NEPSSS is a bi-monthly seminar series showcasing PhD student research in the department and providing the community with a technical and social gathering as well as an opportunity for students to develop their presentation skills. A schedule as well as slides and videotaped talks are available on the at NEPSSS website.
Fall 2016 seminars
December 14, 2016 — 12:00 p.m. in 442 Dana
Title: EEG-assisted Modulation of Sound Sources in the Auditory Scene
Speaker: Marzieh Haghighi
Advisor: Professor Deniz Erdogmus
Noninvasive EEG (electroencephalography) based auditory attention detection could be useful for improved hearing aids in the future. This work is an attempt to investigate the feasibility of online classification of auditory attention using a noninvasive EEG-based brain interface. Proposed online system modulates the upcoming sound sources through gain adaptation which combines decisions from a classifier trained on offline calibration data. For decision making, features are extracted based on cross correlation of EEG and speech envelope at specific time lags that were shown to be useful to discriminate attention in the competing speakers’ scenario. In addition, a new signal modeling approach is introduced which results a lower dimensional sets of features. Attention detection performance of the model and its application to online source modulation is reported in the form of AUCs. On average, for attended speaker classification in training session and application of learned model on online session, the presented approach yields 88% and 82% AUC values respectively, using 20 seconds of data for each decision. A more general probabilistic framework for online closed loop sound source modulation is also introduced which takes new factors of previous attentional states, context information and energy of the sounds into account for future modulation of sound sources.
Title: In-Scene LWIR Downwelling Radiance Estimation
Speaker: Michael Pieper
Advisor: Professor Vinay Ingle
Effective hyperspectral thermal infrared imaging requires accurate atmospheric compensation to convert the measured at-sensor radiance to the ground radiance. The ground radiance consists of the thermal emission of the material and the reflected downwelling radiance. An accurate estimate of the downwelling radiance is required for temperature-emissivity separation (TES) to remove the spectrally sharp reflected atmospheric effects and retrieve a smooth and accurate material emissivity to use for detection.
Determination of the downwelling radiance is difficult due to the fact that a down-looking sensor has no knowledge of the atmospheric properties above its line of sight. As the sensor altitude increases and more of the atmospheric emitters are below the sensor, a relationship forms between the upwelling and downwelling radiances. This relationship comes at the expense of increased pixel size, which increases the likelihood of mixed pixels and nonlinear spectral mixing.
In this paper improvements to methods used to estimate the downwelling radiance of low altitude collections are proposed. The ground radiances of reflective pixels are used to estimate the atmosphere above the sensor. The reflective pixels are identified from their sharp atmospheric spectral features. Using the assumption that emissivity spectra are smooth across the narrow reflected atmospheric downwelling radiance features, the temperatures and emissivities are then separated for the reflective pixels using a look-up-table of downwelling radiances. The downwelling radiance that provides the best overall fit for the reflective pixels is then chosen as the scene downwelling radiance.
Michael Pieper is a PhD. student at Northeastern University. He received his BS and MS degrees in electrical engineering from Northeastern University in 2007. His current research interests include atmospheric compensation and temperature-emissivity separation for LWIR hyperspectral imaging remote sensing.
November 16, 2016 — 12:00 p.m. 442 Dana
Title: OFDM for Multi-carrier Modulation
Speaker: Amir Tadayon
Advisor: Professor Milica Stojanovic
This talk mainly focuses on advanced signal processing techniques for multi-carrier modulation, in particular, orthogonal frequency division multiplexing (OFDM). OFDM promises a substantial increase in data rate and robustness against the frequency selectivity of multi-path channels. For coherent detection, channel estimation is essential for receiver design. In this talk, we will present a receiver design where the channel estimator exploits the sparse nature of the physical channel. We present the most popular subspace algorithm from the array processing literature, namely root-MUSIC, recent sparse identification algorithms in the form of orthogonal matching pursuit (OMP) and basis pursuit (BP), and a hybrid method called path identification (PI) algorithm. We also compare the performance of these estimators with that of the conventional estimators such as least-squares (LS) estimator and linear minimum-mean-squares estimator (LMMSE).
Amir Tadayon is a Ph.D. candidate working with Prof. Milica Stojanovic at Northeastern University, Boston. His current research focuses on channel estimation for OFDM systems. He received his M.S. from Northeastern University in Electrical Engineering and his BS from Tehran University, Iran, in Electrical Engineering.
Title: Recursive Bayesian Coding for AACs
Speaker: Matt Higger
Advisor: Professor Deniz Erdogmus
ALS or Brain Stem Stroke may induce a paralysis from which a person cannot communicate. Because of this it may be difficult to determine if an individual is conscious or not. It is a sad fact that outward signs of intent are subtle enough that it is more often the families of patients than their physicians who discover some tell tale sign of their consciousness.
In this talk, we describe a Recursive Bayesian framework which seeks to aggregate the modest, uncertain output of a person into trustworthy decisions. In particular, we design error models which capture the varying accuracy and confusion of arbitrary user symbols (muscle movements, eye gaze fixations, EEG features, sip-and-puff etc). Such a model allows us to leverage this confusion structure in incorporating evidence into the system's belief of the user’s intended task message (letters in typing applications or destinations in wheelchair control etc). Most importantly, we make use of this confusion structure in optimizing our query scheme to learn the user's intent as quickly as possible. To accomplish this, we motivate a Mutual Information objective and present results which suggest it outperforms other strategies in an Brain Computer Interface spelling task.
Finally, we will share ongoing work on building a system for two men who need such a system. Academic interests aside, there is a team of us who are very interested in building something to help these men. We are currently performing sessions which scan their physiology for signals (EMG, EEG, Eye Gaze) which can be reliably classified. This problem has put us through our paces and our bottlenecks to success, as can be expected, are often outside our wheelhouse. For this reason we warmly welcome any insight or discussion which could yield improvements for our guys.
October 26, 2016 — 12:00 p.m. 442 Dana
Title: Performance Enhancement and Modeling of System Data Compute, Transfer and Storage
Speaker: Janki Bhimani
Advisor: Professor Ningfang Mi & Miriam Leeser
In the current era of big data and cloud computing, everyone wants more speed with no extra cost. Attaining optimal throughput in terms of shortest execution time with better endurance of lifetime sensitive hardware requires making good design choices. The large number of design decisions makes it nearly impossible to obtain the optimal performance point. This emphasizes the need for high-level design decision guidelines built by exhaustive performance engineering. However, this exhaustive method is very expensive in terms of valuable resources like time and cost. So, modeling and predicting performance of an application and optimal number of parallel resources is important. In order to address the above challenges, this research work concentrates on performance enhancement and modeling. The major contributions span performance engineering and modeling of data computation, data communication, memory management and storage.
Janki Bhimani is Ph.D. candidate working with Prof. Ningfang Mi and Prof. Miriam Leeser at Northeastern University, Boston. Her current research focuses on performance prediction and capacity planning for parallel computing heterogeneous platforms and backend storage. She received her M.S. (in 2014) from Northeastern University in Computer Engineering. She received her B.Tech. (in 2013) from Gitam University, India in Electrical and Electronics Engineering.
September 28, 2016 — 12:00 p.m. in 442 Dana
Title: Structured Covariance Estimation and Use of ErrP to Improve RSVPkeyboardTM
Speaker: Paula Gonzalez-Navarro
Advisor: Professor Deniz Erdogmus
Multichannel electroencephalography (EEG) is widely used in non-invasive brain computer interfaces (BCIs) for user intent inference, including augmentative and alternative communication (AAC) systems. EEG can be assumed to be a Gaussian process with unknown mean and autocovariance, and the estimation of parameters is required for BCI inference. However, the relatively high dimensionality of the EEG feature vectors with respect to the number of labeled observations lead to rank deficient covariance matrix estimates for multivariate Gaussian class-conditioned feature density models. Typically, this problem was tackled by applying regularization on maximum likelihood covariance matrix estimators, resulting in regularized discriminant analysis (RDA) for discriminative dimension reduction. In this talk, to overcome ill-conditioned covariance estimation, we propose a structure for the covariance matrices of the multichannel EEG signals. Specifically, we assume that these covariances can be modeled as a Kronecker product of temporal and spatial covariances.
Moreover, EEG signal in current Event related potential (ERP)-based typing systems has very low signal-to-noise ratio and to detect the user intent becomes a challenge. To increase accuracy, through a visual presentation of letters, repeated stimuli are normally used to detect the user intent which decreases the typing speed. Alternatively, here we propose to use the detection of error related potentials (ErrP) in the EEG response and propose different probabilistic approaches to incorporate ErrP evidences in decision making and auto-correction. With simulations on prerecorded real EEG calibration data using our BCI typing system RSVPkeyboardTM, we show that our auto-correction method can improve typing speed without sacrificing accuracy.
Summer 2016 seminars
June 29, 2016 — 12:00 p.m. in 440 Egan
Title: Ultrasonic Networking Technologies for the Internet of Implantable and Wearable Things
Speaker: Enrico Santagati
Advisor: Professor Tommaso Melodia
Wirelessly networked systems of implantable and wearable medical devices endowed with sensors and actuators will be the basis of many innovative, sometimes revolutionary therapies. However, biological tissues are composed primarily of water, and radio-frequency (RF) electromagnetic waves, which are the physical basis of currently used wireless technologies like Wi-Fi and Bluetooth, do not propagate well in water and heat body tissues. Additionally, RF communications can be easily jammed or eavesdropped. This raises major privacy and security red flags, and a risk for the patient.
Given the limitations of RF propagation, we proposed and investigated the use of ultrasonic waves as an alternative carrier of information in human tissues. Compared to RF waves, ultrasonic waves have significantly lower absorption by human tissues and therefore require lower transmission power, resulting in lower energy consumption, in longer battery life and a smaller size for an implantable medical device. Moreover, ultrasonic waves do not easily penetrate through solid materials and do not propagate far in air; therefore, ultrasonic communication systems are inherently more secure than RF with respect to eavesdropping and jamming attacks.
In this talk, we will first discuss the development of the Ultrasonic WideBand (UsWB) technology, the first integrated physical and medium access control protocol developed for networks of implantable devices, and the implementation of the first generation software-defined testbed used to test the UsWB performance. We will then present the design and development of a second generation UsWB prototype that will allow us to prove that ultrasounds are truly a viable lower-power and more secure alternative to RF for medical implants even when all system constraints are considered.
May 25, 2016 — 12:00 p.m. in Forsyth 128
Title: State-action based Link Layer Design for IEEE 802.11b Compliant MATLAB-based SDR
Speaker: Ramanathan Subramanian
Advisor: Professor Kaushik Chowdhury
Software defined radio (SDR) allows unprecedented levels of flexibility by transitioning the radio communication system from a rigid hardware platform to a more user-controlled software paradigm. However, it can still be time consuming to design and implement such SDRs as they typically require thorough knowledge of the operating environment and a careful tuning of the program. In this work, we describe a systems contribution and outline strategies on how to create a state-action based design in implementing the CSMA/CA/ACK MAC layer in MATLAB that runs on the USRP platform, a commonly used SDR. Our design allows optimal selection of the parameters so that all operations remain functionally compliant with the IEEE 802.11b standard (1Mbps specification). The code base of the system is enabled through the Communications System ToolboxTM and incorporates channel sensing and exponential random back-off for contention resolution. The current work provides a testbed to experiment with and enables creation of new MAC protocols starting from the fundamental IEEE 802.11b compliant standard. Our system design approach guarantees the consistent performance of the bi-directional link and we include the experimental results for the three node system to demonstrate the robustness of the MAC layer in mitigating packet collisions and enforcing fairness among nodes.
Spring 2016 seminars
March 30, 2016 — 12:00 p.m. in room 442 Dana
Title: FlashTypeTM: A Responsive cVEP-based BCI Speller
Speaker: Sadegh Salehi
Advisor: Professor Deniz Erdogmus
Brain computer interfaces (BCIs) offer new augmentative and alternative communication (AAC) opportunities for individuals with severe speech and motor impairments. Among different brain activities visually evoked potentials (VEP) are the most effective in BCI design, in terms of accuracy and speed of designed systems, including keyboard applications. In this talk, we describe FlashTypeTM, a brain interface based on code-VEP (cVEP) that utilizes a language model informed keyboard layout with static and dynamic keys.
The proposed system allows the user to move a cursor on a keyboard layout to make a symbol selection with minimum expected number of steps per selection. The static portion of the keyboard can be optimized according to a 1-gram letter probability distribution model for English. This portion is supplemented by a row of dynamically adjusted suggested characters, and a row of dynamically adjusted predicted words, to which the user may navigate with ease, reducing the average time to complete a word. This dynamic adjustment uses a 6-gram letter model for English that is fused with all recent EEG evidence to obtain a posterior probability distribution over the alphabet and dictionary. Moreover, to increase the typing speed and decrease the number of wrong decisions, we investigated two probabilistic graphical models for Bayesian inference, which uses context information and available EEG evidence to obtain the posterior probability distribution over the decision space. The two models will be discussed and performance analyses will be presented.
Although FlashTypeTM uses a cursor-based hierarchical selection method, due to the high accuracy of dynamically adjusted predictions, users tend to make the majority of their selections from the adaptive rows, which significantly reduces average time to type a letter or word by requiring minimal cursor movement steps.
Title: Enabling Protocol Coexistence: High-Level Hardware-Software Co-design of Flexible Modern Wireless Transceivers
Speaker: Benjamin Drozdenko
Advisor: Professors Kaushik Chowdhury and Miriam Leeser
The recent increase in the number of wireless devices has been accompanied by an explosion in the number of protocols for wireless communications, each focusing on different purposes such as execution time reduction, energy reduction, handling higher congestion levels, or operation at different bandwidths. Software-defined radios have introduced new platforms for dynamically modifying wireless system designs, and heterogeneous computing has opened up implementation of such designs on different computing elements. Up to now, researchers have focused on designing complete protocol-specific processing chains. In contrast, our goal is to develop a modeling environment that captures reusability of various processing blocks at the physical layer for several modern protocols, and makes decisions regarding whether processing blocks should be part of reconfigurable hardware or embedded processor software. In this paper, we introduce an integrated profiling approach to implement the 802.11a standard on the Xilinx Zynq system-on-chip. Our approach creates several different MathWorks Simulink model variants for both the transmitter and the receiver, each with a different boundary between hardware and software components. We use these models to generate a bitstream for the FPGA and executable code for the ARM processor. Using this modeling environment, we investigate the HW/SW divide point and identify specific processing blocks to focus on improving. Our results collect such metrics as data path delay, resource utilization, and power usage to demonstrate how exploring variants to processing blocks can further enhance the design.
February 24, 2016 — 12:00 p.m. in room 442 Dana
Title: Revisiting Accelerator-Based CMPs: Challenges and Solutions
Speaker: Nasibeh Teimouri
Advisor: Gunar Schirner
Utilizing Hardware Accelerators (ACCs) is a promising solution to improve performance / power efficiency of Chip Multi-Processors (CMPs). However, new challenges including scalability arise with a trend to shift from few ACCs (with sparse ACCs coverage) to many ACCs (denser ACCs coverage) on a chip. Resources including memory, communication fabric and processor turn into bottlenecks and result in accelerator under-utilization and cripple the performance. The source of this challenge is a lack of clear semantic to communicate with ACCs as well as a processor-centric view for orchestrating the entire system. To open a path toward efficient integration of many ACCs on a single chip, at first we identify 4 major semantic aspects when two ACCs need to communicate with each other: (1) data access
model, (2) data granularity, (3) marshalling, and (4) synchronization. Then, we propose Transparent Self-Synchronizing (TSS) as an efficient architecture realization of those semantic aspects. In principle, TSS proposes a shift from the current processor-centric view to a more equal, peer view between ACCs and the host processors. TSS minimizes the interaction with the host processor and reduces the volume of ACC-to-ACC communication traffic exposed to the
system fabric. Our results using some streaming applications with a variable number of ACC-to-ACC connections demonstrate significant benefits of TSS including 3x speedup over the current ACC based architectures.
Title: A Complete Key Recovery Timing Attack on a GPU
Speaker: Zhen Jiang
Advisor: Yunsi Fei
Graphics Processing Units (GPUs) have become mainstream parallel computing devices. They are deployed on diverse platforms, and an increasing number of applications have been moved to GPUs to exploit their massive parallel computational resources. GPUs are starting to be used for security services, where high-volume data is encrypted to ensure integrity and confidentiality. However, the security of GPUs has only begun to receive attention. Issues such as side-channel vulnerability have not been addressed. The goal of this paper is to evaluate the side-channel security of GPUs and demonstrate a complete AES (Advanced Encryption Standard) key recovery using known ciphertext through a timing channel. To the best of our knowledge, this is the first work that clearly demonstrates the vulnerability of a commercial GPU architecture to side-channel timing attacks. Specifically, for AES-128, we have been able to recover all key bytes utilizing a timing side channel in under 30 minutes.
January 20, 2016 — 12:00 p.m. in room 442 Dana
Title: All-optical magnetic recording for next-generation magnetic storage
Speaker: Feng Cheng
Advisor: Yongmin Liu
The emerging Big Data era demands the ever increasing speed and capacity to store and process information. Recent research has shown that it is possible to realize deterministic and controllable switching of magnetic orders by ultra-fast light pulses. Unlike conventional magnetic storage devices, such an extremely fast and novel reversal mechanism does not require an external magnetic field, which provides us with an opportunity to write data with light. In this talk, I will first introduce the mechanism of all-optical magnetic recording, and our characterization system based on the magneto-optical Kerr effect. Then I will talk about the preliminary experimental results of all-optical switching phenomenon observed in Co/Pt magnetic multilayers. We have explored the influence of different laser repetition rates and peak powers, and robust switching effect is observed. Finally, I will discuss the future work to realize nanoscale magnetization switching for high capacity and low-power data storage by integrating pragmatically designed plasmonic lenses.
Feng Cheng received the BS degree in Opto-electronic engineering from Tianjing University (Tianjin, China) in 2014. He is currently working in the research group of Prof. Yongmin Liu. His research interests include nanophotonics, metamaterials and plasmonics.
Title: K-means-based Consensus Clustering and its Applications
Speaker: Hongfu Liu
Advisor: Yun Fu
Consensus clustering aims to find a single clustering which agrees with several basic partitions as much as possible, which recently attracts increasing attentions. In this talk, I will introduce some basic concepts of consensus clustering and some advanced algorithms. All these algorithms are K-means-based, which are of high efficiency and robustness. Some applications on document clustering, gene expression analysis, constrained clustering are also included.
Hongfu Liu received his B.E. and master degree in Management Information Systems from the School of Economics and Management, Beihang University, in 2011 and 2014, respectively. He is currently a second-year PhD student in Northeastern University. His research interests generally focus on data mining and machine learning, with special interests in ensemble learning.
Fall 2015 seminars
November 12, 2015 — 4:00 p.m. in room 442 Dana
Title: Graphene Enhanced Ultra-high Frequency Piezoelectric Nanoelectromechanical Systems
Speaker: Zhenyun Qian
Advisor: Matteo Rinaldi
Micro and nano-electromechanical systems (MEMS and NEMS) are key drivers behind a number of advanced applications such as radio frequency (RF) wireless communications, single-molecule detection, switches, infrared-imaging, magnetometers, and chemical sensors. Many of these are driven by on-chip piezoelectric actuation and sensing of ultra-high frequency (UHF) vibration in miniaturized free-standing micro and nano mechanical structures thanks to their unique advantages of extremely high sensitivity to external perturbations and ultra-low noise performance. In this context, designing “ideal electrodes” that simultaneously guarantee low mechanical damping and electrical loss as well as high electromechanical coupling in such ultralow-volume piezoelectric nanomechanical structures is a key challenge. In this talk, I will show that mechanically transferred graphene, floating at van der Waals proximity, can closely mimic “ideal electrodes” for UHF piezoelectric nanoelectromechanical resonators with negligible mechanical mass and interfacial strain and perfect electric field confinement. These unique attributes enable graphene-electrode-based piezoelectric nanoelectromechanical resonators to operate at their theoretically “unloaded” frequency-limits with significantly improved electromechanical performance compared to metal-electrode counterparts, despite their reduced volumes. This represents a spectacular trend inversion in the scaling of piezoelectric electromechanical resonators, opening up new possibilities for the implementation of NEMS with unprecedented performance. Furthermore, the transparent and chemically active natures of the atomically-thin graphene electrode enable unique IR detection and chemical sensing capabilities of such graphene enhanced nanomechanical resonators, making them a promising candidate for the development of both high resolution resonant IR detectors and chemical sensors.
Title: Galvanic Coupled Intra-body Communication Technology
Speaker: Meenupriya Swaminathan
Advisor: Kaushik Chowdhury
Implanted sensors will drive the next generation of personalized medicine with in-situ monitoring of abnormal physiological conditions using implanted sensors, and proactive drug delivery using embedded actuators. Autonomous and direct communication among implants through the body tissues is crucial for realization of the above vision, which we achieve using galvanic coupling of weak modulated electrical signals.
This new communication paradigm involves several interesting challenges, including (i) signal propagation characterization of different tissue paths and (ii) identifying the best placements of implants and relays towards extended implant life. We systematically analyze the channel between implanted scenarios (e.g., sensor on skin and actuator in muscle) using suite of equivalent circuit model for three dimensional human arm with four tissue layers - outer dry skin, fat, muscle and bone. Each tissue is modeled using four impedance based on the current paths for various sensor separations, tissue dimensions, hydration levels, operating frequency, noise, and electrode specifications. The results are verified with finite element human arm simulations and empirical measurements using porcine tissue.
Using the channel characters thus obtained, the basic link budget is computed including the channel bandwidth, capacity, quantity of implants that can be accommodated and their life expectancy. Towards the topology formation, we then devise energy efficient relay placement strategy and show that the proposed technique extends the implant life by years.
Maturity in galvanic coupled intra body communication with suitable physical layer and medium access protocols will potentially revolutionize health care with diverse applications arising out of a network of connected implants.
Click here for a video preview of this talk.
October 15, 2015 — 4:00 p.m. in room 442 Dana
Title: A High-Speed Low-Power Hybrid Analog-to-Digital Converter for Wireless Portable Medical Devices
Speaker: Alireza Zahrai
Advisor: Marvin Onabajo
In recent years, telemedicine has become popular because it makes access to healthcare more convenient with lower cost, thereby saving lives through early diagnosis and real-time monitoring. The goal of this research is to design a high-speed analog-to-digital converter (ADC) to be used in portable communication chips for telemedicine applications that require low power consumption to extend the lifetimes of batteries. The high-speed and low-power performance is achieved by devising a hybrid architecture that combines the advantages of two different types of ADCs. In the first stage, a flash ADC resolves the three most significant bits of the analog input signal, and the remaining five bits are determined by four time-interleaved low power SAR ADCs in the second stage, leading to an overall hybrid ADC having 8-bit resolution while operating with a 1GHz sampling clock signal.
The hybrid ADC was designed and simulated with a mix of behavioral models and transistor-level circuit designs in 130nm CMOS technology. The estimated power consumption is 15mW from a 1.2V supply.
Alireza Zahrai is a Ph.D. candidate in Electrical Engineering and research assistant in the Analog & Mixed-Signal Integrated Circuit (AMSIC) Research Laboratory, Northeastern University, Boston. He received his B.Sc. and M.Sc. in Electrical Engineering from the University of Tehran and Iran University of Science and Technology, Tehran, Iran respectively. He was an IC Design Intern at CSR plc, Tempe, AZ in 2013, where he worked on the design of a SAR ADC for an audio SoC. His current research includes low-power high-speed time-interleaving ADC design (GS/s) and on-chip digital calibration systems. His research interests are analog and mixed-signal IC design, high-speed data converters and digitally-assisted analog circuits.
Title: Robust Fault Location for Two and Three Terminal Lines Using Synchronized Phasor Measurements
Speaker: Guangyu Feng
Advisor: Ali Abur
This talk introduces a robust fault location method that utilizes wide area phasor measurements and sparse estimation technique. The proposed method transforms the fault location problem into estimating sparse bus injections in the network, based upon the equivalence in the change of bus voltages between a fault current drawn at an arbitrary point along a line and virtual superimposed current injections at the terminal nodes of the same line. This equivalence not only works for two terminal lines but also three terminal lines. Assuming limited placement of phasor measurements, an underdetermined linear estimation problem whose solution is sparse will be formed. This problem can be solved via sparse estimation especially L1 regularization technique. Considering possible failure of individual measurement units, an extended formulation incorporating sparse error vector is used to increase the method's robustness. Extensive simulation results have verified the effectiveness of the proposed method.
Guangyu Feng received the B.S. degree in electrical engineering from Tsinghua University, Beijing, China in 2013. She's currently pursuing the Ph.D. degree in the Department of Electrical and Computer Engineering at Northeastern University. Her research interests include power system fault identification, computation and optimization methods.
September 17, 2015 — 4:00 p.m. in room 442 Dana
Title: TARS: A Traffic-Adaptive Receiver-Synchronized Medium Access Control Protocol for Underwater Sensor Networks
Speaker: Yu Han
Efficient medium access control (MAC) is desirable for underwater sensor networks (UWSNs). However, designing an efficient underwater MAC protocol is challenging due to the long propagation delay of the underwater acoustic channel and the spatial-temporal uncertainty.
In this work, we propose a novel Traffic-Adaptive Receiver-Synchronized underwater MAC protocol, TARS, a stochastic light-weight channel access scheme that addresses the spatial-temporal uncertainty for maximizing the network throughput. We adjust the packet transmission time (phase) in a slot, which is dependent on the sender-receiver distance, to align packet receptions for collision reduction. Both the sound propagation speed variation and the node mobility are considered in setting the optimal transmission phase and the slot size. We employ a queue-aware utility-optimization framework to determine the optimal traffic-adaptive transmission strategies dynamically, taking into account both the packet interference and the data queue status. Extensive simulation results show that compared to the existing representative underwater MAC protocols, TARS achieves better performance with higher network throughput and lower packet end-to-end delay.
Title: Learning with Robust Data Representations: Methodologies and Applications
Speaker: Sheng Li
Extracting knowledge from high-dimensional and large-scale data plays an important role in many real-world applications. Following a bottom-up framework, we can represent low-level raw features as mid-level codings, or even high-level representations. In this talk, I will introduce some approaches we have designed recently, including mid-level feature learning (e.g., low-rank codings, dictionary learning) and high-level feature learning (e.g., graph construction, subspace learning). Real-world applications such as image classification, person
re-identification, outlier detection and recommender system will be discussed as well.
Summer 2015 seminars
August 6, 2015 — 12:00 p.m. in room 442 Dana
Title: Optimizing 2-Axis Time-Position Tracking Photovoltaic Arrays Under Varying Sky Conditions
Speaker: Stephanie Quinn
Advisor: Bradley Lehman
Two-axis tracking photovoltaic (PV) arrays are 25 to 40% more efficient than fixed PV arrays in collecting solar radiation. They are typically installed in locations with consistently clear skies. But in recent years there has been a rapid increase in the number of these tracking arrays installed in cloudier locations. Commercial 2-axis tracking arrays may be categorized as follows: light-sensing trackers that use irradiance sensors to track the sun, and time-position trackers that use an astronomical algorithm based on the sun’s apparent position. Both types efficiently collect solar radiation under clear, sunny skies, but their performance declines during cloudy intervals. In the case of light-sensing PV trackers, the irradiance sensors may fail to track the sun when the sky is cloudy. Although time-position 2-axis PV trackers continue to track the sun when it is obscured by clouds, more irradiance could be collected if the PV array did not track the sun. The ability to modify the standard tracking angles in response to changing cloud cover allows the PV array to capture more of the total available irradiance, thereby increasing the amount of electricity produced. In this talk, we share our research in finding the optimal tilt angle of a 2-axis time-position tracking PV array so that its efficiency is maximized under varying sky conditions.
Title: Fault Injection Study on the AMD Evergreen Family of GPUs
Speaker: Fritz Previlon
Advisor: David Kaeli
With their numerous processing cores and their impressive parallel processing capabilities, Graphic Processing Units (GPU) have become the accelerator of choice across multiple domains, from scientific computing, bio-informatics and molecular biology to even financial applications. Their presence in the top supercomputers has been steadily growing over the last few years.
With technology scaling, soft errors or single-event upsets (change of state in a device which may lead to wrong program outputs) are becoming a high priority for designers as we’re moving forward. Recently, a study of the Department of Energy has identified soft errors as one of the top 10 challenges to exa-scale computing. We must take measures now in order to come up with ingenious solutions to the problem of soft errors. A key aspect in reliability study is that some soft errors will not cause an error at the output of a program. Therefore, an important step in tackling soft errors in GPUs is to first assess the impact of soft errors and the robustness of the GPUs in the presence of these errors.
In this work, we are presenting an error injection study on the AMD Evergreen family of GPUs using a detailed architectural simulator. Our results show that the GPU can be a highly resilient system. We also present a study of some observed trends in the vulnerability of GPU programs and the GPU memory hierarchy. These trends can be further used by programmers as well as system designers when making decisions about GPU reliability.
July 9, 2015 — 12:00 p.m. in room 442 Dana
Title: Clustering and Ranking in Heterogeneous Information Networks via Gamma-Poisson Model
Speaker: Junxiang Chen
Advisor: Jennifer G. Dy
Information networks are widely applied to represent objects and their interactions in real-world systems in different academic fields. Examples include gene regulatory networks, semantic networks and social networks. As a result, network analysis draws plenty of attention from the research communities. Clustering and ranking are the most widely applied network analysis techniques. These techniques have been successfully applied independently to homogeneous information networks, i.e. networks that contain only one type of objects and links. However, real-world information networks are oftentimes heterogeneous, containing multiple types of objects and links. In addition, recent research has shown that clustering and ranking can mutually enhance each other. In this talk, I will introduce a probabilistic generative model that simultaneously achieves clustering and ranking in heterogeneous information networks, where edges from different types are modeled as samples from Poisson distributions with parameters determined by the “ranking scores” of the vertices in each cluster. The model is applied to two real-world networks extracted from DBLP and YELP data to illustrate its effectiveness.
Title: Dense Electrode Array Current Optimization for Targeted and Directional tDCS
Speaker: Seyhmus Guler
Advisor: Dana Brooks
Dense array transcranial direct current stimulation (tDCS) is an emerging tool to modulate brain function noninvasively via high-density electrode arrays (64-512 electrodes) placed on the subject’s scalp. However, there is need for adequate methods to determine stimulus patterns for such arrays as the degrees of freedom is much higher than conventional two patch electrode configurations. This talk will discuss a method for calculating electrode stimulus patterns for targeted and directional modulation in tDCS. It details the construct and development of an optimization problem that provides an optimal, unique stimulus pattern as a solution that meets a flexible set of safety constraints. The problem is extended to find sub-optimal stimulus patterns that use fewer current sources and thus are more practical in a clinical setting. Exemplary results for both empirical and MR imaging data based brain target regions of interest are shown.
June 4, 2015 — 12:00 p.m. in room 442 Dana
Title: Next-generation Cyber-Physical Systems Utilizing RF-Powered Computing
Speaker: Yousof Naderi
Advisor: Kaushik R. Chowdhury
RF-powered computing is an emerging technology in which small computing devices use ambient and controlled electromagnetic radio frequency (RF) waves for power and communication. This technology can favor a wide range of applications from indoor elderly patient monitoring to outdoor bridge health monitoring for detecting the dangers. However, the coexistence of data communication and energy comes at the cost of new challenges.
This talk is structured around three parts. We first describe RF-powered CPS and some design challenges in these systems, which highlight the need to engineer a system that manage energy interference, coordinate the distribution of wireless energy transfers, control the power, and schedule optimal times for data and energy communications. We share the latest experimental results on evaluating the concurrent low-power data and high-power energy transfer as well as surviving wireless energy interference.
Second, we will introduce an medium access protocol for ET and sensor coordination that jointly selects energy transmitters and their frequencies based on the collective impact on charging time and energy interference, sets the maximum energy charging threshold, requests and grants energy, and decides the access priority of both data and energy.
Finally, we will discuss our ongoing research on a cognitive RF-powered energy harvesting system that allows sensors equipped with multi-band RF harvester to operate continuously by switching between harvesting energy from ambient cellular/TV bands as well as unlicensed bands that have directed power transfer.
Title: Non-invasive Brain Computer Interfaces for Assistive Technologies
Speaker: Mohammad (Sina) Moghadamfalahi
Advisor: Deniz Erdogmus
Brain-computer interfaces (BCIs) promise to provide a novel access channel for assistive technologies, including augmentative and alternative communication systems, for people with severe speech and physical impairments (SSPI). Among various options, non-invasive electroencephalogram (EEG)-based BCIs are considered as safe and more portable solutions which are potentially suitable for home use. The applications of these BCIs can include wheelchair navigation and typing. Research on the subject has been accelerating significantly in the last decade and the research community took great strides toward making non-invasive BCI a practical reality for individuals with SSPI.
Nevertheless, the end goal has still not been reached and there is much work to be done to produce real-world-worthy systems that can be comfortably, conveniently, and reliably used by individuals with SSPI with help from their families and care givers who will need to setup and maintain these systems at home. In the Cognitive Systems Lab, we develop solutions to improve BCIs. Different stimulation strategies can induce unique detectable signatures in the EEG, such as steady state evoked potentials and event related potentials. Brain waves in response to these stimuli can be processed using machine learning and signal processing techniques. Being non-invasive, EEG signals have very low signal to noise ratios. One of the methods to increase the classification performance is to consider the context information in applications. For example, in a typing task, one can use a language model to predict the most probable next letter at each point in a sentence. In a navigation task, such as controlling a wheelchair, historical data can provide context that provides information about probabilities for feasible directions.
Spring 2015 seminars
April 16, 2015 — 4:00 p.m. in room 442 Dana
Title: Performance Evaluation of Hyperspectral Chemical Identification Systems
Speaker: Eric Truslow
Advisor: Vinay K. Ingle
Remote sensing of chemical vapour plumes is a difficult but important task with many military and civilian applications. Hyperspectral sensors operating in the long wave infrared have well demonstrated plume detection capabilities. However, identification of a plume's chemical constituents, using a chemical signature library, is a multiple hypothesis testing problem that standard detection performance metrics do not fully characterize. We propose using an additional performance metric for chemical identification based of the so-called Dice index. Using detection metrics and the proposed performance metric, we demonstrate that the intuitive system design of a detector bank followed by an identifier is justified when the additional metric is considered.
Title: Controlling The Interplay of Electric And Magnetic Resonances for Nanoantenns
Speaker: Kan Yao
Advisor: Yongming Liu
Nanoantennas are able to efficiently couple the energy of free-space radiation to a highly confined region with subwavelength dimensions and vice versa. This promises a wide spectrum of applications, such as near-field microscopy, spectroscopy and photovoltaics, etc. While the size of the optical antennas is one of the major concerns in practice, scaling down the classical microwave antennas is not sufficient for the future miniature. We propose an ultra-compact plasmonic nanoantenna with switchable directionality. The antenna comprises a metallic trimer that can support a highly spectrally tunable magnetic resonance with its amplitude comparable to that of the electric resonance. By moving one of the particles slightly for a few nanometers, the resonant frequency of the magnetic mode will shift dramatically, which leads to a change of the interference conditions and in turn the switching of the radiation direction. The enhancement of spontaneous emission and far-field radiation of a nano emitter coupled to the antenna can reach 4 and 3 orders of magnitude, respectively. Analyses based on a simple dipole model are performed and the reconstructed radiation patterns agree well with the numerical simulations.
April 2, 2015 — 4:00 p.m. in room 442 Dana
Title: Positive Stabilization With Maximum Stability Radius for Continuous-Time Dynamic Systems
Speaker: AmirReza Oghbaee
Positive systems have attracted much attention nowadays due to their numerous applications in modeling and control of physical, biological and economical systems. The state trajectory of such system remains in the nonnegative quadrant of the state space for any given nonnegative initial condition. This class of systems have nice stability and robustness properties. One can take advantage of these interesting properties to robustly stabilize general dynamic systems such that the closed-loop system becomes positive. One of the most important measures in robust control analysis is stability radius. This measure provides the amount of uncertainty that system can cope with before it becomes unstable. There are two types of stability radius defined; complex and real stability radius. Computation of real stability radius is more involved than its complex counterpart. Although the complex and real stability radius are different for a general LTI system, it has been found that they are equal for the class of positive system. In fact, a closed form expression can be obtained to find the stability radius of positive system. In this research, we try to positively stabilize a general uncertain system with the constraint of maximizing stability radius by using a state feedback control law. The necessary and sufficient conditions for the existence of controllers satisfying the positivity constraints are provided. This constrained stabilization problem will be formulated and solved using linear programming (LP) and linear matrix inequality (LMI). With the aid of bounded real lemma, the major contribution is to solve the constrained positive stabilization with maximum stability radius for both regular and time-delay systems.
Title: A Sparse Nonnegative Demixing Algorithm with Intrinsic Regularization for Multiplexed Fluorescence Tomography
Speaker: Vivian Pera
Fluorescence molecular tomography (FMT) is an optical technique that uses near-infrared light to perform quantitative, three-dimensional imaging of fluorophores in whole animals noninvasively. It is becoming an important tool in preclinical imaging of small animals and has been employed to image tumors and assess response to anti-cancer therapeutics. However, the inability to perform high-throughput imaging of multiple fluorescent targets (“multiplexing”) in bulk tissue remains a limitation. Recent work in our group suggests that joint measurement of spectral and temporal fluorophore data can enable robust identification (“demixing”) and localization of at least four concurrent fluorophores. Here we present a novel demixing strategy for this data, which incorporates ideas from sparse subspace clustering and compressed sensing. We will review the basic principles of FMT, present our demixing algorithm, and quantify its performance.
March 19, 2015 — 4:00 p.m. in room 442 Dana
Title: Integration of Phase Change Material Switches to Provide Reconfigurability to AlN MEMS Resonators
Speaker: Gwendolyn Hummel
This talk will explore the unique structures made possible by special materials such as piezoelectric and phase change materials. The use of these materials in devices such as resonators and switches will be explained and then the idea of combining these structures into a single device that can be reconfigured will be discussed and demonstrated.
Title: Low-Rank Transfer Learning and Its Application
Speaker: Ming Shao
For knowledge-based machine learning algorithms, label or tag is critical in training the discriminative model. However, labeling data is not an easy task because these data are either too costly to obtain or too expensive to hand-label. For that reason, researchers use labeled, yet relevant, data from different databases to facilitate learning process. This is exactly transfer learning that studies how to transfer the knowledge gained from an existing and well-established data (source) to a new problem (target). To this end, we propose a method to align the structure of the source and target data in the learned subspace by minimizing the reconstruction error, called low-rank transfer subspace learning (LTSL). The basic assumption is if each datum in a specific neighborhood in the target domain can be reconstructed by the same neighborhood in the source domain, then the source and target data might have similar distributions. The benefits of this method are two-fold: (1) generality to subspace learning methods, (2) robustness by low-rank constraint. Extensive experiments on face recognition, and objection recognition demonstrate the effectiveness of ourmethod.
March 5, 2015 — 4:00 p.m. in room 442 Dana
Title: Detection of OFDM Signals Over Acoustic Channels
Speaker: Yashar Aval
Detection of OFDM over acoustic channels is challenged by estimation of the highly time varying channel and the ensuing inter-carrier interference (ICI). While the common approach to reduce ICI is equalization over the carriers, we will discuss more effective methods which compensate for the channel variations in time domain. The experimental results from a recent experiment in the ocean will be followed by a real-time over-the air demo of the discussed methods which is used to transmit live video over the acoustic channel.
Title: Solving Time Puzzles
Speaker: Caglayan Dicle
I invite you to solve a new kind of puzzle with me which I call "Temporal Puzzles". Just like jigsaw puzzles where pieces of an image are shuffled, in Temporal Puzzles, frames of a video are shuffled and you are asked to sort them such that the final sequence is in correct temporal order. Say it is a skiing video or a video of an ocean wave, the critical question is, how would one develop an algorithm to solve such problems? Set a side difficulty of solving, what might be a plausible goal and how to formulate it?
In my presentation, I will briefly share my intuition to solve above problems. Starting from a fairly universal principle, I will share with you my simple plausible approach to formulate Time Puzzles and some insights about a practical solution. I will also touch upon two direct applications of proposed intuition and solution, namely crowd photography sequencing and similar object tracking.
February 19, 2015:
Title: Packetized Wireless Communication under Jamming: A Game Theoretic Approach
Speaker: Koorosh Firouzbakht
The convenience of wireless communication and its support of mobility has revolutionized the way we access information services and interact with the physical world. Nevertheless, security issues of wireless communications remain a serious concern and among the many security threats that the wireless networks are subject to, physical layer jamming is one of the most prominent and challenging ones. Jamming not only can lead to service interruption or denial of service, but it is often a prelude to other upper layer attack. An important question is to understand the interactions between the communicating nodes and the adversary, determine the long-term achievable performance and the optimal strategies to achieve it. To address this problem, we develop a general game-theoretic framework for a packetized wireless communication link under power limited jamming that can be used to study many jamming problems.
Title: Beyond Parallelism: Exploring multiple levels of concurrency on a modern GPU
Speaker: Yash Ukidave
PUs have gained tremendous popularity as accelerators for a broad range of applications belonging to various computing domains. Many applications have achieved large performance gains using the inherent parallelism offered by GPU architectures. Given the growing impact of GPU computing, there is a growing need for efficient utilization of compute resources and increased application throughput. Applications developed for modern GPUs include multiple compute kernels, where each kernel exhibits a distinct computational behavior and resource requirements. These applications place high resource demands on the hardware, commonly impose timing constraints, and demand concurrent execution of multiple kernels on the device. The growing use of GPUs in cloud engines, data centers, and smart devices demands an effective GPU sharing technique for multiple application contexts. An advanced mechanism is required to support the concurrent and flexible execution of multi-kernel applications. At the same time, such support has to be extended to schedule multiple application contexts on the same GPU. In our work, we explore the multiple levels of concurrency on a GPU, and suggest architectural and runtime enhancements to leverage the concurrency. We implement a dynamic, and adaptive mechanism to manage multi-level concurrency to improve the overall application throughput on the GPU.
The Northeastern ECE PhD Student Seminar Series is organized by PhD students themselves and serves the entire Northeastern ECE community, with multiple goals:
1. To provide students with experience presenting their own research to a general ECE technical audience, such as giving a conference presentation, a seminar, or a job talk at a university or company.
2. To provide students with guidance for clear and concise oral communication of research results to an audience of educated non-specialists.
3. To provide presenters with constructive feedback from designated evaluators on their presentations that they can understand and implement.
4. To provide students with the opportunity to think critically about seminars given by others, as part of their own process of learning to give effective seminars themselves.
5. To provide the entire ECE community with a forum to learn about the broad range of research in the department, in an accessible and interactive environment.
6. To provide a venue by which PhD students from very different areas of ECE can meet, educate each other, and find common interests and even potentially research collaborations.