Energy Efficiency in MIMO WLANs: Today's WLANs (IEEE 802.11n/ac) rely on higher order MIMO and wider channel widths to increase the achievable data rates. We study energy efficiency and channel contention issues, and identify that increasing number of MIMO spatial streams is more energy efficient compared to utilizing wide channel widths. Our work also shows that In heterogeneous channel width environment (coexistence with legacy 802.11a/n) where different links operate at different channel widths, competition to access the medium becomes increasing unfair which results into starvation of the larger channel width links. 
- Papers: [Networking '14], [ETT '15]
   
60 GHz Millimeter-wave Networking: Communications at 60 GHz unlicensed spectrum (57-66 GHz) is becoming popular due to its ability of providing multi-gigabit data rates (IEEE 802.11ad, 802.15.3c) and supporting applications like wireless HD uncompressed video streaming. Due to its directional nature, user mobility and real-time beamtracking is a challenging problem in 60 GHz links. In our research, we propose a joint algorithm for antenna element selection (phased antenna arrays) and beamwidth adaptation based on device mobility to maintain an uninterrupted connectivity.
- Papers: [MASS '15]
   
Visible Light Communication (VLC): LEDs can be exploited to encode data in the emitted light which can then be received by a photodiode or a camera. VLC can enable many novel applications where users can locate a transmitter (i.e. LED) and specifically receive data from it. In our research, we increase the achievable data rates of LED-to-camera communication using RGB LEDs and color shift keying. Our scheme ensures that even when data is transmitted using color symbols, the human perceived color of the LED remains white (flicker-free illumination). Reed-solomon codes are employed to deal with the data loss that occurs while receiving data on a rolling-shutter camera.
- Papers: [CoNEXT '15], [COMSOC-CST '15]
   
Service Continuity and Fault Tolerance (project status - completed): Designing highly-available (availability of two nines or three nines) wireless access networks is extremely challenging due to frequent link failture. In our research, we proved analytcally that in case of probabilistic link failures, network connectivity and data rate capacity are inherently at odds, and performability (performance weighted availability) is a better metric to assess when studying the deployment factors of multi-hop wireless networks. We also developed a relay load balancing scheme for multi-hop wireless networks that utilizes betweenness centrality for variable transmission power control and topology design.
- Papers: [TMC '14], [COMSOC-CST '11], [COMCOM '12], [GlobeCom '10], [WCNC '10]
- Book: [Springer '12]
- Centennial campus outdoor wireless network testbed (NC State): [CentMesh]
- Quail Ridge wireless network testbed (UC Davis): [QuRiNet]
   

Ubiquitous Sensing and Mobile Computing

Continuous Sensing with Low-power Mobile Sensors: Continuous sensing applications on mobile devices (e.g. constantly listening for user's voice command) can result in substantial energy drain. We propose to utilize low-power MEMS sensors (e.g. accelerometer, gyroscope, barometer etc.) to act as proxy sensors for continuous sensing, and triggering the use of energy-hungry sensors only when needed. We show that accelerometer sensor commonly found in today's mobile devices is sensitive to user's voice, and can accurately listen for hotwords like "Okay Google" while consuming very low energy. Similarly, the MEMS barometer available in smartphones can continuously monitor for building door open/close events due to the pressure difference created by HVAC system.
- Papers: [MobiSys '15], [UbiComp '15]
   
Mobile Interaction using Wrist-worn Wearables: The increasing popularity of wrist-worn wearables (smartwatches and fitness trackers) presents an opportunity to recognize user's hand, arm and finger gestures using the motion sensors in the devices. In our research, we demonstrate that due to the movements of wrist tendons while moving the fingers, it is possible to uniquely identify a variety of finger gestures through wrist-worn wearables. The motion identification is also useful in building a finger-writing input system where a user can write characters on a surface while wearing the smartwatch to input text to a connected device or the watch itself.
- Papers: [HotMobile '15]
- Demo: [MobiSys '15], [Video]
   
Context-awareness for First-Person Camera Devices: Many emerging wearables like Google Glass and Narrative life-logger are equipped with point-of-view cameras to capture photos and videos from the first person's perspective. A major privacy concern in adoption of these device is that they can record many sensitive and private moments of user's day-to-day life. We showed that instead of continuously recording audio/video and analyzing it for sensitive context detection, it is more energy efficient to fingerprint the sensitive contexts using low-power sensors like accelerometer, light sensor and magnetometer in the wearables. These sensors can accurately profile user’s surrounding and her activity levels. Using these context fingerprints, we built a cascade of machine learning classifiers that can control the access to the camera sensor.
- Papers: [SECON '15]
   
Contactless RF Sensing: The ubiquitous presence of RF signals for wireless communication can be exploited for contact-free sensing of human motion and activities. Our work showed that the Channel State Information (CSI) of the WiFi link between user's smartphone and the AP can be analyzed at the AP to determine the motion of user's arm. Similarly, when a user is in proximity of a WiFi link, the user's arm gestures, activities (sitting, walking, running etc.) can be determined by mining the signal reflected from her body after a careful multipath removal process. We used these concepts for physical analytics applications such as detecting a shopper's behavior in a retail store. The remote RF sensing is also useful in other applications like device-free pedometer and gait analysis. Our exploration also includes utilizing 60 GHz millimeter-wave signals for remote sensing as they provide an added advantage of reduced multipath.
- Papers: [HotWireless '15], [WPA '15]
   

Large-scale Network Analytics

Characteristics of Multi-device Mobile Users: Over two third of adults in US currently own a laptop, a tablet and a smartphone, and the number of wireless-enabled devices that users carry is consistently increasing. We carried out a study of large user population to understand the Internet access patterns of the multi-device users. We found that when a user owns multiple wireless-enabled devices, the overall network usage increases proportionally with the number of devices, rather than being divided across the devices. Our study also showed that usage of a device remains more or less independent of the existence of user's other devices.
- Papers: [SECON '15],
   
Privacy and Security through Network Traffic Analysis: Utilizing user's contextual location (e.g. home, office) for providing location-based services can protect user's privacy by hiding her exact geographical location. However, determining user's contextual location is difficult. In our research, we propose a solution where user's network traffic can be analyzed for extracting intrinsic access patterns that are reflective of user's contextual location without the need of any deep packet inspection. We have also demonstrated the effectiveness of network traffic mining for detecting access to malicious web domains and how it can augment existing screening methods based on lexical analysis.
- Papers: [INFOCOM '14],