Investigation On Data Acquisition Technologies In Sensor Networks Based On Compressed Sensing

Wireless Sensor Network is a new network which combines data acquisition, transition, and process in one. It can complete automatic information acquisition tasks in a large region by a lot of tiny, intelligent and low-cost nodes. All of the applications in Wireless Sensor Network are based on data acquisition which includes three phases: local sampling by sensing nodes, data upload through wireless channels and information reconstruction in the fusion center. Wireless Sensor Network has some disadvantages such as limited energy, weak computing ability, weak storage capacity, large volume of data and high redundancy. Thus, compressed sensing is introduced in this thesis to complete data acquisition with low energy consumption. Compressed sensing technology compresses the original datas into lower dimensional because of the inner-node and inter-node correlation, and reconstructs the datas with a high probability in the fusion center. It can retain effective information of the original datas and reduce data quantity and communication cost, which means prolonging network lifetime. However the introduction of compressed sensing brings new problems, mainly including three problems: how to find signal sparse representations, how to construct observation matrix and how to choose a high-accuracy and low-delay reconstruction algorithm. The problems need to be solved during the process of local sampling, data upload and information reconstruction. Therefore, this thesis focuses on compressed sensing in Wireless Sensor Network, and proposes solutions to achieve the target of improving data acquisition performance and prolonging network lifetime. The main contributions in this thesis are as follows:1.Considering the situation that the inner-node temporal correlation is strong, an adaptive compressed sensing algorithm based on Kalman prediction is proposed. Because of the strong temporal correlation, each sensing node can predict the local sample by Kalman filter, and decide whether to be active according to the difference between the prediction and sample. The number of active nodes decreases so that it is sparse in the spatial domain and the energy consumption of the sensing nodes is reduced at the same time. The Gauss channels between sensing nodes and relay nodes are used to make observing matrix, and the mixed datas received by relays are the compressed observation of the datas of the active nodes. Moreover, a sequential reconstruction algorithm which chooses the number of relays adaptively is proposed in the fusion center. It receives the datas of relays step by step and tries to recover original datas until it succeeds, which can reduce the energy consumption of the relays. Simulation results demonstrate that, compared with other compressed sensing algorithm based on prediction, the calculation complexity of the sensing nodes and the energy consumption of the relays of the proposed algorithm are much lower without any additional error.2.Concerning the situation that the inter-node spatial correlation is strong, a compressed sensing algorithm based on random projection with unequal probabilities is proposed. Because of the strong spatial correlation, the signals sampled at the same time by all the nodes are sparse on wavelet bases or spatial Fourier bases. Selecting nodes randomly to make the sparse measurement matrix can reduce the energy consumption of the observation nodes effectively. Random projection with unequal probabilities requires that each node chooses its probability of transmission according to its signal intensity. It revises the probability with an energy balance strategy and sends its observation to the fusion center with the revised probability when its upload slot comes. Simulation results demonstrate that compared with random projection with equal probability, the algorithm proposed can lead lower recovery errors in the whole fields and especially the interesting fields without extra energy consumption. Moreover, the introduction of energy balance strategy can avoid that some nodes send datas with higher probabilities for a long time, which means prolonging network lifetime.3.Aiming at the situation that the observed signals are in line with joint sparsity models, a compressed sensing algorithm based on distributed simultaneous reconstruction is proposed. Sometimes there is no need to obtain all the datas in the fusion center. It just needs the linear fusion of the datas of all the nodes. Because the observed signals are in line with JSM-1, distributed reconstruction algorithm can reduce the dimensional of compressed datas for reconstruction. To reduce the reconstruction complexity, sensing nodes compress the original datas locally with the same observation matrix. The fusion center combines fusion and reconstruction and completes four steps: dividing nodes into groups, linear fusion in each group, simultaneous reconstruction and summing reconstructions. The linear fusion of the original datas can be obtained through only one simultaneous reconstruction. Simulation results demonstrate that the algorithm proposed can reduce the reconstruction complexity and delay in the fusion center with no need to increase the dimensional of compressed datas.