Energy-Efficient Distributed Sparse Signal Retrieval and Event Region Detection: Algorithms and Performance Analysis

• Main Authors: Yang, Ming-Hsun, 楊明勳
• Other Authors: Wu, Jwo-Yuh
• Format: Others
• Language: en_US
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/qjab8z

博士 === 國立交通大學 === 電信工程研究所 === 106 === Distributed signal processing is a core enabling technique for modern wireless sensor networks (WSNs). In this thesis we study distributed sparse signal retrieval and distributed dynamic event region detection for energy-constrained WSNs. To balance energy efficiency and data quality control, we first study sensor censoring for distributed sparse signal recovery under the state-of-the-art compressive sensing framework. In the proposed approach, each senor node employs a sparse sensing vector to infer certain partial knowledge on the unknown signal support during the data acquisition process. The optimal local inference problem is formulated from a detection-theory oriented perspective, leading to a ternary censoring protocol and an associated Neyman-Pearson design criterion. A closed-form formula for the optimal censoring rule, as well as a low-complexity implementation, is derived. To further aid global signal retrieval under the proposed censoring scheme, we propose a modified L1-minimization based signal reconstruction algorithm, which exploits certain sparse nature inherent in the received sensory data model. Analytic performance guarantees, characterized in terms of the restricted isometry property of the sensing matrix, are also derived. In the second part of this thesis, we study the problem of dynamic event region detection via WSNs. By exploiting a space-time (S-T) Markov random field (MRF) model of the sensing field, a distributed two-phase decision fusion scheme is proposed, which involves local sensor decision (in Phase I) and a cooperative S-T decision fusion (in Phase II). The design criterion of the local decision rule in Phase I is to minimize average detection error probability subject to a tolerable communication cost constraint. A closed-form optimal detection rule is derived. Unlike most existing distributed cooperative detection methods, which require real-valued data communication among nodes, our proposed scheme just calls for binary information exchange and, thus, can further reduce the communication overhead. The performance of the proposed scheme is demonstrated through numerical simulations. Computer simulations evidence the efficacy of both the proposed signal estimation and event detection schemes.