Research On Information Processing And Transmission Technologies Based On Compressive Sensing In Wireless Sensor Networks

Data-centric is an important feature of Wireless Sensor Networks(WSNs). Such networks are mainly designed to sense a ?eld of interest, process sensed data, and transport them to data management center with as low cost as possible. However, due to the large number of nodes, high-density deployment, and limit resource of nodes including low power, short-distance communications, low computation ability and storage capacity, how to e?ciently process and transmit these data is still a key issue in the research ?eld of WSNs.Considering the above unique characteristics of WSNs and the disadvantages of the traditional methods for sensing information processing and transmission, we develop new technologies based on compressive sensing for data processing and transmission in this thesis. The proposed methods not only simplify the complexity of information processing, thus reducing the requirement of computation resource of nodes, but also overcome the asymmetry of information processing in the traditional compression algorithms. In this thesis, according to different application scenarios, with the goals of improving network capacity, reducing transmission delay and energy consumption,we research on novel information processing and transmission mechanisms, with the emphasis on sampling manner, routing protocol, scheduling strategy and performance analysis. The main contributions of the thesis are summarized as follows:1. Data gathering with compressive sensing in large-scale wireless sensor networksWe investigate the performance in terms of network capacity and transmission delay for the single-sink and multiple-sink networks by introducing the theory of compressive sensing into data gathering application in large-scare wireless sensor networks. In the single-sink networks, we analyze the upper bound of network capacity for data gathering network under the framework of compressive sensing and propose routing and scheduling schemes to achieve optimal low bound of network capacity. We also analyze the performance of transmission delay for the proposed scheme. In the multi-sink networks, by introducing the theory of sparse random projections, we proposed a multi-session data gathering method based on compressive sensing. We analyze the upper bound of network capacity for the multi-sink networks, and then propose the corresponding routing and scheduling schemes by constructing a multi-session spanning tree to achieve the optimal low bound of network capacity. We also analyze the performance of transmission delay for the proposed scheme. Theoretical analysis and simulation results show that compressive sensing is able to signi?cantly improve network capacity and reduce transmission delay in large-scale data gathering networks.2. In-network computation with compressive sensingData transmission based on compressive sensing converts data forwarding into data computation to reduce transmission cost. Therefore, how to characterize the gain of energy saving that compressive sensing brings for data gathering is a signi?cant research topic. In this thesis, we construct a multiround random linear function to e?ciently compute random projections over WSNs so as to evaluate the performance in terms of energy consumption and latency from the perspective of in-network computation. We propose two e?cient computation protocols including tree-based and gossip-based protocols for in-network function computation. In the tree-based computation protocol, we propose routing and scheduling schemes under the optimal computation refresh rate and block computation protocol to further improve the computation performance. In the gossip-based protocol, we propose a broadcast gossip algorithm, which provides a more robust approach to combat with the failure of nodes and links. Theoretical analysis and simulation results show that the proposed computation protocols with compressive sensing are able to reduce energy consumption and latency.3. Data Gathering with compressive sensing and random walksIn this thesis, we propose a random walk based approach with compressive sensing for data gathering in WSNs. This is the ?rst work to provide mathematical foundations to study the feasibility of such an approach from the perspectives of graph theory, Markov chain theory and CS theory. We obtain some important parameters that how many independent random walks and what is the length of each random walk are required for data gathering and also analyze the guarantee that a k-sparse signal can be recovered using ?1minimization decoding algorithm. The proposed scheme relaxes the constraint of uniform sampling in the traditional CS theory and provides a more practical approach for the data gathering application in WSNs. Comparing with the schemes based on traditional CS theory, the proposed scheme has the advantage of small memory requirement,low computation complexity and low transmission cost.