【作者】 张颖；

【导师】 孙桂玲；

【作者基本信息】 南开大学 ， 物理电子学， 2012， 博士

【摘要】 无线传感器网络作为物联网的关键技术之一，在国防、工业、农业、商业、学术等社会各界，都给予了更加广泛的关注，其宗旨，就是以最少的成本和最大的灵活性，解决物联网中的信息感知问题，因此，大规模高密度的网络拓展、海量数据传输和存储、能源有限等关键技术瓶颈的解决迫在眉睫。而压缩传感理论在信息测量传输和重构方面的创新思路，为解决无线传感器网络探索新的信号获取技术及信号处理方式提供了一种全新的技术手段。基于压缩传感这一新兴的研究理论，本文针对无线传感器网络的信息采集、传输和处理技术展开研究：（1）建立CS-based无线传感器网络模型，研究压缩传感理论对无线传感器网络的贡献，使无线传感器网络更具有普适性；（2）分析自然界各种信号的稀疏性，构建了一个更加自适应的稀疏分解平台，有助于无线监测更精确的重建目标信号；（3）研究并梳理了压缩传感理论的重构算法，建立一个新的、更适用于无线传感器网络稀疏度未知的信号重构方法；（4）分析CS-based无线传感器网络在网络全局能耗、负载、传输方式等方面的优势，提出硬件能力更容易实现的传感测量机制。其创新性的研究工作及成果如下：（1）对自然界信号的稀疏性进行研究，针对时域、频域非稀疏信号的稀疏域进行探索，考虑硬件允许的情况下，利用Gabor原子和Chirplet原子分别生成过完备原子库，并结合自适应匹配追踪的思路设计相应的二次寻优算法，使得绝大多数的可压缩信号都可以在稀疏字典中找到一个合适的稀疏基底，进行稀疏表示。研究结果表明，与传统的稀疏分解方法相比，Gabor稀疏字典和Chirplet稀疏字典更具有普适性，稀疏度更强，稀疏表达精度更高，绝大多数的信号都能够获得更好的稀疏效果。此外，Gabor字典二次寻优匹配算法和Chirplet字典二次寻优匹配算法能够使稀疏分解的速度更快，占用的存储空间更少，同时也更有助于重构精度的提升。（2）分析现有的重构算法，对比重构精度、速度的优劣性，并从迭代方式、迭代终止条件、运算复杂度等方面进行全面优化和改进，提出一种更适合大规模无线传感器网络的数据重构算法，回溯自适应阈值迭代匹配追踪算法（BATIMP），在满足RIP约束条件的基础上，速度更快、精度更高、稳定性更强，可以高概率的重构原始信号。研究结果表明，在CS-based无线传感器网络中，能够更加适合自然界中可压缩的各种稀疏度未知的信号，自适应的选择迭代步长，以节省时间提高重构速度，与传统的重构方式相比，在一定程度上降低了对重构端的硬件要求。（3）建立分布式的压缩传感网络模型，提出空间稀疏信号网络模型，研究CS-based无线传感器网络的数据采集和测量机制。针对无线传感器网络的拓扑设计和路由方式，将压缩传感的测量投影过程引入无线传感器网络的信号采集和传输过程中，设计高斯循环、贝努力循环两种半随机测量矩阵，使得CS-based无线传感器网络在感知信息的过程中，就能将信息进行投影测量，在多跳传输的过程中，将网络全局信息直接降维，极大程度的提高了传输带宽利用率，也能使网络负载更加均衡，使多跳传输的无线传感器网络能够真正实现。对比传统的网络传输方式，分析CS-based无线传感器网络在网络负载均衡、传输能力、信号重构等方面的效果，从网络整体的角度，验证不同空间稀疏的信号压缩传感机制的性能。研究结果表明，与传统的无线传感器网络相比，CS-based无线传感器网络无论在局部、全局能耗和负载均衡等方面都有明显优势，并且在网络拓扑的选择方面具有自由性。在CS-based无线传感器网络中实现半随机循环测量传感，既能保持随机测量矩阵与绝大部分的稀疏矩阵不相关的优势，又能使网络节点的设计和部署更加简单，只需要根据网络分配的种子和节点ID产生相应的测量系数，比传统的随机矩阵生成方式简单，更适合存储空间有限的无线传感器网络，同时，网络初始化简单，更适合网络的实际应用需求。总之，压缩传感理论的引入，对本文研究的无线传感器网络提出了测量传输的新理念，为数据的精确重构也提供了更高质量的重建效果，更加适合自然界的各种信号，对应用对象具有“自适应性”。更多还原

【Abstract】 With the conception of the Internet of the thing (IOT) coming up, China makes itthe direction of guiding the whole information industry chain development, to realizethe boom of enconomy. As one of the key technologies, Wireless Sensor Networks(WSN) has been more widespread concerned on solving the information sensingproblems by minimum cost and maximum flexibility. Therefore, large-scalehigh-density network expansion, huge amounts of data transfer and storage, energylimited and other key technology bottleneckes to address immediate. Compressivesensing theory (CS) has innovative ideas on data measuring transmission andreconstruction, and provides a new technical means to solve the wireless sensornetwork to explore new signal acquisition technology and signal processing.Based on CS, we explored research on wireless sensor networks informationcollection, transmission and processing technologies to study the whole networkperformance:(1) estimated CS-based wireless sensor network model, do research onthe contribution of the compressed sensing theory for wireless sensor networks,makes wireless sensor networks more universal;(2) analyze the sparsity of the naturalsignals, built a more adaptable sparse decomposition platform helps wirlessmonitoring more accurate in the reconstruction of the target signals;(3) research andsort out the reconstruction algorithm for compressed sensing theory, establish a new,more suitale for wireless sensor networks unknown sparsity of the signalreconstruction method;(4) analyze the CS-based wireless sensor networksperformation in network global energy consumption, load, transmission and otheraspects advantages, proposed easier implement sensing and measureing mechanism.The exhaustive research work and achievement are as follows:(1) Study the sparsity of the signals in the natural time domain, frequencydomain non-sparse signal sparse domain to explore, considering the limited hardware,make use of Gabor atoms and Chirplet atoms in generating complete dictionary andthe corresponding twice optimization algorithm, making any sparse signal can find a sparse basal sparse representation in the sparse dictionary. The study results show thatcompared to the sparse decomposition method, the Gabor sparse dictionary andChirplet sparse dictionary are more universal sparseness to achieve lower sufficiently,higher sparse precision, the vast majority of signals are able to get a better sparseeffect. In addition, the Gabor dictionary twice optimal matching algorithm and theChirplet dictionary twice optimal matching algorithm can make the sparsedecomposition faster and occupy less storage space, but also more conducive to theenhancement of the reconstruction accuracy.(2) Analyzed the existing reconstruction algorithms, compare reconstructionaccuracy, speed, and iterative method, the iteration termination condition, thecomputational complexity and other aspects of comprehensive optimization andimprovement, and propose a more suitable for large-scale wireless sensor networks,data reconstruction algorithm, backtracking adaptive threshold iterative matchingchasing algorithm (BATIMP), based on the RIP constraints, realized faster, higheraccuracy, greater stability, high probability reconstruction of the original signal. Theresults show that the CS-based wireless sensor networks, more suitable for the natureof a variety of sparse unknown but compressible signal, adaptively select the step size,to save time and improve the reconstruction speed with traditional reconstructioncompared to a certain extent reduces the hardware requirements on the reconstructionside, making a reconstruction algorithm that compressed sensing theory to computerhardware are able to achieve.(3) Estimated the distributed compression of sensor network model, proposedthe space sparse signal network model, research on the CS-based wireless sensornetwork data acquisition and measurement mechanisms. Depending on topology androuting for wireless sensor networks, involved compressed sensing into the wirelesssensor networks measuring projection, signal acquisition and transmission, design aGaussian loop/Bernoulli loop semi-random measurement matrix, making theCS-based wireless sensor networks project number of information in sensing process,during multi-hop transmission, it will be able to not only reduce the dimensions ofglobal network information, the great improvement in the level of bandwidthutilization, but also make network load more balanced and transmission of multi-hop wireless sensor network is able to realize. The results show that compared withtraditional wireless sensor networks, the CS-based wireless sensor networks in termsof local and global energy consumption and load balancing can have a high degree ofoptimization, and has the freedom in the choice of network topology. Thesemi-random cycle measurement sensor in the CS-based wireless sensor networks, itcan keep the random measurement matrix and most of the sparse matrix-relatedadvantages, but also make more simple the design and deployment of the networknodes, only corresponding coefficient of the measurement according to the networkdistribution of seeds and node ID than the traditional random matrix generated ismore suitable for wireless sensor networks with limited storage space, at the sametime, the network initialization, more suitable for practical applications of thenetwork.In short, involving of compressive sensing theory,we introduce the new conceptof measuring the transmission for wireless sensor networks, the exact reconstructionof the data also provides a higher quality of reconstruction effect, which is moresuitable for a variety of the natural signals, has the "adaptive".更多还原