in some transform domain. First, we present a new sum frequency generation imaging microscope using compressive imaging that has been developed for surface studies. Then a hyperspectral video imaging system designed with the single-pixel and single-modulator architecture is demonstrated. Last we employ machine learning methods to do inferences directly in the compressed domain of the scene instead of on raw image pixels. It enables object recognition and related tasks employing a single detector using compressed samples taken by the single-pixel camera.

Contextual Multi-Armed Bandits for Personalized Learning Selection
Indu Manickam, Andrew Lan, Richard Baraniuk
Optimizing the selection of learning resources and practice questions to address each individual student’s needs has the potential to improve students’ learning efficiency. In this paper, we study the problem of selecting a personalized learning action for each student (e.g. watching a lecture video, working on a practice question, etc.), based on their prior performance, in order to maximize their learning outcome. We formulate this problem using the contextual multi-armed bandits framework, where students’ prior concept knowledge states (estimated from their responses to questions in previous assessments) correspond to contexts, the personalized learning actions correspond to arms, and their performance on future assessments correspond to rewards. We propose three new Bayesian polcies to select personalized learning actions for students that each exhibits advantages over prior work, and experimentally validate them using real-world datasets.

Decentralized data detection for massive MU-MIMO base-station on multi-GPU systems
Kaipeng Li, Charles Jeon, Joseph Cavallaro, Christoph Studer
Achieving high spectral efficiency in realistic massive multi-user (MU) multiple-input multiple-output (MIMO) wireless systems requires computationally-complex algorithms, such as data detection in the uplink (users transmit to base-station) transmission. Most existing algorithms are designed to be executed on centralized computing hardware at the base-station (BS), which results in prohibitive complexity for systems with hundreds or thousands of antennas and generates raw baseband data rates that exceed the limits of current interconnect technology and chip I/O interfaces. We propose a novel feed-forward decentralized baseband processing architecture that alleviates these bottlenecks by partitioning the BS antenna array into clusters, each associated with independent radio-frequency chains, analog and digital modulation circuitry, and computing hardware. For this architecture, we develop novel decentralized data detection algorithms that only access local channel-state information and require only one-shot message passing among the clusters for global detection results. We study the associated trade-offs between error-rate performance, computational complexity, and interconnect bandwidth, and we demonstrate the hardware efficiency at Gbps level data throughput and the design scalability to support hundreds of BS antennas with a reference implementation on multi-GPU supercomputers.

Digital Predistortion for Massive MIMO
Chance Tarver and Joseph R. Cavallaro
Power amplifiers (PAs) are a critical but often problematic component of the RF physical layer. Their inherent nonlinearities contribute to spectral regrowth around RF signals which can violate emission masks set by standards and regulatory agencies such as 3GPP and the FCC. This is frequently combated by using a digital predistortion (DPD) algorithm where the nonlinearities are estimated so that they can be corrected before the transmission. In emerging 5G and beyond 5G systems with massive multiple-input and multiple-output (MIMO), there may be hundreds of antennas in the basestation to allow for greater capacity. However, as we increase the number of antennas, we also increase the number of PAs, and the complexity of correcting for their nonlinearities grows. In this preliminary work, we explore the application of DPD to massive MIMO communications so that we can reduce the computational complexity.

Directional Training for FDD Massive MIMO
Xing Zhang, Lin Zhong and Ashutosh Sabharwal
A key challenge for frequency-division duplexing (FDD) Massive MIMO is the large overhead in acquiring channel state information (CSI) for transmit beamforming. In this paper, we propose a scalable method called directional training to obtain downlink CSI. Directional training is motivated by two empirical results derived from massive MIMO channel measurements. First, the number of dominant angle-of-arrivals (departures) is much smaller than and nearly independent of the number of base-station antennas. Second, there is a strong correlation between uplink arrival and downlink departure angles even in FDD systems, which leads to the idea of directional training where a small number of training symbols can be sent to estimate the dominant components of the downlink channel. Therefore, directional training measures much fewer complex coefficients than full-training based methods, and as a result, compared to full-training, the overall channel acquisition overhead for directional training scales much slower with the number of base-station antennas. We evaluate directional training with extensive experiments with a 64-antenna base-station at two bands separated by approximately 72 MHz. Our results show that directional training based downlink beamforming outperforms full-training systems by 150% in terms of average spectral efficiency, and loses only 5.3% average spectral efficiency from genie-aided systems.

Efficient Nanophotonics-Enabled Solar Membrane Distillation for Off-the-grid Water Treatment
Pratiksha D. Dongare, Alessandro Alabastri, Oara Neumann, Peter Nordlander, Naomi J. Halas
According to a recent study by Mekonnen and Hoekstra (Science Advances, 2016), 4 billion people around the world face water scarcity for at least one month of the year. To meet this increasing water demand, it becomes important to purify alternative water sources like saline seawater, wastewater and polluted water. Many of the traditional water purification solutions become expensive and require high maintenance for treating these alternative water sources. As a solution, we have developed a completely off-the-grid and scalable Nanophotonics-enabled Solar Membrane Distillation (NESMD) system. NESMD utilizes the efficient localized heating of cost-effective carbon black (CB) nanoparticles embedded in poly(vinyl alcohol) matrix on top of a polyvinylidene difluoride (PVDF) membrane. Saline water (feed) and purified water (distillate) at ambient temperature are flown on top and bottom of this coated membrane respectively. The broadband absorbing CB nanoparticles absorb more than 90% of the incident sunlight and create localized heating in a 10-15 microns thick layer. The saline feed water in contact with the CB layer heats up creating a localised temperature difference across the membrane. The resulting vapor pressure difference leads to evaporation of saline water at the feed-membrane interface and condensation at the distillate-membrane interface. As the water vapor travels from the saline feed to purified distillate, it leaves the salts and minerals behind, thus purifying water. The condensing water vapor transfers the enthalpy of condensation to the distillate increasing its temperature. This heat can be transferred to input saline feed through a heat exchanger to preheat the input water providing an energy source in addition to sunlight. At solar intensity of 700 W/m2 we obtain purified water flux of 0.3 kg/m2h (30 % efficiency) with NESMD without heat exchnager which increases to 0.6 kg/m2h (60 % efficiency) with an heat exchanger with input feed and distillate flux of 5 mL/min and 50 mL/min respectively. The salt rejection of the system is > 99.5 %. These initial results point towards NESMD being a useful and practical solution to treat alternative water sources to meet increasing water demands of the world.

Estimating Gameplay Engagement from User-contributed Videos
Xu Chen, Li Niu, Ashok Veeraraghavan and Ashutosh Sabharwal
Fueled by the advent of digital media, people are increasingly surrounded by numerous screen-based applications such as mobile shopping, eLearning, online chatting, and gaming. Understanding the extent to which users engage with those screen media is able to facilitate more applications toward optimizing user experience, which in turn can also benefit content providers by increasing their revenues, user retention rates and so forth. However, existing attempts to estimate user engagement in activities either rely on self-report measures sometimes lacking objectivity, or physiological signal-based methods requiring invasive and costly devices, both of which are impractical for real-world deployments (e.g., home use). In this work, we focus on a representative screen-based activity — gaming, and present a deep learning-based framework that can infer user gameplay engagement non-intrusively and automatically. We base our study on a labeled, challenging and realistic YouTube gameplay video dataset containing over a thousand of picture-in-picture gaming video clips recorded by amateur gamers themselves in unconstrained conditions. The results from extensive experiments demonstrate that our deep learning-based engagement estimator can achieve the accuracy of over 85%, and that the estimator can be generalized across different users and diverse games.

Face Detection with the FlatCam Lensless Imaging System
Jasper Tan, Li Niu, Jesse Adams, Vivek Boominathan, Jacob Robinson, Richard Baraniuk, Ashok Veeraraghavan
In this demo, I show a live video feed from the FlatCam lensless camera. In addition, I show that the presence and locations of human faces can be obtained from the reconstructed FlatCam video.

Feasibility of Passive Eavesdropping in Massive MIMO: An Experimental Approach
Chia-Yi Yeh, Edward Knightly
Massive MIMO has the potential to thwart passive eavesdropping as the signals transmitted by a large antenna array become highly focused. Indeed, the impact of passive eavesdropping has been shown to be negligible when the number of base station (BS) antennas approaches infinity for independent Rayleigh channels. In this paper, we experimentally explore eavesdropping in Massive MIMO incorporating real-world factors including a limited BS antenna array size, potential correlation in over-the-air channels, and adaptation of modulating and coding schemes (MCS) over a discrete and finite set. Using a 96-antenna ArgosV2 BS, we (i) explore scaling the array size; (ii) identify eavesdropper advantages due to channel correlation and the resulting increase in array size required to mitigate this advantage; (iii) identify the “MCS saturation regime” as a vulnerability even with high SNR, (iv) characterize transmit power control counter strategies at the BS, and (v) explore the impact of a nomadic eavesdropper that moves to find the most favorable position.

FLASH – Family Level Assessment of Screen-use in the Home
Leo Meister, Ewa Nowara, June Chen, Ashutosh Sabharwal, Ashok Veeraraghavan
Nowadays, it is common for individuals to own several electronic devices ranging from tablets and smartphones to computers and televisions. However, due to a lack of tools to continuously and unobtrusively monitor screen usage, the extent of the health consequences from using these devices daily is unknown. This study proposes a machine learning technique based on image- and video-processing algorithms to simultaneously monitor the screen usage of multiple people in a home-like setting. Such a system will be the first unobtrusive, second-by-second accurate means of monitoring human screen usage. Running in real-time will prevent the need to store the video, preserving users’ privacy. Currently, to achieve that level of accuracy requires human observers – both costly and intrusive – while non-intrusive techniques like self-reporting tend to be inaccurate. Front-facing cameras on mobile devices will collect video of their users and external cameras mounted to TVs will record TV viewers. To learn when users are engaging with the screens in front of them and to monitor each user’s total screen-time, the system will detect and recognize users and track their gaze and head pose. The system will pave the way for future studies seeking to understand the long-term health effects of screen usage at a more granular level in the age of ubiquitous screens.

Focused Ultrasound: A Treatment for Neuropsychiatric Disorders
Boqiang Fan, Behnaam Aazhang
Neuropsychiatric disorders, such as depression and obsessive compulsive disorder, affects millions of individuals in the world. Traditional treatment includes deep brain stimulation (DBS), which stimulates deep brain tissues with high spatial resolution by utilizing electrodes plugged into the brain. Despite the efficacy, plugging electrodes into brains may introduce surgery damage and side effects such as infection and rejection. Therefore, non-invasive DBS methods attracts researchers’ attention recently. Ultrasound has been proved to be able to affect brain activities and is viewed as a potential treatment for neuropsychiatric disorders. In this research, we aim to design the ultrasound transducer and the stimulation signal pattern for modulating brain activities with very high spatial resolution. Simulation results of ultrasound propagation in medium is presented.

Fresnel Lens Imaging by Wavefront Coding
Yicheng Wu, Manoj Kumar Sharma, Ashok Veeraraghavan
We introduce a wavefront coding technique to use a Fresnel lens to achieve sub-diffraction limited imaging performance. A Spatial light modulator (SLM) is placed in front of the sensor to randomly modulate the phase of the incident light field. For the reconstruction, we first apply a phase retrieval algorithm to recover the complex field on the sensor plane. And then, we back-propagate this field to object plane, with pre-calibrated phase distribution on the Fresnel lens, to get both amplitude and phase information of the unknown object. This method improves the spatial resolution of the Fresnel Lens by more than 100 times.

Functional Architectures of Aligned Carbon Nanotubes
Natsumi Komatsu, Weilu Gao, Peiyu Chen, Cheng Guo, Aydin Babakhani, and Junichiro Kono
Carbon nanotubes possess unique electrical, thermal and mechanical properties, but creating macroscopic devices by assembling while preserving their extraordinary properties has been challenging. In this poster, wafer-scale monodomain films of spontaneously aligned single-walled carbon nanotubes and precise control of the levels and locations of dopants would be introduced.

Gas Tracking Wireless Sensor Network
Nadya Mohamed, Joseph Cavallaro
Wireless sensor networks (WSNs) are regarded as a revolutionary information gathering method that will greatly improve the way of human living, working and change world with their advanced architecture and wide variety of daily life applications. The oil and gas industrial environment is one of the most prevailing environments for the application of WSN. The deployment of WSN technologies in the oil and gas fields aid the oil industries in improving safety, optimizing operations, and reducing downtime and productions costs. However, there are still some issues and challenges that need to be studied and one of the main challenges is the node deployment which highly affect the system performance. The objective of this project is to study the sensing coverage and network connectivity to find an optimal node deployment strategy that provide a reliable cost effective quality of service (QoS) for gas sensing applications.

GPLink: Interference-free Reuse of Guard Periods in OFDM-based Networks
Niranjan M Gowda and Ashutosh Sabharwal
In this paper, we propose and study a simple idea of re-using the channel during guard periods of an active OFDM link (MainLink), which we refer to as GPLink (short for Guard Period Link). GPLink is active only during the guard periods of the OFDM symbols. Since the samples in the guard periods are discarded at the MainLink receiver, the interference caused in the guard periods by the GPLink is non-consequential. The main contribution is the design and analysis of GPLink for the zero-knowledge case, in which the GPLink transmitter has no knowledge about the locations or the number of MainLink users. Our numerical study show that with 20 MHz normal-LTE, which has 7% cyclic-prefix overhead, data rates on zero-knowledge GPLink can exceed 20Mbps when the distance between GPlink transmitter and receiver is less than 1km. For an OFDM link with 25% CP overhead (example: extended LTE or 802.11), data rates on zero-knowledge GPLink can exceed 100 Mbps even when GPlink users are 2 km apart.

Hardware Transactional Persistent Memory

Ellis Giles, Kshitij Doshi, Peter Varman

Recent years have witnessed a sharp shift towards real-time data-driven and high-throughput applications, impelled by pervasive multi-core architectures and parallel programming models. This has spurred a broad adoption of in-memory databases and massively-parallel transaction processing across scientific, business, and industrial application domains. However, these applications are severely handicapped by the difficulties in maintaining persistence on typical durable media like HDDs and SSDs without sacrificing either performance or reliability. Two emerging hardware developments hold enormous promise for transformative gains in speed and scalability of concurrent data-intensive applications. The first is the arrival of Persistent Memory, or PM, a generic term for byte-addressable non-volatile memories, such as Intel’s 3D XPoint\TM technology. The second is the availability of CPU-based transaction support known as Hardware Transactional Memory, or HTM, which makes it easier for applications to exploit multi-core concurrency without the need for expensive lock-based software. This poster presents Hardware Transactional Persistent Memory, the first union of HTM with PM without any changes to known processor designs, allowing for high-performance, concurrent, and durable transactions. The techniques presented are supported on three pillars: handling uncontrolled cache evictions from the processor cache hierarchy, logging to resist failure during persistent memory updates, and log ordering to permit consistent recovery from a machine crash. We develop pure software solutions that work with existing processor architectures as well as software-assisted solutions that exploit external memory controller hardware support. We also introduce relaxed versus strict durability, allowing individual applications to tradeoff performance against robustness, while guaranteeing recovery to a consistent system state.

Hybrid optical-computational CNN with DMD based imaging system
Jianbo Chen, Yibo Xu, Matthew Herman, and Kevin F. Kelly
We designed a hybrid convolutional neural network system using digital micro mirrors as the first convolution layer for lower cost and more computation efficient object classification applications.

Imaging Through Scattering Media
Manoj Sharma, Chris Metzler, Rich Baraniuk, Ashok Veeraraghavan
We use transmission matrices to characterize the scattering process associated with complex scattering media. We then use phase retrieval algorithms to “invert” the scattering process and see through frosted glass.

Information Theory in Neuroscience
Joseph Young, Valentin Dragoi, Behnaam Aazhang
Information theory has gained increasing popularity in the analysis of neural data. Mutual information, typically used to quantify the capacity of a communications channel, has been used in neuroscience to analyze the relationship between external stimuli and neural behavior. We focus on a newer formulation of mutual information that allows for analysis of the frequency relationships between recorded neural signals, and also investigate the application of directed information to such signals. Both of these tools are extremely powerful, as they do not make any model assumptions and account for nonlinearity. We have used such tools to investigate a particular set of recordings from the visual cortices of monkeys while they were performing a visual matching task.

Learned D-AMP: Principled Neural Network Based Compressive Image Recovery
Chris Metzler, Ali Mousavi, Richard Baraniuk
Well-understood iterative algorithms can be “unrolled” to form deep neural networks. In this work we unroll a high-performance compressive sensing recovery algorithm to form a deep neural network. This network outperforms all other recovery algorithms in terms of performance, performs high resolution reconstructions in near real-time, and, unlike other neural network-based techniques, can handle a variety of measurement models.

Light-Induced Bandgaps In Graphene
Bryan Anthonio, Weilu Gao, and Junichiro Kono
Graphene has recently emerged as one of the most studied materials in condensed matter physics due to its extraordinary electrical and mechanical properties that may be useful for many technological innovations. However, the fact that it lacks a bandgap renders it unfeasible for certain applications including optoelectronics and other semiconducting technologies such as fast-switching transistors. Recent theoretical calculations have elucidated the possibility of opening a sizable bandgap by irradiating graphene with intense pulses of circularly polarized mid-infrared or far-infrared radiation through a coherent modification of the topological properties of electronic states. These studies predict that the size of the induced bandgaps varies with the wavelength and intensity of light excitations, which can be utilized to develop tunable and switchable optoelectronic devices. In this talk, we will describe our current efforts to experimentally demonstrate laser-induced bandgaps in graphene.

Low-Latency, Closed-Loop System for Hippocampal Sharp-Wave Ripple Detection
Shayok Dutta, Etienne Ackermann, Chun-Ting Wu, Caleb Kemere
Transient neural activity pervades electrophysiological activity within the hippocampus. Particularly, sharp-wave ripples (SWRs), transient coordinated bursts of ~150-250 Hz oscillations present in the local field potential in hippocampal area CA1 last approximately 60-150 ms. Within the network, SWRs co-occur with ensemble spiking of pyramidal neurons in the area. Together, SWRs and concomitant neural activity are associated with information propagation relating to memory consolidation, recall, and memory-guided decision making. Selective interaction with the brain upon online hippocampal SWR detection has been established to cause behavioral and cognitive alterations in animal memory consolidation and working memory. However, null results of behavioral or physiological alterations have also been reported. Additionally, investigations of cortical regions post-SWR have been studied in order to examine the extent of SWR based coordination within the brain. As such, to reverse engineer SWR contributions to learning and memory processes, we evaluate the performance and discuss the capabilities of an open-source, plug-and-play, online ripple detection system. Our system has been developed to interface with an open-source software platform (Trodes – http://www.spikegadgets.com/software/trodes.html) and two hardware platforms (OpenEphys – http://www.open-ephys.org and SpikeGadgets – http://www.spikegadgets.com). We characterize system performance and show algorithmic limitations and discuss potential effects it has on further experiments.

Magnetoelectric materials for miniature, wireless neural interfaces
Amanda kenskens, Ben Avants, Josh Chen, Nishant Verma, Shayok Dutta, Josh Chu, Ariel Feldman, Caleb Kemere, Jacob Robinson
Developments in wireless neuromodulation lead to new treatments for neurological disorders and new studies in probing neural circuits in humans and animal models. Miniaturized and wirelessly powered biomedical implants are being developed to reduce negative host response caused by larger implant and leads. Wireless Power Transfer (WPT) methods cable of transmitting sufficient power in small form factors are required to implement these new biomedical devices. Conventional WPT techniques, when miniaturized, suffer from reduced power transfer, high angle dependence, and require high ~MHz frequency electromagnetic fields to carry the power, all of which limits the applications for any given device. Here we show magnetoelectric devices capable of transforming external magnetic fields to controllable electric fields strong enough to wirelessly stimulate targeted neural regions in freely moving rats with no genetic modification. We found that by coupling a piezoelectric and magnetostrictive material at an acoustic resonance, magnetoelectric films can stimulate cells in vitro when we apply an external magnetic field. We further show that these electric fields are strong enough to stimulate activity wirelessly by powering implanted electrodes in freely moving rats. Furthermore, in contrast to traditional inductive coupling, we show magnetoelectric materials are scalable and still capable of generating large voltages with a small device footprint. Our results demonstrate that magnetoelectric materials can be used to develop versatile lightweight wireless neural implants. We lay the foundation for further developing these materials to be used for many different applications in neuroscience.

Magnetogenetic: mechanism and new candidates
GUILLAUME DURET, SRUTHI POLALI, A. MARTIN BELL, BEN AVANTS, CONSTANTINE N. TZOUANAS, & JACOB T. ROBINSON
The wireless stimulation of specific neurons using magneto-sensitive channels will improve the exploration of neural networks, the mapping of regions of the brain that cannot be reached with optogenetic tools, and the stimulation of neurons in freely moving animals. Significant progresses have been achieved by associating TRP channels with artificial and biogenic magnetic nanoparticles such as ferritin. Channel gating occurs when these constructs are stimulated by alternative or static magnetic field. Despite yielding responses, the mechanism by which these nanoparticle-functionalized TRP channels respond to the magnetic stimulation is not known. The heat produced by ferritin in an alternating magnetic field is too weak to activate the nearby thermosensitive TRP channel and we know that the mechanical force between adjacent ferritin is at least eight orders of magnitude too weak for a mechanical stimulation. However, magnetic fields can produce a change in the magnetic entropy of biogenic nanoparticles, which in turn generates sufficient heat to gate temperature-sensitive ion channels. Based on our calculation and experimental data, we believe that this magnetocaloric effect is the underlying mechanism explaining the functional biogenic magnetogenetics proteins. Understanding this route of activation allows the development of new magneto-responsive proteins, and offers a putative new mechanism for magnetosensation in migratory animals.

Magnetogenetics for Drosophila
Charles Sebesta, Guillaume Duret, Constantine N. Tzouanas, Herman Dierick, and Jacob T. Robinson
Magnetically sensitive genetically encoded channels enable remote activation of specific neural networks deep within the brain of freely moving animals. Recent progress coupling various TRP channels and the biogenic magnetic nanoparticle ferritin in mammalian cells shows significant promise for creating such “magnetogenetic” technology, but this approach relies on ferritin assembly by endogenously expressed subunits. This reliance on endogenously expressed ferritin subunits makes it difficult to study the efficacy of ferritin formation and iron loading and prevents transfer of the technology developed with mammalian ferritin to species like Drosophila that express divergent ferritin proteins. In addition, magnetogenetics in Drosophila provides two key advantages for optimizing the magnetic sensitivity of the proteins: 1) because Drosophila ferritin is secreted and circulates throughout the animal within hemolymph, the particles can be more easily isolated for characterization, 2) once isolated, we can bind these ferritin nanoparticles to other proteins with minimal reduction of iron loading. By developing magnetogenetic channels for Drosophila with an extracellular binding site for functionalized ferritin or superparamagnetic nanoparticles, we can study the responses of the channels to low frequency magnetic stimulation with fewer unknowns and create driver lines expressing functionalized magnetogenetic ferritin nanoparticles. Drosophila also enable this technology to be rapidly tested in various cell types by taking advantage of large depositories of UAS/Gal4 drosophila lines. We believe developing these magnetogenetic channels and functionalized biogenic magnetic nanoparticles will create a testbed for behavioral studies in drosophila and enable researchers to rapidly test responses to stimuli throughout the animal while gaining a deeper understanding of the physics behind magnetogenetics.

Microfluidics for Electrophysiology, Imaging, and Behavioral Analysis of Hydra
Krishna N. Badhiwala, Daniel L. Gonzales, Daniel G. Vercosa, Benjamin W. Avants, Jacob T. Robinson
The cnidarian Hydra vulgaris provides an opportunity to discover the relationship between animal behavior and the activity of every neuron in a highly plastic, diffuse network of spiking cells. However, Hydra’s deformable and contractile body makes it difficult to manipulate the local environment while recording neural activity. Here, we present a suite of microfluidic technologies capable of simultaneous electrical, chemical, and optical interrogation of these soft, deformable organisms. Specifically, we demonstrate devices that can immobilize Hydra for hours-long simultaneous electrical and optical recording, and chemical stimulation of behaviors revealing neural activity during muscle contraction. We further demonstrate quantitative locomotive and behavioral tracking made possible by confining the animal to quasi-two-dimensional micro-arenas. Together, these proof-of-concept devices show that microfluidics provide a platform for scalable, quantitative cnidarian neurobiology. The experiments enabled by this technology may help reveal how highly plastic networks of neurons provide robust control of animal behavior.

mmWave Beam Acquisition and Steering with Practical Phased Array Antennas
Muhammad Kumail Haider, Yasaman Ghasempour, Edward W. Knightly
We present X60, the first SDR-based testbed for mmWave WLANs, featuring fully programmable MAC/PHY/Network layers, multi-Gbps rates, and a user-configurable 12-element phased antenna array. Combined these features provide an unprecedented opportunity to re-examine the most important aspects of signal propagation and performance expected from practical mmWave systems. Leveraging the testbed’s capabilities, we conduct an extensive measurement study, looking at different aspects of indoor mmWave links. We find that slight translational and rotational mobility can misalign antenna beams, severely degrading throughput. We study various adaptation schemes to address the issue of beam acquisition and alignment. Further, we present iTrack, an example strategy where we can steer mmWave beams by tracking indicator LEDs on wireless APs. Additionally, our comparison of different beam adaptation strategies reveals how beam steering even at one end of the link can often be sufficient to restore link quality.

Modeling Multi-User WLANs under Closed Loop Traffic
Peshal Nayak, Michele Garetto and Edward W. Knightly
In this project we perform the first cross-layer analysis of wireless LANs operating under downlink multi-user MIMO (MU-MIMO), considering the fundamental role played by closed loop (TCP) traffic. In particular, we consider a scenario in which the access point transmits on the downlink via MUMIMO, whereas stations must employ single-user transmissions on the uplink, as is the case in IEEE 802.11ac. With the help of analytical models built for the different regimes that can occur in the considered system, we identify and explain crucial performance anomalies that can result in very low throughput in some scenarios, completely offsetting the theoretical gains achievable by MU-MIMO. We discuss solutions to mitigate the risk of this performance degradation and alternative uplink strategies allowing WLANs to approach their maximum theoretical capacity under MU-MIMO.

Multi-Lobed Point Spread Functions Lead to Diameter-Dependent Heterogeneous Localization Bias for Plasmonic Nanowire – Dye Complexes
Rashad Baiyasi, Seyyed Ali Hosseini Jebeli, Qingfeng Zhang, Liang Su, Johan Hofkens, Stephan Link, Christy F. Landes
Plasmonic nanostructures offer a wealth of novel properties as nanocatalysts, but optimizing their structure-function relationship using superlocalization techniques is hindered by the formation of abnormal, non-Gaussian point spread functions. The resultant localization bias and extra localizations that result are investigated here for Ag nanowires labeled with Alexa-647 and imaged under remote excitation. Two main classes of abnormal PSFs are focused on: single-lobed PSFs exhibiting a variable localization bias based on position around the nanowire, and bi-lobed PSFs occurring near the top edge of the nanowire. The difference in localization bias, measured by the apparent width of the nanowire, is found to diverge for these two populations in the case of nanowire diameters below 300 nm. This indicates both that there is a larger population of bi-lobed PSFs under experimental conditions than predicted by simulation, and that nanowires with diameters in this range have the greatest potential for distinguishing between single-lobed and bi-lobed PSFs in experiment. We present a fitting method for these abnormal PSFs based on a basis of Hermite-Gaussian beam modes, and show that orientation information that would otherwise be inaccessible is encoded in bi-lobed PSFs. Future work focusing on detecting and localizing abnormal PSFs will enable the use of superlocalization techniques to identify and track fluorescent dyes and other quantum emitters on plasmonic nanowires.

Nanogapped Au Antennas for Ultrasensitive Surface-Enhanced Infrared Absorption Spectroscopy
Liangliang Dong, Xiao Yang, Chao Zhang, Benjamin Cerjan, Linan Zhou, Ming Lun Tseng, Yu Zhang, Alessandro Alabastri, Peter Nordlander, and Naomi J. Halas
Surface-enhanced infrared absorption (SEIRA) spectroscopy has outstanding potential in chemical detection as a complement to surface-enhanced Raman spectroscopy (SERS), yet it has historically lagged well behind SERS in detection sensitivity. Here we report a new ultrasensitive infrared antenna designed to bring SEIRA spectroscopy into the few-molecule detection range. Our antenna consists of a bowtie-shaped Au structure with a sub-3 nm gap, positioned to create a cavity above a reflective substrate. This three-dimensional geometry tightly confines incident mid-infrared radiation into its ultrasmall junction, yielding a hot spot with a theoretical SEIRA enhancement factor of more than 107, which can be designed to span the range of frequencies useful for SEIRA. We quantitatively evaluated the IR detection limit of this antenna design using mixed monolayers of 4-nitrothiophenol (4-NTP) and 4-methoxythiolphenol (4-MTP). The optimized antenna structure allows the detection of as few as ∼500 molecules of 4-NTP and ∼600 molecules of 4-MTP with a standard commercial FTIR spectrometer. This strategy offers a new platform for analyzing the IR vibrations of minute quantities of molecules and lends insight into the ultimate limit of single-molecule SEIRA detection.

Nanophotonic light sheet probes for deep imaging in scattering media
Fan Ye, Benjamin W. Avants, Ashok Veeraraghavan, Jacob T. Robinson
Optical techniques that measure changes in calcium or voltage provide a promising route toward to large scale measurement of neural activity. However, delivering light to specific regions deep within the brains is limited by the scattering of neural tissue. Here we show an implantable integrated photonic probe about 20 microns thick that produces a plane of illumination. With this planar illumination we can image more than 500 microns below the surface of a brain tissue phantom. Additionally, using the integrated photonic elements we scan this plane of illumination using no moving parts.

Non-Invasive Neurostimulation with High-Frequency Electromagnetic Wave
Shi Su, Behnaam Aazhang
Modulation of the brain circuits and impacting the electrical signaling of the brain is an important treatment for neuropsychiatric disorders. In this work, we propose a novel non-invasive neurostimulation approach using the high-frequency electromagnetic wave and beamforming with multiple antennas. By delivering the brain multiple electromagnetic beams at frequencies differ within the dynamic range of neural firing, we can modulate neural activities in the superposition of the beams with an envelope signal at the difference frequency. Such a method can achieve high spatial resolution in neurostimulation. Results from electromagnetic simulations are presented to demonstrate the electromagnetic propagation in the brain.

On the impact of blockage on the throughput of multi-tier millimeter wave network
Shuqiao Jia, Behnaam Aazhang
We take a look on the throughput scaling of the network we proposed. We analyze two blockage scenarios. The gold standard for the network throughput is the linear scaling. In the reasonable blockage scenario, the cutset upper-bound of multi-tier mmWave network achieves this gold standard. Even in the adverse blockage scenario, the throughput performance is decent. In these two figures, we compare the cut-set upper bound with the throughput achieved by our transmission strategy. The gap between two are log(n), which means our protocol achieves the optimality. Note that the throughput of the network was divided into two parts by this sweet spot. The first half, the throughput goes up with the AP density. In the second half, it goes down with the AP density. The intuition behind the left of the sweet spot is that the density of infrastructure help solving the blockage issue. The right part is limited by the number of RF chains.

Optical Force Reshapes Al-Au Nanodisk Heterodimers
Chao Zhang, Thejaswi Tumkur, Jian Yang, Minhan Lou, Liangliang Dong, Linan Zhou, Peter Nordlander, and Naomi J. Halas
Active control of the geometry of plasmonic nanostructures is of high interest in optical engineering, sensing, colorimetry, and laser printing. Conventionally, very intense laser pulses are required to melt the plasmonic (mostly Au) nanoparticles and the surface tension transforms the nanoparticles into more spherical shapes. In this work, we fabricated large scale, densely-dispersed, and uniform Al-Au nanodisk heterodimers with inter-disk spacing below 10 nm in average. Optical absorption of the Au nanodisk is enhanced by the adjacent Al nanodisk due to “forced plasmon” effect. When illuminating these heterodimers with laser pulses of moderate intensities at longitudinal polarization, we observed very noticeable reshaping of the Au nanodisks in the heterodimers, while the Al nanodisks remained unchanged. The temperature increase of the Au nanodisks was not enough to induce such reshaping. Instead, it only softens the Au lattice and the optical force plays a major role in the reshaping process. By tracking the reshaping of individual heterodimers, we found a statistical trend that the degree of reshaping increases with decreasing gap size, which is consistent with the distance-dependent forced plasmon picture. The change of the optical property of the heterodimers due to reshaping was also studied using darkfield scattering spectroscopy.

Performance Analysis of Spatial Massive MIMO with Zero-forcing
Xu Du, Ashutosh Sabharwal
Zero-forcing beam-forming is one of the beam-former candidate of 5G massive MIMO due to its near optimal performance (at high SNR). It this work, we present first results on a closed-form rate characterization of a spatially correlated massive MIMO downlink system with zero-forcing.

Photoluminescence of Gold Nano Rods
Behnaz Ostovar, Stephan Link
In this research, we provide a systematic study on photoluminescence (PL) in gold nanorods (AuNRs) with different sizes. We investigated the physical origin of shortband peak in PL spectra by measuring PL spectra of different sizes of AuNR. In order to better understand the underlying mechanism for this peak, we also conducted power dependence and polarization dependence experiments. The expected outcome of this project is the better understanding of the underlying process of electronic decay pathways in plasmonic nanostructures.

Photon Emission by Inelastic Tunneling from Plasmonic Junctions
Ali Mojibpour, Palash Bharadwaj
Electrical excitation of plasmonic nanoparticle provides on-chip plasmon sources that will have applications from sensing to optical interconnects. Here, we study single plasmonic-nanoparticle tunnel junctions.

Physics-Based Renderer for Time-of-Flight Cameras
Adithya Kumar Pediredla, Ioannis Gkioulekas, Ashok Veeraraghavan
Transient imaging or Time-of-flight imaging is the recent imaging revolution, in which we measure both the number of photons and the time-of-arrival of each photon. Time-of-flight cameras are finding applications in various fields ranging such as chemistry (fluorescence lifetime imaging), archaeology (mapping of ancient cities), meteorology (cloud profiling), automobiles (ranging and remote sensing), defense (imaging around the corners), and health (imaging through the skin). The proliferating field, with several techniques to measure time-of-flight, currently suffers from the lack of accurate physics-based simulators that can help in the co-design of hardware and algorithms. We develop a time-of-flight renderer that can simulate any arbitrary time-of-flight camera on the market for a given arbitrary scene (varying geometry, scale, material properties including specularity, texture, and subsurface scattering). The renderer is physics-based, multi-threaded, parallelizable on a cluster and is open-sourced at https://github.com/cmu-ci-lab/MitsubaToFRenderer/. Additionally, we show inverse rendering applications by reconstructing the shape of a hidden room if all we can observe is one wall (presumably from a window).

PPGSecure: Biometrics Presentation Attack Detection Using Photoplethysmograms
Ewa Magdalena Nowara, Ashutosh Sabharwal, Ashok Veeraraghavan
Authentication of users by exploiting face images or videos as a biometric is quite commonplace and becoming more widespread due to the advances made in face recognition technologies. While face recognition has made rapid advances in its performance, such face-based authentication systems remain vulnerable to biometric presentation attacks, such as the presentation of a video or photograph on a display device, the presentation of a printed photograph or the presentation of a face mask resembling the user to be authenticated. In this paper, we present PPGSecure, a novel methodology that relies on camera-based physiology measurements to detect and thwart such biometric presentation attacks. PPGSecure uses photoplethysmogram (PPG), which is an estimate of vital signs from the small color changes in the video observed due to minor pulsatile variations in the volume of blood flowing to the face. We demonstrate that the temporal frequency spectra of the estimated PPG signal for real live individuals are distinctly different than those of presentation attacks and exploit these differences to detect presentation attacks. We demonstrate that PPGSecure achieves significantly better performance than existing state of art methods on photograph, image and video based presentation attacks.

PulseCam: High-resolution and motion-robust blood perfusion imaging using a camera and a pulse oximeter
MAYANK KUMAR, J AMES W. SULIBURK, ASHOK V EERARAGHAVAN, AND ASHUTOSH S ABHARWAL
Blood perfusion is the flow of oxygen-rich blood to the end organs and tissue through the blood vessels. Measuring blood perfusion has widespread clinical applications, e.g., to monitor blood flow into surgical sites, to diagnose peripheral arterial disease, and to monitor patients in the ICU. Recently, a new camera-based blood perfusion imaging modality has been proposed which can be used to continuously and non-invasively measure spatial maps of peripheral perfusion. But, current camera-only blood perfusion imaging modality suffers from two core challenges: (i) motion artifact due to movement of skin surface relative to the camera, and (ii) small signal recovery in the presence of measurement noise and large surface reflection or shunt signal. In this paper, we propose a new multi-sensor blood perfusion imaging methodology named PulseCam that robustly fuses measurements from a video camera and a conventional pulse oximeter to perform blood perfusion imaging in the presence of motion artifacts and outliers in the video recordings. We also propose a new brightness-invariant optical flow computation approach for blood perfusion imaging that helps us reduce error in blood perfusion estimate below 10% in different motion scenarios compared to 30 40% error when using current approaches. Finally, we demonstrate the feasibility of using our proposed PulseCam methodology to detect statistically significant differences (p<0:005) in blood perfusion estimate during arterial and venous blood flow occlusion experiment done on 12 participants of varying skin tones. Thus, PulseCam can have potential clinical application in monitoring blood perfusion in different surgical sites including lower extremity revascularization and free-flap re constructive surgery.

Reducing Network Latency Shenanigans
Fatima Ahsan and Ashutosh Sabharwal
Random access is a crucial building block for nearly all wireless networks, and impacts both the overall spectral efficiency and latency in communication. In our research, we analytically show that the spatial degrees of freedom, e.g. available in massive MIMO systems, can potentially be leveraged to reduce random access latency. Using one-ring propagation model, we evaluate how the random access collision probability depends on the aperture size of the array and the spread of user’s signal Angle-of-Arrivals (AoAs) at the base-station, as a function of the user-density and the number of random access codes. Our numerical evaluation shows that for practically sized large arrays in outdoor environments, a significant reduction in collision probability is possible, which in turn can decrease the random access latency.

Seizure Prediction via Extracting Qualitative Information from Signal Dynamics
Negar Erfanian, Dr. Behnaam Aazhang, Dr. Nitin Tandon
Suffering from unprovoked, frequent seizures in which hypersynchronous neural activity spreads from one or more small diseased areas on the brain to a broader portion of it, around three million patients in the US are affected by epilepsy. As a result, predicting seizures in these patients will provide the chance to help them deal with the seizures in a more comfortable way or in some cases prevent the seizures from happening. Extracting information from dynamics of the signals recorded from the exposed surface of the brain of these patients, with the aim of seizure prediction, is the main goal of this work. In this project, I am evaluating the changes in the dynamics of the epileptic data recorded from the seizure onset zone of the brain, in both single channel and across different channels to find out about the different states that are existing on the epileptic recordings. In order to do so, I bring my time-series data from the time domain to the phase domain, using phase space reconstruction methods, which reveals the evolution of states of the data. Extracting and defining individual states in this domain is feasible by estimating Lyapunov Exponent of the epileptic data, which gives us information on how much the future state of the time series is predictable. By differentiating the preseizure state from other periods of the epileptic data, we will have a tool to predict the seizures.

Single Photon Emitters in Hexagonal Boron Nitride and Optothermal Characterization of Two Dimensional Alloys
James Kerwin, Palash Bharadwaj
Our group is pursuing research in the optical characterization of two dimensional materials. This includes the creation and engineering of single photon emitters hosted in the wide band gap semiconductor, hexagonal boron nitride. Also, we are working to explore the thermal and optothermal properties of tungsten and molybdenum transition metal dichalcogenide alloys.

Single-Frame 3D Fluorescence Microscopy with Ultra-Miniature Lensless FlatScope
Jesse K. Adams, Vivek Boominathan, Benjamin W. Avants, Daniel G. Vercosa, Fan Ye, Richard G. Baraniuk, Jacob T. Robinson, Ashok Veeraraghavan
Increased interest in fluorescence imaging of neural networks has driven the demand for smaller, lighter weight, and larger field-of-view microscopes. Miniaturized lensed-microscopes take advantage of advances in fabrication technology but still suffer a fundamental tradeoff between less light collection or smaller field-of-view. Overcoming this fundamental trade-off, we present a new 3D fluorescence imaging concept that replaces lenses with an optimized amplitude mask positioned only a few hundred microns above the imaging sensor and an efficient computational algorithm that can convert a single exposure into high-resolution 3D images. This innovation, we call FlatScope, is a lensless microscope scarcely thicker than an image sensor (roughly 0.2 grams in weight and less than 1 mm thick), but able to produce micron-resolution, high-frame-rate, large field-of view 3D fluorescence movies covering a total volume of several cubic millimeters.

SocialSense: Estimate “physical social network” from unconstrained audio recordings
Jian Cao, Ashutosh Sabharwal
Sociability is an important factor that influences the quality of life, the health and wellbeing of an individual, as well as the workplace performances and learning outcomes. However, current methods for sociability measure mainly rely on surveys and interviews, and hence are often subjective and biased. In this project, we propose to design and evaluate the SocialSense framework, which constructs the “physical social network” of a person from audio recordings of unconstrained conversations over time. The framework uses semi-supervised approach to gradually identify the social contacts and grow the egocentric network through longitudinal data across multiple conversations. Through this project, we hope to provide an alternative approach for objective, quantitative and privacy-preserving sociability measure.

STRUCTURE-DIRECTED IMAGE MODELING AND ITS APPLICATION TO IMAGE INTERPOLATION
Lantao Yu, Michael Orchard
This poster explores the development and the use of a complex-valued, multi-resolution image representation to model and exploit local image structure in image-processing applications. Our approach distinguishes itself from prior approaches in the way we exploit both grayscale smoothness and smoothness of curves and shapes in images. Coeffi- cients of our representation have magnitudes that measure the image energy within a specific region in both space and frequency, and phases that carry information about the distance of that energy from a local reference position. Our work proposes to model statistical relationships, both across spatial regions and across different frequency bands, among the field of coefficient magnitudes, and among the field of coefficient phases. To illustrate the advantages of modeling these relationships, we present an algorithm for interpolating a natural image by a factor of two, both horizontally and vertically. Relationships among phases and magnitudes of available bands of coefficients are exploited to estimate local edge parameters (e.g. location, orientation, sharpness) that provide information about higher-frequency coefficients that are not available in the original image. Compared with state-of-the-art interpolation algorithms, our work shows improved PSNR around edges, better preserves the edge sharpness, and produces more pleasing contour smoothness.

The Best of Both Worlds: ‘Antenna-Reactor’ Nanostructures for Plasmonic Photocatalysis
Dayne F. Swearer, Hossein Robatjazi, Linan Zhou, Chao Zhang, Rowan K. Leary, Sadegh Yazdi, Hangqi Zhao, Paul A. Midgley, Peter Nordlander, Emilie Ringe, Naomi J. Halas
Developing materials that can efficiently harvest solar radiation and drive chemical reactions is one route to realize the wide scale adoption of solar-driven chemical manufacture. This achievement could help wean the chemical industry off of fossil fuels, a major source of global energy consumption and anthropogenic atmospheric pollution. Yet this has been a challenge, since transition metal nanoparticles traditionally used in heterogeneous catalysis do not effectively couple with light. On the other hand, plasmonic metals such as Au, Ag, and Al, interact strongly with light but are often unsuitable to drive diverse chemical reactions on their surfaces. We are developing a new platform of modular ‘antenna-reactor’ nanostructures that combine the best of both worlds: a plasmonic ‘antenna’ and catalytic ‘reactors’ into a single structure with optimized optical and catalytic properties. We utilize aluminum nanocrystals (AlNCs) as plasmonic antennae because Al is the most abundant metal in the Earth’s crust, and sustains strong plasmon resonances throughout the ultraviolet, visible, and near infrared regions of the electromagnetic spectrum. The AlNCs are decorated with 3-5 nm islands of transition metals, of which over a dozen varieties have been synthesized. AlNCs have a 2-4 nm self-limiting native oxide layer that passivates the underlying crystal, and acts as a thin physical barrier between the antenna and reactor. Material characterization using high-resolution electron microscopy, energy dispersive X-ray spectroscopy, and electron tomography revealed the morphological details and true 3D nature of these materials; important for understanding structure-function relationships within this new class of photocatalyst. The close proximities between the antenna and reactor results in significant absorption enhancements in the catalytic materials. This increased absorption in the optically lossy catalytic metal leads to high hot-carrier distributions within the materials, which leads to new mechanistic pathways that show potential for selective chemical transformations. For example, on Al-Pd, hydrogenation of acetylene resulted in significantly increased selectivity of ethylene production over ethane. Similarly, CO2 reduction to CO was achieved on Al-Cu2O with 99.97% selectivity over CH4, which is the main product under traditional thermal conditions. This interdisciplinary work shines light on the fundamentals of plasmon-mediated chemistry, nanomaterial synthesis and characterization, heterogeneous catalysis, electron transfer processes, and nanoscale optics. The modularity in material design and the growing library of photocatalyzed reactions using antenna-reactor nanostructures may one day allow for light-driven chemistry to make the food, medicine, and materials of tomorrow.

The Rotating Magnetocaloric Effect as a Mechanism for Magnetosensation
A. Martin Bell and Jacob T. Robinson
Many animals demonstrate the ability to detect and respond to magnetic fields. This magnetic sense is used to navigate and orient with respect to the earth’s magnetic field by detecting a combination of field strength, orientation, inclination, and polarity. While the existence of a magnetic sense is well-established, the mechanism or mechanisms underlying magnetosensation are not known, though models have been proposed that depend on radical pair generation or on mechanical detection of magnetic force. Here we propose a mechanism of magnetosensation based on the magnetocaloric effect, where thermally sensitive ion channels respond to energy generated by the entropy changes resulting from rotation of a magnetically anisotropic material in the presence of a magnetic field. We explore how the magnetocaloric effect could be used as a natural sense and identify paths to experimentally test this model. We establish a parameter space for candidate magnetic materials, and outline experiments to distinguish magnetic senses resulting from the magnetocaloric mechanism from those resulting from chemical or magnetomechanical mechanisms. The magnetocaloric hypothesis provides a new hypothesis for a magnetosensation mechanism that may be able to resolve some of the unanswered questions about how animals sense the earth’s magnetic field.

Theseus: A State Spill-free, Runtime Composable Operating System
Kevin Boos, Lin Zhong
The underdiagnosed problem of state spill remains a barrier to realizing complex systems that are easy to maintain, evolve, and run reliably. We share our experience building Theseus from scratch, an OS with the guiding principle of eliminating state spill. Theseus takes inspiration from distributed systems to rethink state management, and leverages Rust language features for maximum safety and efficient isolation. We intend to demonstrate Theseus as a runtime composable OS, in which entities are easily interchangeable and can evolve independently without reconfiguring or rebooting, even in the heart of the kernel.

Two-Dimensional Active Tuning of an Aluminum Plasmonic Array for Full-Spectrum Response
Michael Semmlinger, Ming Lun Tseng, Jian Yang, Chao Zhang, Peter Nordlander, and Naomi J. Halas
Color pixels composed of plasmonic nanostructures provide a highly promising approach for new display technologies, capable of vivid, robust coloration and incorporating the use of low-cost plasmonic materials, such as aluminum. Here we report a plasmonic device that can be tuned continuously across the entire visible spectrum, based on integrating a square array of aluminum nanostructures into an elastomeric substrate. By stretching the substrate in either of its two dimensions, the period and therefore the scattering color can be modified to the blue or the red of the at-rest structure, spanning the entire visible spectrum. The unique two-dimensional design of this structure enables active mechanical color tuning, under gentle elastic modulation with no more than 35% strain. We also demonstrate active image switching with this structure. This design strategy has the potential to open the door for next-generation flexible photonic devices for a wide variety of visible-light applications.

Upgrading an Open-Source Miniaturized Fluorescent Microscope for Neural Closed-Loop Stimulation in Freely-Behaving Rodents
Jill Juneau, Caleb Kemere
The goal of my research is to study learning and memory circuit pathways in the brain by investigating neural network activity using brain-machine interfaces. Understanding changes in neural networks during learning in rodent models can provide insight into human cognitive performance and memory related pathways. Building on this understanding, we can develop and test innovative therapies geared to alleviate symptoms of memory related trauma. The brain-machine interface I specifically use is a head-mounted fluorescent microscope that can image neural activity of a freely behaving mouse. This technology allows for visualizing the network activity of a learning/memory related brain region, such as the hippocampus, while the animal completes a learning task. Important to studying the plasticity of learning and memory circuits, this technology allows long-term imaging of the same region of the brain compared to the short-term recording windows available with current electrophysiology methods. Our goal is to use this technology to create a real-time neural detection/stimulation system. However, the current microscope has limitations in that it has slow frame rates and low sensitivity levels. Even though the system works well with imaging the bright fluorescent kinetics of the genetically encoded calcium indicator GCaMP, it cannot image the faster and lower fluorescent signals from genetically encoded voltage indicators. Therefore, we present an updated image sensor module that aims at addressing these shortcomings. Furthermore, the demo portion of this presentation will demonstrate the current microscope as well as presenting in vivo microscope imaging video.

Postdocs/guests/startups

Internet of Things: Wearable Anger Management Tool
Indhusri Gunda; Mentor: Professor Ray Simar
I am working on a wearable device to help with anger management. It is a “bracelet” that uses a muscle flex sensor and pulse sensor that connect to a microcontroller. These sensors help detects signs of anger such as making a fist or an increasing pulse. When it detects these signs, it alerts the user through a message on the display of the wearable. This device will connect through Bluetooth to an iPhone application that will display daily/weekly graphs of the sensor readings so the user will have access to the data being collected. I want to make this device because oftentimes when people are angry, they act in a manner or say things they later regret later. I hope my device can act as a reminder or even a distraction to people to help them remember to think about the consequences of their words or actions. Currently I am at the prototyping stage and will discuss my progress and future plans.

Microfluidic actuation of flexible microelectrodes for neural recording
Daniel G. Vercosa, Flavia Vitale, Alexander V. Rodriguez, Sushma Sri Pamulapati, Frederik Seibt, Erik M. Lewis, Jiaxi Stephen Yan, Krishna Badhiwala, Mohammed Adnan, Gianni Royer-Carfagni, Michael Bei
Ultra-small, flexible neural electrodes with diameters the size of individual cells significantly increase the quality and longevity of neural recordings by reducing neural injury during chronic implantation; however, flexible electrodes are traditionally difficult to implant and position without causing acute damage. Researchers typically apply stiffening agents that temporarily increase the overall size and rigidity of the electrode in order to overcome the buckling force upon implantation. The resulting increased electrode footprint leads to cell loss and glial activation during implantation that persists even after the stiffening agents are removed or dissolved. Here, we show minimally invasive implantation of flexible microelectrodes using a specially designed microdrive utilizing microfluidics that eliminates the need for stiffening agents or shuttles. Using microfluidic vent channels, we can minimize the amount of fluid directed towards the brain during insertion, further reducing potential trauma. We show the versatility of this technique by demonstrating in vitro recordings and precise actuation in mouse brain slices and acute in vivo recordings from anesthetized rats. Additionally, we show that we are able to record neural activity from several locations. Future work will focus on implanting flexible, multichannel electrodes in vivo and quantifying differences between microfluidic and traditional implant methods.

Skylark Wireless: Massive MIMO Software-Defined Radio Systems for Next-Generation Wireless
Ryan E. Guerra; Clayton W. Shepard
Skylark Wireless LLC was spun out of Rice University ECE to develop advanced next-generation wireless solutions for research and development and rural broadband markets. In 2015, the company started developing a commercial 5G Massive Multiple-Input, Multiple-Output (Massive MIMO) wireless solution for last-mile rural broadband. Skylark began installation of a proof-of-concept multi-cell research network, “ArgosNet” on Rice University campus in late 2017. Skylark’s proprietary many-antenna multi-user beamforming technology and systems enable unprecedented long-range and high-speed point-to-multi-point communication, making it economically viable and sustainable to provide true broadband internet connectivity to over 40 million under-served Americans and over 3 billion people internationally without any internet access.