This collaborative project investigates Opportunistic Sensing (OS) and Compressive Sensing (CS) in Wireless Sensor Networks (WSNs). OS refers to a paradigm in which a WSN can automatically discover and select sensor modalities and sensors based on an operational scenario, resulting in an adaptive network that automatically finds scenario-dependent, objective-driven opportunities with optimized performance. CS is a novel sensing/sampling paradigm that goes against the common wisdom in data acquisition. Both OS and CS help improve efficient operations and performance of WSNs significantly. In particular, OS aims at reduction from space by selecting a subset of sensors and modalities for efficient data fusion, whereas CS targets reduction in sampling by selecting a subset of samples non-uniformly. Therefore, theoretical foundations and algorithms for opportunistic and compressive sensing are essential for advancing the state of the art in WSNs that not only ensure effective utilization of sensing assets but also provide robust optimal performance. This project addresses fundamental research issues from information theoretic viewpoint to evaluate joint OS and CS distortions, develop OS and collaborative CS schemes for better performance of WSNs, and cross-layer design to adapt to the non-uniform sampling in CS-based WSNs. This project will make a significant contribution to the theory and applications of opportunistic and compressive sensing to WSNs and will have a broad and deep social impact in homeland security and defense. Under-represented and female students from different societies will be recruited, and seminars and in-house short courses will be offered to local industry via IEEE.
The major objective of this project is to develop theory and algorithms for Opportunistic and Compressive Sensing (OCS) in wireless sensor networks that jointly address sensor adaptation, data collection/sampling, and data fusion in constrained environments. The project outcome can be summarized as follows: We have found out that compressiving data gathering is ineffective when outlying sensor readings and broken links exist in a WSN. We demonstrated that the compressiveness of the sensor data is lost in an imperfect WSN. To provide robust compressive data gathering, we developed two compressive sensing based approaches, with one focusing on detecting and recovering outlying sensor readings and the other inferring the broken links. We have identified the possibility of information leakage during data aggregation via statistical inference in compressive data gathering. We showed that if there exists a node compromised by an attacker, the data of the subnetwork controlled by the compromised node could be released to the attacker via the following two statistical attacks: controllable event triggering attack and random event triggering attack. We then designed a new Secure Compressive Data Gathering Scheme to prevent possible statistical inference based attacks during the data gathering process under the existence of compromised nodes. Opportunistic Senisng (OS) aware connected dominating set (CDS) construction is a NP-hard problem. We have found that making CDS construction OS-aware can expand the applications of CDS in WSNs. Therefore we made effort to construct CDSs that consist of connected sensors with higher residual energy, lower traffic load, or higher network connectivity, to facilitate routing and scheduling, data collection, in-network information processing, and network lifetime elongation. We provided a compressive sensing based problem formulation for target counting and localization in sensor networks, and proved that the product of the measurement matrix and the target decay matrix obeys the so-called RIP property with a high probability. This is a significant result as it provides a theoretical foundation of applying CS to target counting. We brought CS to RSNs and demonstrated that a set of stepped-frequency (SF) waveforms can be used as pulse compression codes for transmit sensors, and as the sparse matrix to compress the signal in the receiving sensor. This can significantly decrease the amount of information that should be collected for target detection in RSNs. Two students supported by this project have obtained their PhD degrees. We also have organized a bi-weekly group seminar to help graduate students gain relevant knowledge and train their presentation skills. Our research findings have been published and presented in journals and focused conferences/workshops.
