The major objective of this project is to study the challenges and techniques of applying the emerging Compressive Sensing (CS) theory to various fundamental sensor network problems under typical network settings. The proposed research activities include: i) temporal and spatial compressive sampling to decrease the transmitted data volume while preserve the information level; ii) compressive data gathering to decrease the communication overhead while preserve high-fidelity data recovery; iii) mission-critical sensor network applications such as outlier detection and target counting to illustrate the CS formulations of the problems and the strength of CS as a technical approach; and iv) testbed and field deployment for validation purpose. These research activities are motivated by the observations that a) sensor networks are typically deployed to measure various natural signals that are usually compressible and are temporally and spatially correlated; b) fundamental sensor network problems such as topology control and in-network data aggregation and compression investigate the compressibility and correlation among sensor readings for resource conservation; and c) CS provides an approach to recover the compressible data by acquiring just the important information via non-adaptive random projections.
The expected results include novel algorithms that can contribute significantly to both CS theory and sensor networking. Our research could motivate a new wave of exploration via sparse signal recovery on a wide range of fundamental sensor network problems that have been investigated through traditional approaches for many years. Research outcomes will be disseminated through high-quality publications as well as presentations in focused workshops and conferences.
Upon the completion of this project, the following important outcomes were achieved: In compressive data gathering, outlying sensors and imperfect wireless links both lead to sparsity violations of the sensor signals, making compressive data gathering completely fail. But outlying sensor readings and broken/lossy links can be identified via compressive sensing too as they are sparse events in the temporal domain. Thus robust compressive data gathering can be achieved to collect high-fidelity sensory data at low communication overhead based on the compressive sensing theory. Two statistical inference attacks, namely controllable event triggering attack and random event triggering attack, are identified in compressive data gathering, and their effectiveness in disclosing the measurement matrix information of the sensor network to the adversary is investigated. To provide secure compressive data gathering, a secure protocol to establish a measurement matrix that varies from time to time is proposed and its superior performance is extensively studied. When point targets in a relatively large deployment area are considered, compressive sensing theory can be applied to accurately count and locate the targets. The target signal decay matrix, which is proved to satisfy the RIP property, is used as the measurement matrix, and a greedy pursuit sparse recovery algorithm, which takes advantage of the fact that only a finite number of targets can be located at any unit grid, is developed and studied. A comparison based performance validation indicates that the proposed compressive sensing based target counting and positioning achieves very high counting and positioning accuracy for both heterogeneous and homogeneous targets. The channel impulse responses of Radar sensor signals (with narrowband 200MHz, 400MHz, and UWB) in a foliage environment are sparse. An amplitude based compressive sensing algorithm is developed to effectively compress the UWB noise radar signal (500-1000MHz). In radar sensor networks, compressive sensing theory can be applied to estimate the target radar cross section and the recovery can be very precise if the number of measurements is no less than the total number of sensors in the network. This project has made contributions to the theory and applications of compressive sensing to wireless sensor networks and its outcomes have a broad and deep impact on big sensory data collection and processing. One female PhD student and a high-school student (summer intern) were recruited to participate in this project. The research outcomes were distributed to the broad community via conference and journal presentations, and invited (focused) talks.
