Poster/Demo Session

Undergraduate Projects

Automated Control and Monitoring System for Water Waste Reduction in Scale Oil Field (Senior Design)
Caleb Lu, Ethan Myers, Jonathan Nguyen
Faculty Advisors: Matthew Elliot, Gary Woods
When oil and gas are produced from the ground, large amounts of waste water is produced as well. Waste water mitigation and disposal is a billion dollar per year problem for the oil and gas industry. We have developed a scalable solution that monitors and automatically controls a model oil field that reduces the amount of produced waste water. Authors: Ankush Agrawal, Andrew Elsey, Caleb Lu, Ethan Myers, Jonathan Nguyen, Paul Saad, Matthew Elliott, Gary Woods.

Clinically Viable Motion Tracking and Analysis Device for Cerebral Palsy Patients (Senior Design)
Erica Chen, Romeo Garza, Ethan Heng, Young Yul Kim, Jason Liu, Tianyi Yao
Faculty Advisors: Dr. Gary Woods
Various devices on the market assess the functional dexterity (Functional Dexterity Test) of Cerebral Palsy patients. However, none of them are suitable for everyday clinical use; they are either too expensive and immobile, or they lack in accuracy. Our project aims to design and implement a lightweight, portable, and reliable motion path tracking and analyzing device that is accurate enough for clinical use. Our device should achieve 3 key functionalities: object tracking, 3D location calculation, and motion-path reconstruction. During FDT, patients pick up a peg, rotate it, and place it back in a certain location. Throughout the process, our device recognizes the peg the patient is holding, calculates the 3D location of the peg, and reconstructs the motion path. We incorporate the Microsoft Kinect with the inertial measurement units to measure the position, and the linear and rotational velocity. The inertial measurement units inside the peg transmit data wirelessly using Wi-fi, and the data of the Kinect and the IMU are statistically processed to accurately predict the actual motion path taken. The resulting motion path from the given data is successfully close to the actual path within 1 cm. Even when some intermediate points are missing, the predicted values from the statistical processing are within 3 cm from the actual path. This project suggests a potential solution for a clinically viable assessment system at a much lower price than the existing solutions. In the future, this project will allow the quantitative analysis of the resulting parameters. These may be researched further for their relevance to the disease and its treatment. Authors: Dr. Gary Woods, Erica Chen, Romeo Garza, Ethan Heng, Young Yul Kim, Jason Liu, Tianyi Yao.

Digital Gym: Automated Evaluation of Squash Games
Betty Huang
Faculty Advisor: Ashutosh Sabharwal
As a part of the Rice Digital Gym project, we are developing a computer vision based system to automatically evaluate squash games played in the Rice recreational center. Our objective is to extract the location and orientation of players and the moment in time when contact is made between the racket and the ball. In order to do so, we combine methods from computer vision and machine learning in the ongoing algorithm development. Authors: Betty Huang, Ashu Sabharwal.

Efficiency sensor for energy optimization
Manuel Pacheco, Zephyr Smith, Irene Zhang, Collin Allen
Faculty Advisors: Gary Woods, Guru Naik
The REV team at rice builds an electric car to compete in the Shell Eco-Marathon every year. Our goal is to maximize efficiency for the car. By creating a sensor for power and for speed, we can find the instantaneous efficiency (miles / Joule), feed this data real-time to the driver, and optimize motor controls to make our car’s energy usage more efficient. Authors: Manuel Pacheco, Zephyr Smith, Austin Meng, Clark Zha, Colin Shi, Collin Allen, Damien Ng, Irene Zhang, Louis Smidt.

FPGA System for Real-Time Seizure Prediction (Senior Design)
Erik Biegert, Sarah Hooper, Marissa Levy, Justin Pensock, Luke Van der Spoel, Xiaoran Zhang
Faculty Advisors: Behnaam Aazhang, Gary Woods
Of the 3 million Americans who suffer from epilepsy over a third cannot be treated with medication. Our goal is to create an alternative treatment by developing an implantable neurostimulator that applies low-frequency stimulation to suppress seizures. In past years this Vertically Integrated Project team has developed a predictive machine learning algorithm that detects when a seizure is approaching. This year we will develop custom hardware to implement this algorithm in real time. Authors: Erik Biegert, Sarah Hooper, Marissa Levy, Justin Pensock, Luke Van der Spoel, Xiaoran Zhang.

Hippotherapy Simulator (Senior Design)
Paul Byrnes, Leah Chong, Shengliang Zhu, Linus Shih, Chandler Burke, Sihan Zeng
Faculty Advisors: Gary Woods, Matthew Elliott
Hippotherapy uses horseback riding to improve coordination, balance, and core strength, and has been used to treat various conditions such as autism, cerebral palsy, arthritis, Down syndrome, and PTSD. The aim of this project is to build a mechanical hippotherapy device that provides accurately-simulated therapeutic riding experience for patients to increase the accessibility of hippotherapy. Our device will be safe for human trials and adaptable to different demographics. Our team started with a prototype of last year’s team. This year, our goal is to improve the safety, accuracy of simulation, adaptability, affordability and durability of last year’s prototype. Authors: Paul Byrnes, Leah Chong, Shengliang Zhu, Linus Shih, Chandler Burke, Sihan Zeng.

Improving Respiratory Support Outcomes in Premature Babies (Senior Design)
Aaron Velasquez-Mao, Alice Xie, Chris Sabbagh
Faculty Advisor: Gary Woods
Nearly half of neonates placed on CPAP Therapy for respiratory support fail treatment. We hypothesize that CPAP failure is caused by misalignment or obstruction of the airway nasal prongs. Currently, doctors check routinely for air supply obstruction via stethoscope, but the infrequency of these checks causes extended periods of poor nasal prong positioning. Thus, we have designed a continuous and automated system for monitoring CPAP use in neonates receiving respiratory support to improve response time and to create a testing platform by which analytical data can be obtained on CPAP effectiveness. Our system uses an acoustic sensor that attaches to patients via an FDA-approved sticker, a microcontroller that sends data to a central repository within Texas Children’s Hospital, and a signal detection algorithm for alerting improper use. Here, we present our comprehensive system, preliminary detection algorithms, and our testing plan for applying this system to clinic. Authors: Dr. Gary Woods, Aaron Velasquez-Mao, Alice Xie, Jennifer Dawkins, Chris Sabbagh, Tony Ren, Anna Norris.

LITSCOPE: Mobile Fourier Ptychographic Microscopy (Senior Design)
Brian Brenner, John Haug, Morganne Lerch, Mia Polansky, Vinay Raghavan, Sujay Tadwalkar
Faculty Advisor: Ashok Veeraraghavan, Gary Woods
Fourier ptychography, an emerging technique in imaging, reconstructs high resolution and high field-of-view images from a series of low-resolution images. Our team aims to employ Fourier ptychography to provide field researchers and medical workers in low-resource areas access to high-quality imaging without the bulk and the price tag associated with conventional microscopy. Authors: Brian Brenner, John Haug, Morganne Lerch, Mia Polansky, Vinay Raghavan, Sujay Tadwalkar, Ashok Veeraraghavan, Gary Woods.

PalliAssist: Palliative Care Software for Low Resource Settings (Senior Design)
Senthil Natarajan, Steven Tsai, Cassie Wang, Aakash Shah
Faculty Advisors: Maria Oden, Gary Woods
Palliative care — the prioritization of pain management over aggressive treatment of severe illness — is rapidly gaining popularity in Brazil and other countries throughout the world. The Barretos Cancer Center in Barretos, Brazil, is one of hospitals at the forefront of this type of treatment. Nonetheless, it faces challenges regarding communication between patients and their palliative care physicians. To help remedy this, the hospital has recently adopted WhatsApp as a means of communication. However, even though WhatsApp is free, ubiquitous, and capable of transferring images and text, the app is limited in the scope of the data it can transfer, lacks organization of information, and cannot ensure a secure, central, and real-time storage. We are developing software that is accessible by web and mobile to help patients better convey their health status, symptoms, and needs to their doctors. This project will enable efficient, organized two-way communication and create a holistic clinical tracking interface for managing palliative care data between doctors and patients. Our software solution is split into two components – a core mobile application and a satellite web portal for doctors to manage and keep track of all of their patients. The software will allow the patients to specify in detail the symptoms they are experiencing to communicate to their doctor. The doctors will receive these notifications in real-time and can refer to the patient’s medical history and records from the same software. The doctors will then be able to use the software to communicate their advice and any necessary medical treatment back to the patients. Patients can also use the software to manage their medical prescriptions and usage history. While there is no singular or perfect solution, we aim for our software to be a standard for at-home palliative care management in Barretos and other low-resource environments. Authors: Senthil Natarajan, Steven Tsai, Cassie Wang, Aakash Shah.

Patient Specific, Data-Driven, Multisite Pacemaker (Senior Design)
Greg Harper, Daniel Zdeblick, Yamin Arefeen, Jorge Quintero, Philip Taffet
Faculty Advisors: Mehdi Razavi, Benhaam Aazhang, Joseph R. Cavallaro
Current wired-pacemakers involve physical leads that connect electrodes inside the heart to a control unit. While multi-site, the number of transvenous leads is limited, with each additional lead increasing the risk of complications, as well as the inability of these leads to reach all 4 chambers of the heart. Current wireless-pacemakers can only pace and sense in one chamber of the heart. Our team is designing and creating the external control unit that will leverage both multi-site and wireless technologies to remove lead-based complications and provide holistic sensing and pacing. Authors: Senthil Natarajan, Steven Tsai, Cassie Wang, Aakash Shah.

Realtime Vibrotactile Feedback System For Prosthetic Users (Senior Design)
Ryan Li, Keiko Kaplan, Fanny Huang, David Fraga
Faculty Advisors: Gary Woods, Matthew Elliott, Eric Richardson
Current prosthesis products has limited capabilities in providing users sufficient control of and feedback from the prosthetic limb. Our sponsor, NeoSensory, has developed the Versatile Extra-Sensory Transducer (VEST), a system based on sensory substitution device that conveys information through vibration on the VEST. Using the VEST we designed a low-cost, non-invasive device that can provide real-time and interpretable vibrational feedback to lower-limb prosthesis users. Authors: Dr. Garry Woods Ryan Li Keiko Kaplan Fanny Huang David Fraga Samuel Soyebo Isabella Yang.

Rice Digital Gym
Carl Henderson, Titus Deng, Aidan Curtis
Faculty Advisor: Ashutosh Sabharwal
Rice Digital Gym project aims to turn Rice gym into a massive health sensor. The Digital Gym will serve as a foundation to measure how exercise impacts the wellness and health of a whole community, in this case, the Rice community. To convert the gym into a massive sensor, we are developing mountables. Mountables can be mounted on gym equipment or gym space and will convert “analog” equipment/spaces to “networked digital” gym equipment/spaces, to capture various exercise activities in the gym. Combined with a smartphone/tablet app where the users opt-in, the end result is that we can capture a whole slew of exercise behavior without any wearables. Our opt-in procedure ensures complete privacy of gym users who do not want their exercise behaviors measured. Authors: Carl Henderson, Aidan Curtis, Titus Deng, Irene Zhang, Jose Pacheco, Ashwin Hareesh, Ashutosh Sabharwal.

Text-based Financial Prediction Models
Ragib Mostofa
Faculty Advisors: Vivek Sarkar, Anshumali Shrivastava
We attempt to predict stock prices using learning algorithms that incorporate financial news with other features. In particular, we try to build a model that allows the machine to extract relevant information from news related to the stock market and utilize that information in addition to information gleaned from different patterns that appear in historical stock price movements, to predict the direction in which the next day’s stock price moves towards. This is a 3-class classification problem – the price is expected to RISE (Bull Market), FALL (Bear Market) or STAY constant (Stagnant Market). Our results show that when just textual features are used the learning algorithm can successfully predict the movement of stock prices with an accuracy of 60.29%. On the other hand, an accuracy of 61.45% was seen when only pattern features were used. However, when the two types of features were used simultaneously, the prediction accuracy underwent an increase to 64.93%. In terms of simple accuracy, this final model has an accuracy level that exceeds a simple coin toss accuracy for a binary classification model (50%) by about 14.93%. Moreover, our proposed model performs considerably better than most other models that have been proposed as a solution to this research problem in the past. We hope that trading entities of all sizes, take advantage of our model and enjoy a competitive edge in the trading world today. Authors: Ragib Mostofa, Joshua Phipps.

Wrigley: Team DISSECT’s Fun Sized Long Lasting PCB
Christopher Chivetta, Chris Chee
Faculty Advisor: Ray Simar
Wrigley is a new PCB design by the Vertically Integrated Project Team DISSECT. It measures in at the size of a piece of chewing gum, and expands on the capabilities of the Texas Instruments Tiva C Series Launchpad, commonly used in the Rice ECE department. With less than half the size and a 250% increase in pin connections from the TI Tiva C Series Launch Pad, Wrigley is a powerful and conveniently sized embedded hardware device that aims to be a robust and integral part of any Rice Engineering project. Authors: Christopher Chivetta, Christopher Chee.

Graduate Projects

A Characterization of State Spill in Modern Operating Systems
Kevin Boos
Faculty Advisor: Lin Zhong
Understanding and managing the propagation of states in operating systems has become an intractable problem due to their sheer size and complexity. Despite modularization efforts, it remains a significant barrier to many contemporary computing goals: process migration, fault isolation and tolerance, live update, software virtualization, and more. Though many previous OS research endeavors have achieved these goals through ad-hoc, tedious methods, we argue that they have missed the underlying reason why these goals are so challenging: state spill. State spill occurs when a software entity’s state undergoes lasting changes as a result of a transaction from another entity. In order to increase awareness of state spill and its harmful effects, we conduct a thorough study of modern OSes and contribute a classification of design patterns that cause state spill. We present StateSpy, an automated tool that leverages cooperative static and runtime analysis to detect state spill in real software entities. Guided by StateSpy, we demonstrate the presence of state spill in 94% of Android system services. Finally, we analyze the harmful impacts of state spill and suggest alternative designs and strategies to mitigate them. Authors: Kevin Boos Emilio Del Vecchio Lin Zhong.

A Latent Factor Model For Instructor Content Preference Analysis
Jack Wang
Faculty Advisor: Richard Baraniuk
Existing personalized learning systems (PLSs) have primarily focused on providing learning analytics using data fromlearners. In this paper, we extend the capability of current PLSs by incorporating data from instructors. We propose a latent factor model that analyzes instructors’ preferences in explicitly excluding particular questions from learners’ assignments in a particular subject domain. We formulate the problem of predicting instructors’ question exclusion preferences as a matrix factorization problem, and incorporate expert-labeled Bloom’s Taxonomy tags on each question as a factor in our statistical model to improve model intepretability. Experimental results on a real-world educational dataset demonstrate that the proposed model achieves superior prediction performance compared to several other baseline methods commonly used in recommender systems. Additionally, by explicitly incorporating Bloom’s Taxonomy, the model provides meaningful interpretations that help understand why instructors exclude certain questions. Since instructor preference data contains their insights after years of teaching experience, our proposed model has the potential to further improve the question recommendations that PLSs make for learners. Authors: Jack Wang, Andrew Lan, Phillip Grimaldi, Richard Baraniuk.

A near-infrared gas sensor system based on tunable laser absorption spectroscopy and its application in CH4/C2H2 detection
Qixin He, Dr. Weilin Ye
Faculty Advisor: Frank Tittel
A near-infrared (NIR) dual-channel differential gas sensor system was experimentally demonstrated based on tunable laser absorption spectroscopy (TLAS) and wavelength modulation spectroscopy (WMS). The sensor consists of four modules, including distributed feedback (DFB) lasers for the detection of targeted gasses, a custom portable DFB driver compatible for butterfly-packaged DFB lasers, a 20cm-long open-reflective gas-sensing probe and a custom cost-effective lock-in amplifier for harmonic signal extraction. The optical and electrical modules were integrated into a standalone sensor system, which possesses advantages of user-friendly operation, good stability, small volume and low cost. With different DFB lasers, the sensor system can be used to detect different gasses. Two DFB diode lasers with emission wavelengths of 1.65 µm and 1.53 µm were used to detect CH4 and C2H2, respectively. Standard CH4 and C2H2 samples were prepared and experiments were carried out to evaluate the performance of the two-gas TLAS sensor system. The relation between the second harmonic amplitudes (2f) and gas concentrations was obtained for the two gasses by means of calibration. Both the detection error and the limit of detection (LoD) were determined experimentally. The sensor system will be useful in industrial trace gas monitoring due to its use of a low-loss optical fiber and an open-reflective gas-sensing probe. Authors: Qixin He, Chuantao Zheng, Huifang Liu, Yiding Wang, Frank K. Tittel.

Absorption-Induced Image Resolution Enhancement in Scattering Media
Mehbuba Tanzid
Faculty Advisor: Naomi J. Halas
Highly scattering media pose significant challenges for many optical imaging applications due to the loss of information inherent to the scattering process. Absorption can also result in significant degradation of image quality. However, absorption can actually improve the resolution of images transmitted through scattering media in certain cases. Here we study how the presence of absorption can enhance the quality of an image transmitted through a scattering medium, by investigating the dependence of this enhancement on the medium’s scattering properties. We find that absorption-induced image resolution enhancement is substantially larger for media consisting of isotropic scatterers (e.g., dielectric nanoparticles) than for strongly forward-scattering media (e.g. biological tissue). This work leads to a broader understanding, and ultimately control, of the optical properties of strongly absorbing, scattering media. It has consequences for applications in imaging through media that exhibits both scattering and absorption, such as biological tissue, as well as for engineering media for optimal imaging and image transmission, or communication. Authors: Mehbuba Tanzid, Nathaniel J. Hogan, Ali Sobhani, Hossein Robatjazi, Adithya K. Pediredla, Adam Samaniego, Ashok Veeraraghavan and Naomi J. Halas.

Al−Pd Nanodisk Heterodimers as Antenna−Reactor Photocatalysts
Chao Zhang
Faculty Advisor: Naomi J. Halas
Photocatalysis uses light energy to drive chemical reactions. Conventional industrial catalysts are made of transition metal nanoparticles that interact only weakly with light, while metals such as Au, Ag, and Al that support surface plasmons interact strongly with light but are poor catalysts. By combining plasmonic and catalytic metal nanoparticles, the plasmonic “antenna” can couple light into the catalytic “reactor”. This interaction induces an optical polarization in the reactor nanoparticle, forcing a plasmonic response. When this “forced plasmon” decays it can generate hot carriers, converting the catalyst into a photocatalyst. Here we show that precisely oriented, strongly coupled Al−Pd nanodisk heterodimers fabricated using nanoscale lithography can function as directional antenna−reactor photocatalyst complexes. The light-induced hydrogen dissociation rate on these structures is strongly dependent upon the polarization angle of the incident light with respect to the orientation of the antenna−reactor pair. Their high degree of structural precision allows us to microscopically quantify the photocatalytic activity per heterostructure, providing precise photocatalytic quantum efficiencies. This is the first example of precisely designed heterometallic nanostructure complexes for plasmon-enabled photocatalysis and paves the way for high-efficiency plasmonic photocatalysts by modular design. Authors: Chao Zhang, Hangqi Zhao, Linan Zhou, Andrea E. Schlather, Liangliang Dong, Michael J. McClain, Dayne F. Swearer, Peter Nordlander, Naomi J. Halas

An Injection-Locked Picosecond Pulse Receiver for Wireless Clock Synchronization
Babak Jamali
Faculty Advisor: Aydin Babakhani
A picosecond pulse receiver is presented which is based on a three-stage divide-by-8 injection-locked frequency divider. The receiver operates for pulses with center frequency of 77 GHz and locks its output to the 9.6-GHz repetition rate with an effective locking range of 142 MHz. This receiver is used to demonstrate wireless clock synchronization with a 0.29ps RMS timing jitter and indicates an estimated sensitivity of −65.5 dBm in detecting picosecond pulses. Authors: Babak Jamali, Aydin Babakhani.

Analyzing Language Connectivity Networks during Articulation from Human ECoG data using Mutual Information in Frequency
Sudha Yellapantula
Faculty Advisor: Behnaam Aazhang
We humans have a remarkable ability to communicate through language by learning its underlying structure and grammar. Unfortunately, millions of people suffer from language disorders due to strokes and other brain injuries. Developing fine-grained connectivity maps of the human language system will pave the way for remediating these disorders. In this project, we are analyzing the connectivity between inferior frontal gyrus and pre frontal cortex (broadly related to Broca’s region and motor cortex), during articulation in an object naming task. To calculate connectivity between brain regions, we use human ECoG data obtained from cortical recording electrodes from patients undergoing epileptic surgeries, who undergo a multitude of language tests. This data is used for developing a platform for modeling language systems using information theoretic connectivity metrics like Mutual Information (MI). MI has been shown to be particularly effective in modeling nonlinear relations underlying the data. Our methods are data-driven, non parametric and by using a k-nearest neighbors method for estimating entropies, they even work well in higher dimensions. We are analyzing changes in MI in different frequency bands, to understand the dynamics of the brain networks while performing this task. Authors: Sudha Yellapantula, Nitin Tandon, Behnaam Aazhang.

A Power Harvesting System for Scalable Wireless Neural Recording Devices with an On-chip Antenna
Hamed Rahmani
Faculty Advisor: Aydin Babakhani
In this work, we demonstrate the first fully on-chip power harvesting system. The system receives RF waves at 3GHz through a wireless link that attenuates the transmitted power from the external source by 27 dBm. The received sinusoidal waves are rectified and energy is stored over a 1.2 nF on-chip capacitor. A power management unit divides the operation of the system into two phases with an average current consumption of 10nA. The system is capable of delivering up to 1mW to an external load when the power management unit enables the voltage regulator of the system. An array of four low noise amplifiers is also implemented on the same silicon chip that can amplifies EEG signals by 40 dB. The system is fabricated in 180nm SOI CMOS technology occupying 1.6×1.6mm2 including an on-chip loop antenna. Authors: Hamed Rahmani, Aydin Babakhani.

Broadband Beamforming of THz Pulses with Single-Chip Arrays in Silicon
Mahdi Assefzadeh
Faculty Advisor: Aydin Babakhani
This poster presents a broadband THz frequency-comb spectroscopic imager based on a fully-integrated 4×2 picosecond Direct Digital-to-Impulse (D2I) radiating array. By employing a novel trigger-based beamforming architecture, the chip performs coherent spatial combining of broadband radiated pulses achieving an SNR>1 BW of 1.03THz (at the receiver) with a pulse peak EIRP of 30dBm. Time-domain radiation is characterized using a fsec-laser-based THz sampler and a pulse width of 5.4ps is measured. Spectroscopic imaging of metal, plastic, and cellulose capsules (empty and filled) are demonstrated. This chip achieves signal generation with an available full-spectrum of 0.03-1.03THz. The 8-element single-chip array is fabricated in a 90nm SiGe BiCMOS process. Authors: Mahdi Assefzadeh, Aydin Babakhani.

Coherent Inverse Scattering via Transmission Matrices: Efficient Phase Retrieval Algorithms and a Public Dataset
Sudarshan Nagesh, Manoj Kumar Sharma
Faculty Advisor: Ashok Veeraraghavan
Transmission matrices describe the input-output relationship of a complex wavefront as it passes through/reflects off a multiple scattering medium, such as frosted glass or a painted wall. Knowing a medium’s transmission matrix allows one to image through the medium, or even use it as a lens. The double phase retrieval method is a recently proposed technique to learn a medium’s transmission matrix that avoids difficult-to-capture interferometric measurements. Unfortunately, to perform high resolution imaging existing double phase retrieval methods require (1) a large number of measurements and (2) an unreasonable amount of computations. In this work we focus on the latter of these two problems and reduce computation times with two distinct methods: First, we develop a new phase retrieval algorithm that is significantly faster than existing algorithms, especially when used with an amplitude-only spatial light modulator (SLM). Second, we calibrate the system using a phase-only SLM, rather than an amplitude-only SLM which was used in previous double phase retrieval experiments. This seemingly trivial change allows us to use a far faster class of phase retrieval algorithms. As a result of these advances, we achieve a 100 reduction in computation times, thereby allowing us to image through scattering media at state-of-the-art resolutions. In addition to these advances, we also release the first publicly available transmission matrix dataset. This contribution will allow phase retrieval researchers to apply their algorithms to real data. Of particular interest to this community, our measurement vectors are naturally i.i.d. subgaussian; no coded diffraction pattern is required. Authors: Chris Metzler, Manoj Kumar Sharma, Sudarshan Nagesh, Richard Baranuik, Oliver Cossairt, Ashok Veeraraghavan.

Directional Training for FDD Massive MIMO
Xing Zhang
Faculty Advisor: Ashutosh Sabharwal
In this work, we propose directional training for FDD massive MIMO systems, to reduce the large downlink CSI acquisition overhead. Directional training first leverages the AoA/AoD reciprocity between uplink and downlink to locate the AoD set of downlink channel utilizing uplink CSI and then trains the downlink channel using the AoD set only. We conduct extensive channel measurement employing a 64-antenna base station to evaluate the downlink beamforming performance of directional training. The results show that in the perfect CSI case, directional training performs close to full training in the line-of-sight scenarios and leads to about 17% achievable rate loss in the non-line-of-sight scenarios when serving two mobiles. In contrast, for the imperfect CSI case, directional training outperforms full training by 155% in the line-of-sight scenarios and 100% in the non-line-of-sight scenarios in terms of spectral efficiency when serving two mobiles. Hence, directional training is a promising scheme for FDD massive MIMO systems to obtain downlink CSI at the base station. Authors: Xing Zhang, Ashutosh Sabharwal.

Distributed Signal Processing for Large-scale Multiple-antenna Systems
Kaipeng Li
Faculty Advisor: Joseph Cavallaro
Achieving high spectral efficiency in realistic massive multi-user (MU) multiple-input-multiple-output (MIMO) wireless systems requires computationally complex algorithms for data detection in the uplink (users transmit to base station) and beamforming in the downlink (base station transmits to users). Most existing algorithms are designed to be executed on centralized computing hardware at the base station (BS), which both 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. This work proposes a novel distributed 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 distributed data detection and beamforming algorithms that only access local channel-state information and require low communication bandwidth among the clusters. We study the associated trade-offs between error-rate performance, computational complexity, and interconnect bandwidth, and we demonstrate the scalability of our solutions for massive MU-MIMO systems with thousands of BS antennas using reference implementations on a graphic processing unit (GPU) cluster. Authors: Kaipeng Li, Rishi Sharan, Yujun Chen, Tom Goldstein, Joseph R. Cavallaro, Christoph Studer.

Downlink Transmissions for mmWave MIMO Transmitters with Reconfigurable Antenna Arrays
Shi Su
Faculty Advisor: Behnaam Aazhang
Millimeter-wave communication systems equip large antenna arrays which enable directional beamforming to combat pathloss. However, the hardware limitations in digital beamforming devices make analog processing of signals in RF domain, referred to analog beamforming, more attractive than the traditional digital beamforming. We develop a reconfigurable antenna array structure with low hardware complexity, as well as the corresponding analog beamforming strategy. Transmitters with the proposed structure and strategy can reach a throughput close to the upper bound of the system. The proposed system can also achieve a considerable performance increase of downlink transmissions to multiple users than conventional antenna sub-array methods. Authors: Shi Su, Behnaam Aazhang.

Electrochromic Devices Based on Molecular Plasmonics
Adam Lauchner, Grant J. Stec
Faculty Advisor: Naomi Halas
Polycyclic aromatic hydrocarbon (PAH) molecules, the hydrogen-terminated, subnanometer scale version of graphene, support plasmon resonances with the addition or removal of a single electron. Typically colorless when neutral, they are transformed into vivid optical absorbers in either their positively or negatively charged states. Here we demonstrate a low-voltage, multistate electrochromic device based on PAH plasmon resonances that can be reversibly switched between nearly colorless (0 V), olive (+4 V), and royal blue (-3.5 V). The device exhibits highly efficient color change compared to electrochromic polymers and metal oxides, lower power consumption than liquid crystals, and is shown to reversibly switch for at least 100 cycles. We also demonstrate the additive property of molecular plasmon resonances in a single-layer device to display a reversible, transparent-to-black device. This work illuminates the potential of PAH molecular plasmonics for the development of color displays and large-area color-changing applications due to their processability and ultralow power consumption. Authors: Adam Lauchner, Grant J. Stec, Yao Cui, Peter Nordlander, and Naomi J. Halas.

Essential Nonlinear Properties in Neural Decoding
Qianli Yang
Faculty Advisor: Xaq Pitkow
To decode task-relevant information from sensory observations, the brain must eliminate nuisance variables that affect those observations. For natural tasks, this generally requires nonlinear computation. Here we contribute new concepts and methods to characterize behaviorally relevant nonlinear computation downstream of recorded neurons. Linear decoding weights can be inferred from correlations between neurons and behavior. However, these weights do not adequately describe the neural code when, due to nuisance variation, mean neural responses are poorly tuned to the task while higher-order statistics of neural responses are well tuned. The task-relevant stimulus information can then be extracted only by nonlinear operations. For example, detecting an object boundary in an image requires contrast invariance: an edge appears when the foreground object is darker or lighter than the background, yet any linear function will exhibit opposite responses in these two cases. We generalize past weight-inference methods to determine the brain’s nonlinear neural computations from joint higher-order statistics of neural activity and behavioral choices in perceptual tasks. This method is based on a new statistical measure we call nonlinear choice correlation, defined as the correlation coefficient between behavioral choices and nonlinear functions of measured neural responses. Importantly, the exact neural transformations may not be uniquely identifiable, since many neural nonlinearities can generate the same behavioral output. This is expected when sensory signals are expanded into a larger cortical response space, creating a redundant code. We exploit this redundancy to define a new concept of equivalence classes for neural transformations. We then demonstrate how to quantify essential properties of these equivalence classes, and provide simulations that show how these properties can be extracted using neural data from behaving animals. Finally, we explain the functional importance of these nonlinearities in specific perceptual tasks. Authors: Qianli Yang, Xaq Pitkow.

Estimating Gameplay Engagement in User-contributed Videos via Motion Signature
Xu Chen
Faculty Advisor: Ashutosh Sabharwal
Modeling task engagement of individuals is conducive to a broad array of applications, such as the optimization of cutting-edge interactive systems and the prevention of cognitive disorders. However, existing methods that exploit the physiological modalities (e.g., EEG signals, pupil dilation, and heart rate variability) to approximate task engagement are either dependent on intrusive and costly devices, or robust only in lab settings. In this work, we focus on an important interactive task – gameplay, and develop a non-intrusive framework Engagementometer to estimate the gameplay engagement in user-contributed videos captured by off-the-shelf webcams. We build a database that contains over 1,000 picture-in-picture gameplay video excerpts from YouTube with the labeled engagement levels of more than 10 users. Based on this database, Engagementometer extracts multiple motion signatures associated with task engagement (e.g., eye motion, head motion, etc.) from users, and then combines these motion tags training a multimodal classifier that automatically estimates gameplay engagement. We assess this model, by carrying out interview-based field study using sample videos in the database. Experimenters are asked to rank the engagement levels (i.e., high to low) of users in the selected clips, which are then fed into our classifier to generate a series of confidence scores. The rankings of videos given by experimenters will be compared with the rankings of confidence scores, for determining agreement. We demonstrate that Engagementometer can estimate engagement with decent accuracy, and that this estimation model can be generalized across different users and different games. Authors: Xu Chen, Ashutosh Sabharwal

Estimating Heart Rate variability by weighted frequency demodulation of imaging photoplethysmogram
Amruta Pai
Faculty Advisor: Ashutosh Sabharwal
As the heart rate is a non-stationary signal, the beat to beat interval (time period of the cardiac cycle) varies. The variation is majorly due the parasympathetic and sympathetic nervous system. This heart rate variability can be extracted and is indicative of various cardiac diseases present or might occur in an individual. It can also be an indicator of the stress level and mental health of an individual. Currently heart rate variability is accurately measured using contact devices which are inconvenient and can cause discomfort when used for long periods of time. Recently research in the field of non-contact Photoplethysmography (PPG) has made vital sign measurement using just the video recording of a person’s face possible. The current signal processing method of extracting heart rate variability using peak detection works for contact based systems to a certain extent but fails in case of non-contact Imaging PPG systems as the latter has low signal quality. The current method is also not robust to motion artifacts. I will present a new method for extracting the heart rate variability from the Imaging PPG signals by modelling them as spatially diverse frequency modulated signals. The results obtained will be validated using a pulse oximeter which is the gold standard device currently used. Authors: Amruta Pai, Ashutosh Sabharwal.

Fast HARQ: Low Latency (H)ARQ for Massive MIMO
Xu Du
Faculty Advisor: Ashutosh Sabharwal
Cellular systems employ Hybrid Automatic Repeat Request (HARQ) to ensure reliable packet delivery. Mobile stations in current LTE systems experience at least 4ms delay to receive negative acknowledgment or re-transmitting a failed packet, which fails the 1 ms target latency requirement for 5G. In this work, we propose a low-latency HARQ control channel design by leveraging the spatial degree of freedom of massive MIMO base station. Our analytical and simulation results show that our strategy achieves sub-ms HARQ retransmission latency and outperforms the leading 5G URLLC design for general traffic in extended coverage range. Authors: Xu Du, Ashutosh Sabharwal.

FlatCam: Computational Lensless Camera
Vivek Boominathan, Jesse K. Adams
Faculty Advisors: Ashok Veeraraghavan, Jacob Robinson, Richard Baraniuk
The basic design of a camera has remained unchanged for centuries. To acquire an image, light from the scene under view is focused onto a photosensitive surface using a lens. Unfortunately, lenses also introduce a number of limitations. Majorly, while image sensors are typically thin, cameras end up being thick due to the lens complexity and the large distance required between the lens and sensor to achieve focus. In this poster/demo, we demonstrate a thin form-factor lensless camera that is constructed by attaching a coded mask directly onto a CMOS/CCD sensor. The lack of focusing lens is then compensated using computational algorithms to reconstruct a clear and recognizable image. Authors: Vivek Boominathan, Jesse K. Adams, M. Salman Asif, Ben Avants, Jacob T. Robinson, Richard G. Baraniuk, Aswin C. Sankaranarayanan, Ashok Veeraraghavan.

Hardware Transactional Non-Volatile Memory
Ellis Giles
Faculty Advisor: Peter Varman
Emerging byte-addressable non-volatile memory, called Storage Class Memory, combined with the expanding use of parallel programming and concurrency control mechanisms will need to address not only the consistency of concurrent programming, but also the persistence consistency of the non-volatile, persistent storage to SCM. This research presents both software only approaches using Transactional Locking techniques and also Hardware-Software techniques leveraging off Intel’s Restricted Transactional Memory. The techniques are implemented in a library called TSCM, Transactional Storage Class Memory. We evaluated TSCM using both micro-benchmarks and the STAMP benchmark suite and found good speedup can be achieved while atomically persisting concurrent transactions to SCM safely. As Storage Class Memories enter the marketplace and are combined with High Performance Computing, programming challenges to couple high concurrency transactions with atomic persistence to SCM can be handled using techniques presented in this paper that allow for continued high performance. Authors: Ellis Giles, Dr. Kshitij A. Doshi (Intel), Dr. Peter Varman.

High Frequency Oscillation detection in epileptic data
Negar Erfanian
Faculty Advisor: Behnaam Aazhang
BiMTLE is a disease that causes a high percentage of all epilepsies in adults while remaining resistant to drug therapies. Moreover, surgical methods such as resection or ablation are not reasonable treatments for this type of disease either, since they might degrade memory in patients that may already suffer from memory loss due to their disease. In light of these approaches’ shortcomings, I will investigate novel neural modulation as a remedy for the aforementioned symptoms via three stages: I will begin by systematically analyzing the characteristics of an epilepsy network as well as a memory network, then will determine the best equipment and tools for characterization of the mentioned networks, and finally will propose a method to simultaneously reduce seizures and enhance memory performance in BiMTLE patients. Considering that we will work on this project in 5 years, this semester I will focus on analyzing the data collected via precisely placed electrodes and other implanted recording devices from patients’ brains, using MATLAB. The characteristics of the epileptic data recorded from these patients when facing seizures will be interfered by analyzing the changes occur in the following states: 1- pre-seizure state 2- during seizures 3-between seizures 4-post-seuzires. Both ictal and inter-ictal time windows will be collected to better characterize the epileptic network. Authors: Dr. Behnaam Aazhang, Dr. Nitin Tandon, Negar Erfanian.

Hippocampal cells representation during mental navigation
Sibo Gao
Faculty Advisor: Caleb Kemere
Can you estimate coherent translation of a person’s spatial environmental behavior merely by measuring and analyzing one’s neural activity? We can easily navigate mentally through memorized routes using our own mental maps. However, the effect of cognitive variables such as attention on what hippocampal cells look like remains unknown. In this poster, we investigate the representation of rodent hippocampal place cells under various simulated cognitive variables through neural decoding model to infer spatial behavior. Authors: Sibo Gao, Caleb Kemere

Inferring Inference
Rajkumar Vasudeva Raju
Faculty Advisor: Xaq Pitkow
How can we determine what the brain computes? Here we describe a framework to infer canonical structure in this computation. It is based on the theory that the brain uses nonlinear computation by redundant, recurrently connected population codes to perform approximate probabilistic inference. Specifically, these computations implement a message-passing algorithm operating on a probabilistic graphical model whose interactions are encoded by overlapping probabilistic population codes. We describe an analysis method that aims to identify this algorithm from complex neural data elicited during perceptual inference tasks. To recover the message-passing algorithm from neural recordings, we must simultaneously find (1) the representation of task-relevant variables, (2) interactions between the decoded variables that define the brain’s internal model of the world, and (3) the global parameters that define the message-passing inference algorithm. We hypothesize that the global parameters are canonical – that is, common to all parts of the graphical model regardless of interaction strength – so that they generalize to new graphical models. We have applied this analysis method to artificial neural recordings generated by a simple model brain that uses an advanced mean field method to perform approximate inference. We formulate a method of learning the inference algorithm from the given neural dynamics as an optimization problem, and successfully recover our model inference algorithm. This success encourages us to apply these methods to more sophisticated brain models, trained neural networks, and, eventually, to large-scale neural recordings to uncover canonical properties of the brain’s distributed nonlinear inferential computations. Authors: Rajkumar Vasudeva Raju, Xaq Pitkow.

Inferring Spectral and Spatiotemporal Dependencies from Data and its Application to Epilepsy
Rakesh Malladi
Faculty Advisor: Behnaam Aazhang
A fundamental problem in many science and engineering disciplines is inferring the characteristics of a physical or biological system from the dependencies in data recorded from the system. The dependencies in data, particularly in case of signals recorded from brain, are commonly believed to be nonlinear and the underlying model is often unknown. This thesis focusses on developing novel information-theoretic approaches to detect and quantify spectral and spatiotemporal dependencies from data in a data-driven manner and applies them to electrocorticographic (ECoG) recordings from epilepsy patients to unravel epileptic brain networks. Frequency components in a signal or between two signals, not necessarily at the same frequency, are spectrally dependent if they are not statistically independent. Two signals are temporally dependent if the past measurements at one decrease the uncertainty in predicting the other. First, we define a novel metric, mutual information in frequency, to detect spectral dependency and quantify it using a data-driven estimator. We then develop a data-driven estimator of mutual information between dependent data using mutual information in frequency. Next, we develop a model-based and a data-driven estimator of directed information to detect and quantify the temporal dependencies in data. Finally, we apply the proposed metrics to ECoG recordings from epilepsy patients to identify seizure onset zone (SOZ), to learn the spatiotemporal characteristics of seizures and to infer the cross-frequency coupling in SOZ. We observe that seizure onset zone drives the rest of brain to a seizure during pre-seizure and seizure periods, while it acts as a sink during post-seizure periods. In addition, high frequency coupling increases during seizures within an ECoG channel and between channels in the same anatomical region in SOZ, but not between different regions in SOZ. This suggests different anatomical regions in the SOZ are independently driving the seizure activity and any treatment should potentially target these regions simultaneously. Going forward, the dependencies unraveled by the proposed metrics should be further analyzed to optimize the parameters of closed-loop electrical stimulation based treatments for epilepsy. Authors: Rakesh Malladi, Don H Johnson, Giridhar P Kalamangalam, Nitin Tandon.

Low Latency, Closed-Loop, Real-Time System for Detecting Hippocampal Sharp-Wave Ripples
Shayok Dutta
Faculty Advisor: Caleb Kemere
Reverse engineering the brain in order to establish causal relationships often begins with lesioning particular brain regions which may result in a general cognitive impairment. In order to identify the particular neural substrates responsible for specific behavioral and cognitive tasks, modulation of specific patterns of brain activity is required. Here, we devolop and evaluate a real-time system that detects hippocampal sharp-wave ripples — transient neural oscillations present in electrophysiological recordings from the hippocampus that are associated with learning and memory consolidation. Our results indicate that we have a true positive rate of event detections of greater than 95% while having less than 5 false detections per minute with a closed-loop latency of 2 ms upon event detection. Authors: Shayok Dutta, Caleb Kemere.

Measuring Spike-LFP Synchronization via Mutual Information
Joseph Young
Faculty Advisor: Behnaam Aazhang
Despite numerous years of research on skill learning, incredibly little is understood about the neuronal explanation for why skills develop with practice. This lack of understanding leaves those afflicted with learning disabilities suffering, as today’s pharmaceutical solutions induce side effects and target the symptoms instead of the cause. To develop understanding of skill learning and treat the cause of such disorders, initial research on skill learning focused on the role of individual neurons. This yielded minimal insight, motivating our hypothesis that learning is a result of specific changes at the network level. Spike-field coherence (SFC) has been used previously to study network level interactions between individual local field potentials (LFPs) and spike times. We have developed a new probabilistic tool, mutual information (MI) in the frequency domain for a continuous process and a discrete process, that analyzes such interactions with greater flexibility. For this tool, we estimate the MI between a two-dimensional frequency representation of windows of LFPs and a discrete binary random variable representing the presence or absence of a spike at the center of the aforementioned LFP windows. Simulations reveal that MI is able to capture not only single phase spike-LFP synchronization, but also captures multiphase synchronization much more robustly than SFC. Previous SFC analysis of LFPs and spike times recorded from rhesus monkeys learning a visual rotation recognition task revealed increased spike-LFP synchronization during learning. As expected, our MI results match SFC, but MI is better equipped to capture nonlinear interactions and is truly model-free. Ultimately, we will develop data-driven models of skill learning to use as blueprints for electrically stimulating learning in rhesus monkeys. Such stimulation will actually accelerate the rate of learning and allow for treatment of learning disorders without the side effects inherent to pharmaceutical solutions. Authors: Joseph Young, Rakesh Malladi, Valentin Dragoi, Behnaam Aazhang.

Miniaturized mm-Wave and THz Impulse Radiators in Silicon for High-Resolution 3D Imaging
Peiyu Chen
Faculty Advisor: Aydin Babakhani
In this poster, I’ll present three integrated chips (IC) using silicon technologies to produce picosecond mm-wave and THz impulse radiation, ranging from 1.14ps to 60ps. The proposed IC technologies deploy the power of integration to perform much more complicated functions than the traditional photocondutive antenna method. Synthetic-array-based 3D imaging has been demonstrated by using one of the chips. With the help of digital beamforming, high-resolution 3D images of both metallic objects and dielectric objects have been produced. Future directions of my research will be discussed as well. Authors: Peiyu Chen, Aydin Babakhani.

Multi Component Carrier, Sub-Band DPD and GNURadio Implementation
Chance Tarver
Faculty Advisor: Joseph Cavallaro
Digital predistortion (DPD) is an effective way of mitigating spurious emission violations without the need of a significant backoff in the transmitter, thus providing better power efficiency and network coverage. In this paper, the IM3 subband DPD, proposed earlier by the authors, is extended to more than two component carriers (CCs) through a sequential learning solution. The DPD learning is iterated over each spurious emission generated by each pair and trio of CCs. We train and apply the DPD coefficients for the intermodulation distortion (IMD) products until a satisfactory performance is achieved. The algorithm is tested in simulations using MATLAB and in a novel, real-time implementation on a CPU via a software version of the algorithm using GNURadio. Authors: Chance Tarver, Mahmoud Abdelaziz, Joseph R. Cavallaro.

Nanophotonics-enabled Solar Membrane Distillation for Off-grid Water Purification
Pratiksha Dongare
Faculty Advisor: Naomi J. Halas
Membrane distillation is being used for water desalination for more than 30 years. In conventional membrane distillation system, vapor pressure difference on the two sides of a porous hydrophobic membrane drives water vapor across the membrane, thus separating salts from water. This vapor pressure difference is either created by flowing water with difference in temperature on two sides of the membrane or using vacuum pumps. This process is energy intensive as it heats up bulk water to maintain temperature difference across the membrane. This results in significant increase in the size of the system and limits the scalability of the membrane distillation module, making it difficult to be used in remote areas. In this project, we have created a sustainable and scalable nanoparticle-enabled solar membrane distillation (NESMD) system, which uses conventional polyvinylidene fluoride (PVDF) membranes electrospun with broadband absorbing carbon black nanoparticles that absorb more than 90% of the incident solar radiation. In sunlight, this locally heats up the water on one side of the membrane creating vapor pressure difference driving water through it even when there is no difference in the temperature of the bulk water on the two sides of the membrane. Diffuse reflectance measurements and Monte Carlo simulations are used to obtain the optical properties of the electrospun membrane. Through outside solar experiments, we have confirmed the effectiveness of the lab scale NESMD prototype. This system is portable and scalable as the temperature difference is created locally on the membrane. Authors: Pratiksha D. Dongare, Alessandro Alabastri, Seth Pedersen, Katherine Zodrow, Nathaniel J. Hogan, Oara Neumann, Jinjian Wu, Tianxiao Wang, Qilin Li, Peter Nordlander, Naomi J. Halas.

On the Impact of Blockage on the Throughput of Multi-tier Millimeter-Wave Networks
Shuqiao Jia
Faculty Advisor: Behnaam Aazhang
We characterize upper bounds on the throughput capacity of a multi-tier millimeter wave (mmWave) network with diverse blockage probability, p(n), scaling scenarios. Communication links in the network are divided into three communication tiers sharing a single mmWave frequency. The bottom tier considers transmissions between n terminals and M(n) access points (APs). The top tier includes links between the APs and the backhaul. The bottom tier and the top tier are connected by the AP tier, referring to the communications between APs. We derive the optimal APs scaling orders M(n) for the multi-tier mmWave network under various blockage scaling scenarios p(n). To show the effectiveness of the multi-tier mmWave network, we benchmark its performance against the general mmWave network topology. By comparing the throughput scaling with and without blockage, we show that under certain blockage probability scalings, the performance degradation of the multi-tier mmWave network can be avoided by choosing the appropriate M(n). Authors: Shuqiao Jia; David Ramirez; Lei Huang; Yi Wang; Behnaam Aazhang.

Ouroboros Wear-leveling: A Two-level Hierarchical Wear-leveling Model for NVRAM
Qingyue Liu
Faculty Advisor: Peter Varman
Emerging non-volatile RAM (NVRAM) technologies have a limit on the number of writes that can be made to any cell, similar to the erasure limits in NAND Flash. This motivates the need for wear-leveling techniques to distribute the writes evenly among the cells. We propose a novel hierarchical wear-leveling model called Ouroboros Wear-leveling. Ouroboros uses a two-level strategy whereby frequent low-cost intra-region wear-leveling is combined with predictive inter-region wear-leveling at a larger time interval and spatial granularity. Ouroboros is a hybrid migration scheme that exploits correct demand predictions in making better wear-leveling decisions, while using randomization to avoid wear-leveling attacks by deterministic access patterns. Several experiments are performed on both specially crafted memory traces and two block-level storage traces generated by Microsoft and FIU. The results show that Ouroboros Wear-leveling can successfully distribute writes smoothly across the whole NVRAM with very small space and time overheads for a 512 GB memory. Authors: Qingyue Liu, Peter Varman.

Pace to the beat of your own heart! (Smarter Pacemakers)
Paz Zait-Givon
Faculty Advisor: Behnaam Aazhang
Your heart works best when all parts of it are in communication with each other and everyone waits their turn to do their part. This poster will explain why it is important that the whole heart pumps in order synchronously as well as why it might fail to do that and how we hope to fix it.

PPGSecure: Biometric Presentation Attack Detection Using Photopletysmograms
Ewa Nowara
Faculty Advisors: Ashok Veeraraghavan, Ashutosh Sabharwal
Authentication of users by exploiting face as a biometric is gaining widespread traction due to recent advances in face detection and recognition algorithms. While face recognition has made rapid advances in its performance, such face-based authentication systems remain vulnerable to biometric presentation attacks. Biometric presentation attacks are varied and the most common attacks include 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 a 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 the art presentation attack detection methods. Authors: Ewa Nowara, Ashutosh Sabharwal, Ashok Veeraraghavan.

PulseCam: Design, development, and characterization of a new multi-sensor blood perfusion imaging modality
Mayank Kumar
Faculty Advisors: Ashok Veeraraghavan, Ashutosh Sabharwal
Blood perfusion is the flow of oxygen-rich blood to the end-organs and tissues through the blood vessels. Peripheral arterial disease (PAD) causes narrowing of the blood vessels that supply blood to the periphery (e.g. hand and leg) leading to reduced blood perfusion. Around 12%-20% of the people above 60 years of age are affected by PAD, and the risk is significantly higher among diabetic patients and smokers. PAD is mostly asymptomatic, and if left untreated, can lead to complications such as foot ulcers, and may eventually lead to foot amputations as well. In this thesis, I will present the design, development, and characterization of a new multi-sensor blood perfusion imaging modality, named PulseCam, to measure peripheral blood perfusion by combining a video camera and a pulse-oximeter. Earlier work in this field has demonstrated the feasibility of camera-only blood perfusion imaging, but most of them produce low-resolution peripheral perfusion maps due to very low signal-to-noise ratio (< -16 dB per pixel) of the blood perfusion-related signal. In my preliminary work, I demonstrated the feasibility of high-resolution blood perfusion imaging by utilizing a contact pulse oximeter as a source of reference blood volume waveform. I showed that having access to a reliable blood volume waveform improves the accuracy of perfusion estimate and leads to reliable estimation of optical flow vectors which are needed to compensate movement of the skin surface. I propose to characterize the performance of PulseCam for people having different skin tones, under varying lighting conditions, and in the presence of motion. Specifically, I will propose new experiment designs to measure the sensitivity of PulseCam system, i.e. to quantify how small a change in pulsatile blood perfusion can be detected using PulseCam, and to identify the tradeoff between sensitivity and spatial and temporal resolution. Such characterization will be essential to understanding the utility of PulseCam in diagnosing peripheral arterial diseases under varying operational scenarios (skin tone, lighting, and motion), and will provide system specification and benchmark for future algorithm and device development. Authors: Mayank Kumar, Ashok Veeraraghavan, Ashutosh Sabharwal.

Reconstructing Rooms Using Photon Echoes: A Plane based model and reconstruction algorithm for looking around the corner
Adithya Kumar Pediredla
Faculty Advisor: Ashok Veeraraghavan
Can we reconstruct the entire internal shape of a room if all we can directly observe is a small portion of one internal wall, presumably through a window in the room? While conventional wisdom may indicate that this is not possible, motivated by recent work on `looking around corners’, we show that one can exploit light echoes to reconstruct the internal shape of hidden rooms. Existing techniques for looking around the corner using transient images model the hidden volume using voxels and try to explain the captured transient response as the sum of the transient responses obtained from individual voxels. Such a technique inherently suffers from challenges with regards to low signal to background ratios (SBR) and has difficulty scaling to larger volumes. In contrast, in this paper, we argue for using a plane-based model for the hidden surfaces. We demonstrate that such a plane-based model results in much higher SBR while simultaneously being amenable to larger spatial scales. We build an experimental prototype composed of a pulsed laser source and a single-photon avalanche detector (SPAD) that can achieve a time resolution of about 30ps and demonstrate high-fidelity reconstructions both of individual planes in a hidden volume and for reconstructing entire polygonal rooms composed of multiple planar walls. Authors: Adithya Kumar Pediredla, Mauro Buttafava, Alberto Tosi, Oliver Cossairt, Ashok Veeraraghavan.

Reducing Latency in LTE Random Access Channel (RACH) by the Size-of-Texas!
Fatima Ahsan
Faculty Advisor: Ashutosh Sabharwal
At any random time, a user may pick up her phone to get internet or call service from the cellular Base Station (BS), hence, called Random Access (RA) process in wireless communication literature. To handle concurrent RA user requests, current LTE standard specifies 64 orthogonal codes for accessing the BS. The users randomly pick up a code from a pool of 64 and send association request to the BS for data transfer during LTE-RACH time-frequency slot. If more than one users pick up the same code, it causes collision at the BS and necessitates further steps for collision resolution that add to the latency of the users communicating with the BS. In future, the user demand is going to increase considerably and the probability of more than one users picking up the same code will increase. To deal with this issue, we leverage the fact that future cellular networks will consist of Massive MIMO antenna arrays. Hence, the signals that are not separable in time and frequency domain, can be separated in spatial domain via their disparate Angle-of-Arrivals (AoA) at the BS. This will tantamount to providing more number of codes within the same time-frequency LTE-RACH resource. Thereby, decreasing the number of collisions and, hence, the latency in connecting the users to the BS. Authors: Fatima Ahsan, Ashutosh Sabharwal

Republic: Revisiting Data Multicast in Hybrid Data Centers
Xiaoye Steven Sun
Faculty Advisor: T.S. Eugene Ng
In recent years, there has been a surge in interest in building hybrid data center network prototypes using various emerging technologies, including optical circuit switch, free-space optics, millimeter wave wireless, etc. These physical layer innovations have fundamentally changed the communication capability of data center interconnections, especially the capability for one-to-many group communication. This paper revisits the issue of data multicast under the scenario of multicast-featured hybrid data centers. We propose Republic, a complete system providing a high-performance and flexible data multicast service in hybrid data centers. Applications use a unified Republic API to transfer multicast data. We have implemented Republic and deployed Republic in our hybrid data center testbed. The evaluation shows that Republic can improve data multicast in Apache Spark machine learning applications by up to 8.2$\times$. Authors: Xiaoye Steven Sun, Yiting Xia, Simbarashe Dzinamarira, Xin Sunny Huang, Dingming Wu, T. S. Eugene Ng.

Scheduling Schemes for Multi-tier mmWave Networks with Blockage
Boqiang Fan
Faculty Advisor: Behnaam Aazhang
Due to abundant frequency resources, millimeter wave (mmWave) frequencies have drawn much attention as a solution to bandwidth scarcity. However, many characteristics of mmWave transmissions, such as blockage and reduced coverage, make schedulers designed for conventional wireless communications very inefficient for use in mmWave networks. We consider a multi-tier mmWave network composed of an AP to Backhaul tier, an AP to AP tier and an end-user to AP tier, which allows for relaying around blockages and serves as a model for a possible network deployment. A scheduling scheme with polynomial complexity is developed for the multi-tier mmWave network, limiting the number of hops so that the delay and the computational complexity are controlled. Three benchmark schedulers and a tight upper bound are presented for performance comparison. Simulation results show that the performance of our scheduling scheme approaches the upper bound even with blockages. In networks with imperfect beamforming, i.e. not all power are focused on the receiver, our algorithm also significantly outperforms the benchmark. Authors: Boqiang Fan, Behnaam Aazhang.

Scoring Sequences of Hippocampal Activity using Hidden Markov Models
Etienne Ackermann
Faculty Advisor: Caleb Kemere
We propose a novel sequence score which we expect will enable us to detect and analyze hippocampal replay more effectively than current approaches. In particular, we show how hidden Markov models (HMMs) can be used to model and analyze sequences of neural activity, and how the resulting joint probability of an observation sequence and an underlying sequence of states naturally lead to the development of a two component sequence score in which the sequential and contextual information are decoupled. We also show how this score can discriminate between true and shuffled sequences of hippocampal neural activity. Authors: Etienne Ackermann, Caleb Kemere.

Semi-supervised Learning and Network Pruning with the Deep Rendering Mixture Model
Tan Nguyen, Wanjia Liu, Ethan Perez
Faculty Advisors: Richard Baraniuk, Ankit Patel
Semi-supervised learning algorithms reduce the high cost of acquiring labeled training data by using both labeled and unlabeled data during learning. Deep Convolutional Networks (DCNs), which have achieved great success in supervised tasks, have recently made progress in semi-supervised learning. However, since a probabilistic generative model underlying DCNs is missing, there is no principled way to enable DCNs to learn from unlabeled data. In this work, we develop a new semi-supervised learning algorithm based on the recently developed Deep Rendering Mixture Model (DRMM), a probabilistic generative model whose inference algorithm corresponds to the computations in a DCN. We derive the Expectation Maximization algorithm for the DRMM and use it to learn from both labeled and unlabeled data. We employ variational inference and a novel non-negativity constraint inspired by the DRMM theory to dramatically improve performance. We also develop a new synapse and neural pruning method that works with our semi-supervised learning algorithm to reduce the model complexity and enable efficient inference and learning. Our DRMM-based semi-supervised learning algorithm achieves state-of-the-art performance on the MNIST and SVHN datasets and competitive results on CIFAR10 amongst all methods that do not use data augmentation. We show that after pruning, the model is significantly compressed while still achieving the same classification accuracy. Authors: Tan Nguyen, Wanjia Liu, Ethan Perez, Richard G. Baraniuk, Ankit Patel.

SocialSense: Construct “Physical” Social Network from Smartphone Data
Jian Cao
Faculty Advisor: Ashutosh Sabharwal
Sociability is an important factor that influences the quality of life, as well as physical and mental well-being. However, current approaches for sociability measures mainly rely on recall-based self-report. Hence, it’s time-consuming and often bias. Recently, the smartphone has shown its potential as a wearable device for context-rich sensory data accumulation. Therefore, we propose to combine smartphone audio, sensor and usage data, to quantitatively measure sociability, and to construct a “physical” social network of a certain user. The aim of this project is two-folded: first, we will derive a set of sociability metrics, that are most correlated with social interaction quality, health outcomes, etc.; second, we will develop robust algorithms to automatically estimate these sociability metrics, such as the physical social network of the user, the turn-taking behavior during conversations, the daily trajectory and lifelog, etc. By the subjective and passive sociability measure, we hope to better support aging-in-place, psychiatric disease monitoring, and mental well-being. Authors: Jian Cao, Ashutosh Sabharwal.

Spatial IHC analysis to quantify tumor-immune cell interactions and predict outcomes
Souptik Barua
Faculty Advisors: Arvind Rao, Ashok Veeraraghavan
The distribution and type of immune cells in the tumor micro-environment has been shown to dictate the progress of cancer. We design data-driven metrics that quantify the immune infiltration level of a patient from immunohistochemistry images of tumor biopsies. First, we show that our spatial metric is a robust and accurate representation of the infiltration level we see visually. Second, our metric has been shown to be an independent predictor of patient outcome for pancreatic cancer. We envisage our metric to serve as a fast and accurate diagnostic checkpoint to guide immune-based treatment post-surgery. Authors: Souptik Barua, Dr. Arvind Rao.

Spatial Map of Pulse Transit Time
Akash Kumar Maity
Faculty Advisor: Ashutosh Sabharwal
The goal of my project is to build a system to accurately estimate spatial map of Pulse Transit Time (PTT) with the help of a camera which can be used in clinical environment. Knowing PTT is important because it is a measure of arterial stiffness and thus can be used to monitor risks of cardiovascular diseases in humans. The basic approach of estimating PTT is by obtaining photoplethysmogram (PPG) signal from two locations of the body and then calculating the time delay in these two signals. Non-contact based PTT estimation overcome many physical and technical limitations encountered by contact-based methods. But, the main challenge in non-contact based methods are low signal-to-noise ratio and tracking under motion. While traditional methods records signal from two diverse locations, say proximal and distal sites of the body, my goal is to estimate a spatial map of PTT within a region of the subject, preferably the foot region. In this project, a high frames per second camera will be used to record the foot region of a subject, and then visualize propagation of blood flow in the corresponding region. This can be particularly useful in studies for better understanding of blood microcirculation in a foot region for diabetic subjects. Unlike the existing studies, it will be shown that the system works on people with different skin tones with a considerable increase in accuracy as well as spatial resolution. Authors: Dr. Ashutosh Sabharwal.

Two Cameras Are Better Than One
Ewa Norawa, Leo Meister, June Chen
Faculty Advisors: 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. Authors: Ashok Veeraraghavan, Ashutosh Sabharwal, Ewa Norawa, Leo Meister, June Chen.

Universal microbial diagnostics using random DNA probes
Amirali Aghazadeh
Faculty Advisor: Rich Baraniuk
Early identification of pathogens is essential for limiting development of therapy-resistant pathogens and mitigating infectious disease outbreaks. Most bacterial detection schemes use target-specific probes to differentiate pathogen species, creating time and cost inefficiencies in identifying newly discovered organisms. In this talk I will present a novel universal microbial diagnostics (UMD) platform to screen for microbial organisms in an infectious sample, using a small number of random DNA probes that are agnostic to the target DNA sequences. Our platform leverages the theory of sparse signal recovery (compressive sensing) to identify the composition of a microbial sample that potentially contains novel or mutant species. We validated the UMD platform in vitro using five random probes to recover 11 pathogenic bacteria. We further demonstrated in silico that UMD can be generalized to screen for common human pathogens in different taxonomy levels. UMD’s unorthodox sensing approach opens the door to more efficient and universal molecular diagnostics. Authors: Amirali Aghazadeh, Adam Y. Lin, Mona A. Sheikh, Allen L. Chen, Lisa M. Atkins, Coreen L. Johnson, Joseph F. Petrosino, Rebekah A. Drezek and Richard G. Baraniuk.

Wireless Powered Implantable Pacemaker with On-Chip Antenna
Yuxiang Sun
Faculty Advisor: Aydin Babakhani
We present a battery-less mm-sized wirelessly powered pacemaker microchip with on-chip antenna in 180nm CMOS process. The microchip harvests RF radiation from an external source in the X-band frequency, with the size of 4mm by 1mm. The in-vivo experiment is demonstrated successfully on a live pig heart. The pacemaker can be wirelessly powered with a distance of 2cm. It generates a stimulation pulse signal with a voltage magnitude of 1.3V. The wireless pacing testing was successfully demonstrated by changing the heart rhythm frequency from 1.67Hz to 2.87Hz. Authors: Yuxiang Sun, Brian Greet, David Burkland, Mathews John, Mehdi Razavi, Aydin Babakhani.

Wireless Synchronization and Spatial Combining of Widely-Spaced Mm-wave Arrays in 65nm CMOS
Charles Chen
Faculty Advisor: Aydin Babakhani
This paper presents the first Wirelessly Synchronized Multi-chip Array (WSMA) in 65-nm CMOS. The proposed architecture makes use of a central wireless signal to synchronize a mm-wave array, eliminating the need for connecting wires between the array elements. Wireless injection locking of a single chip is successfully demonstrated and a line-width of 400 Hz at a carrier frequency of 50 GHz is achieved (stability ratio of 8 ppb). In addition, a 2-element WSMA with an array aperture greater than 20 wavelengths is demonstrated using the proposed transceiver architecture. The reported transceiver includes a receiving on-chip antenna, a low-noise amplifier, an injection-locked voltage-controlled oscillator, a buffer amplifier, an in-phase/ quadrature generator, a phase shifter, a power amplifier, and a transmitting on-chip antenna. The chip is fabricated in a 65-nm CMOS process and occupies an area of 1.7 mm x 3.8 mm. This work sets the foundation for increasing the array aperture through wireless injection locking, extending traditional array systems into the high-resolution, narrow-beamwidth regime. Authors: Charles Chen, Aydin Babakhani.

Community Projects

A compact mid-infrared dual-gas CH4/C2H6 sensor using a single interband cascade laser and custom electronics (Postdoctoral Project)
Dr. Weilin Ye, Qixin He
Faculty Advisor: Frank Tittel
A compact mid-infrared (MIR) dual-gas sensor system was demonstrated for simultaneous detection of methane (CH4) and ethane (C2H6) using a single continuous-wave (CW) interband cascade laser (ICL) based on tunable laser absorption spectroscopy (TDLAS) and wavelength modulation spectroscopy (WMS). Ultracompact custom electronics were developed, including a laser current driver, a temperature controller, and a lock-in amplifier. These custom electronics reduce the size and weight of the sensor system as compared with a previous version based on commercial electronics. A multipass gas cell with an effective optical length of 54.6 m was employed to enhance the absorption signal. A 3337 nm ICL was capable of targeting a C2H6 absorption line at 2996.88 cm-1 and a CH4 line at 2999.06 cm-1. Dual-gas detection was realized by scanning both the CH4 and C2H6 absorption lines. Based on an Allan deviation analysis, the 1 σ minimum detection limit (MDL) was 17.4 ppbv for CH4 and 2.4 ppbv for C2H6 with an integration time of 4.3 s. TDLAS based sensor measurements for both indoor and outdoor mixing ratios of CH4 and C2H6 were conducted. The reported single ICL based dual-gas sensor system has the advantages of reduced size and cost without influencing the mid-infrared sensor detection sensitivity, selectivity and reliability. Authors: Weilin Ye, Qixin He, Chuantao Zheng, Frank K. Tittel, Nancy P. Sanchez, Aleksander K. Gluszek, Arkadiusz J. Hudzikowski, Minhan Lou, Lei Dong, Robert J. Griffin.

Optical Society of America
Benjamin Clark, Mehbuba Tanzid, Jian Yang, Aswathy Vijayakumaran Girija and Gururaj Naik
Come join the Rice University Optical Society of America (RU-OSA) Student Chapter which has a mission to enrich its members through professional networking, and by providing career development and technical programming. RU-OSA also raises awareness of optics and science learning through participation in events at Rice University. Through organizing youth education outreach activities such as science fairs, lab tours, and classroom demonstrations, RU-OSA creates opportunities for getting involved in optics and science education in the local community.

Rice University Engineering Professional Master’s Program
Dagmar Beck, Agustina Fernandez-Moya
The Professional Master’s degree is designed for the student who wishes to expand his or her engineering knowledge but isn’t interested in pursuing an academic or research career. The degree emphasizes the practical aspects of engineering, combined with management and communication coursework. Graduates of our program have gone on to work for some of the leading engineering companies in the world, including Schlumberger, National Instruments, Intel, Microsoft, and ExxonMobil.

Shape Control of Al Nanocrystals
Christopher DeSantis
Faculty Advisor: Naomi Halas
Over the course of the past two decades many researchers have demonstrated the remarkable morphological tunability of nanocrystals. The library of compositions of nanocrystals is immense, ranging from noble metals such as Au, and Pd to semiconducting crystals such as GaAs to metal oxides such as CuO. In these compositions, various shapes ranging from low-index {111} terminated octahedrons and tetrahedrons have been prepared along with high-index faceted crystals in shapes such as trisoctahedron and tetrahexahedron. These new shapes have expanded optical features, improved catalytic properties, and are the basis for many interesting assembly designs. Recently, nanoscale Al has become of interest due to its abundance and ability to support a plasmon from the ultraviolet, visible, and infrared region of the electromagnetic spectrum. Unfortunately, colloidal synthesis of Al nanocrystals has proven challenging due to the high susceptibility of oxidation of alane precursors. Here, we demonstrate morpholocial control of Al nanocrystals is achievable through alane polymerization with coordinated ethereal or aminated solvents. Importantly, we find that controlling precursor formation, we can achieve homogeneous samples of a singly-twinned Al crystal with {111} and {110} facets. These results provide the foundation for a low-cost plasmonic material with tunable structural features and will open up new commercially viable pathways to light-based applications such as photocatalysis and water distillation. Authors: Christopher DeSantis*, Benjamin Clark, Michael McClain, David Renard, and Naomi Halas.