Enabling Energy Efficient Sensing and Computing Systems by He Ba Submitted in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy Supervised by Professor Wendi B. Heinzelman Department of Electrical and Computer Engineering Arts, Sciences and Engineering Edmund A. Hajim School of Engineering and Applied Sciences University of Rochester Rochester, New York 2015 ii Biographical Sketch The author was born in Beijing, China. He received his Bachelor of Science degree in Electrical Engineering from Beijing Institute of Technology in 2008 and his Master of Science degree from the University of Rochester in 2011. He pursued his doctoral research under the direction of Professor Wendi Heinzelman. He worked as an Informatics Intern at UCB Pharma from February 2012 to June 2012. He worked as a Systems Engineering Intern at ASSIA Inc. from June 2014 to October 2014. His research interests lie in the areas of wireless sensor networks, signal processing and mobile computing. After graduation, he will start working at KPMG as a Software Engineer in Big Data from March 2015. The following publications were a result of work conducted during doctoral study: He Ba, Ilker Demirkol, and Wendi Heinzelman, \Feasibility and Benefits of Passive RFID Wake-up Radios for Wireless Sensor Networks," in IEEE Global Telecommunications Conference, 2010. He Ba, Jeff Parvin, Luis Soto, Ilker Demirkol, and Wendi Heinzelman, \Passive RFID-based Wake-up Radios for Wireless Sensor Network," in Wirelessly Powered Sensor Networks and Computational RFID, Springer Publishers, 2012. He Ba, Na Yang, Ilker Demirkol, and Wendi Heinzelman, \BaNa: A Hybrid Approach for Noise Resilient Pitch Detection," IEEE Statistical Signal Processing Workshop, 2012. Li Chen, He Ba, Wendi Heinzelman and Andre Cote, \RFID Range Extension with Low-power Wireless Edge Devices," in Proceedings of International Confer- ence on Computing, Networking and Communications, 2013. iii Li Chen, Stephen Cool, He Ba, Wendi Heinzelman, Ilker Demirkol, Ufuk Muncuk, Kaushik Chowdhury and Stefano Basagni, \Range Extension of Passive Wake-up Radio Systems through Energy Harvesting," in Proceedings of IEEE ICC, 2013. He Ba, Wendi Heinzelman, Charles-Antoine Janssen and Jiye Shi, \Mobile Computing - A Green Computing Resource," in Proceedings of IEEE Wireless Communications and Networking Conference, 2013. Tolga Soyata, He Ba, Wendi Heinzelman, Minseok Kwon and Jiye Shi, \Cloudlets: Extending the Utility of Mobile Computing," in Communication Infrastructures for Cloud Computing: Design and Applications, IGI Global, 2013. Rajani Muraleedharan, Ilker Demirkol, Ou Yang, He Ba, Surjya Ray, Wendi Heinzelman, \Sleeping Techniques for Reducing Energy Dissipation," in The Art of Wireless Sensor Networks, Springer Berlin Heidelberg, 2014. Na Yang, He Ba, Weiyang Cai, Ilker Demirkol, Wendi Heinzelman, \BaNa: A Noise Resilient Fundamental Frequency Detection Algorithm for Speech and Music," in IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2014. Minseok Kwon, Zuochao Dou, Wendi Heinzelman, Tolga Soyata, He Ba, Jiye Shi, "Use of Network Latency Profiling and Redundancy for Cloud Server Selec- tion," in IEEE International Conference on Cloud Computing, 2014. Colin Funai, Cristiano Tapparello, He Ba, Bora Karaoglu, and Wendi Heinzel- man, \Extending Volunteer Computing through Mobile Ad Hoc Networking," in IEEE Global Telecommunications Conference, 2014. Cristiano Tapparello, Colin Funai, Shurouq Hijazi, Abner Aquino, Bora Karaouglu, He Ba, Jiye Shi, and Wendi Heinzelman, \Volunteer Computing on Mobile De- vices: State of the Art and Future Research Directions," in Enabling Real-Time Mobile Cloud Computing through Emerging Technologies, IGI Global, 2015 iv Acknowledgements Throughout my study at the University of Rochester, I have received infinite help and support from a number of people, without whom this dissertation would not be completed. First and foremost, I would like to express my deepest appreciation to my advisor, Prof. Wendi Heinzelman. Her knowledge, wisdom, insight, patience and working attitude will not only lead me towards my doctoral degree but will also benefit me beyond my graduation. My gratitude also goes to Prof. Mark Bocko, Prof. Zhiyao Duan and Prof. Kai Shen for serving as my thesis committee members and providing invaluable comments and suggestions to my research and thesis. I would like to express my thanks to Dr. Jiye Shi, for offering me the internship opportunity at UCB Pharma and for providing insightful guidance and support during our collaboration since 2012. Over the years, I have had the pleasure to work with many exceptional fac- ulty members and researchers. Among them, I would like to specially thank Prof. Melissa Sturge-Apple and Prof. Zeljko Ignjatovic from Project CONNECT, Dr. Ilker Demirkol from Project GENIUS, Prof. Tolga Soyata, Prof. Minseok Kwon and Dr. Rajani Muraleedharan from Project MOCHA, and Dr. Cristiano Tap- parello from Project GEMCloud. I am grateful to all my brilliant colleagues in the Wireless Communications and Networking Group. Specifically, I would like to acknowledge Ou Yang, Tianqi Wang, Chen-Hsiang Feng, Surija Ray, Bora Karaoglu, Li Chen, Na Yang and Colin Funai for their inspirations and collaborations. They and many other friends have v made my life in Rochester bright, colorful and filled with memories that I will always cherish. Last but not least, I would like to give my special thanks to my dear parents Hong Ba and Xiang Shen. Without their love, encouragement and support, I would not be able to come to the US and achieve my goals. vi Abstract Wireless electronic devices are becoming more and more powerful while re- maining portable and affordable. Some of the smartphones and tablets on the market today are equipped with multi-core CPUs and GPUs and have compara- ble computing capabilities to PCs, along with the improved network bandwidth and connectivity provided by cellular networks. Given this state of current tech- nology, in this dissertation we develop a variety of techniques to enable energy efficient sensing and computing systems. As an example of such a system, consider a personal healthcare system, where wireless sensors are used to gather physio- logical data and send the data to a local cloudlet. The local cloudlet preprocesses the data and transmits the data to a remote cloud server for computation and storage purposes. Data analysis results are sent back to the cloudlet or directly to the user's smartphone for display. One of the challenges in developing such a system is the data acquisition and processing. For applications like emotion classification or personal health, sensed data are not always gathered in clean environments and therefore are often corrupted by noise. Noisy sensing signals must be processed to improve the signal- to-noise ratio (SNR) in order to extract relevant information. For example, in an emotion classification application, speech data may contain babble noise from people talking in the background. In order to extract pitch, which is a key feature in emotion classification algorithms, from noisy speech data, we developed a hybrid pitch detection algorithm named BaNa. The BaNa algorithm combines the idea of using the ratios of harmonic frequencies and the Cepstrum approach to find the pitch from a noisy signal. We tested our BaNa algorithm on real human speech samples corrupted by various types of realistic noise. Evaluation results show the vii high noise resiliency of BaNa compared to other state-of-the-art pitch detection algorithms. The second challenge comes from considering the energy availability. Wireless sensors are usually battery powered and hence have limited lifetime. To extend the battery life of a sensor node, power management approaches have to be utilized. The energy of a node can be saved by putting its radio and other components into sleep mode occasionally. To wake up a sensor node so that it can perform its functionalities, traditionally, a duty cycling approach is used, where an internal timer fires to wake up the sensor node from the sleep state. In this case, the sensor's energy efficiency may suffer from idle listening since it has no knowledge of the channel while sleeping. We created a passive wake-up radio sensor node named WISP-Mote by using a programmable RFID tag as an external wake- up radio for a Tmote Sky sensor node. The wake-up radio reduces the energy wasted on idle listening and hence improves the energy efficiency of the sensor node. We characterized the WISP-Mote's performance by measuring its energy consumption for different operations and assessing its wake-up probabilities in different environments for various WISP-Mote to reader distances. MATLAB simulation results show that the energy efficiency of a sensor network using the WISP-Motes is much greater than when using traditional duty-cycling nodes. Computation of energy efficient sensing and computing systems can be local on the node, or, for more intense applications, computing can be off-loaded to external computing resources, such as cloud-based resources, to save the energy of the node. However, a traditional cloud is composed of powerful but energy- hungry workstations. The growth of the population of mobile devices such as smartphones and tablets provides a huge amount of idle computing power. We describe the design and implementation of a mobile computing system prototype named GEMCloud that utilizes energy efficient mobile devices (e.g., smartphones and tablets) as computing resources. The computing