Research On Compressive Sampling And Data Reconstruction Methods In Wireless Sensor Network

Wireless sensor network(WSN) as a kind of brand-new information acquisition technology, has been widely used in environmental monitoring, fire detection, infrastructure detection, precision agriculture, etc. But because of the limitation of node’s energy and storage space which constrained the practical application of wireless sensor network in the long-term and a widely range target monitoring. Compression sampling which implemented in the process of data sampling is an effective means to solve the problem of wireless sensor network energy limited. Therefore, this paper studies from the compressive sampling and data reconstruction, and put forward the compressive sampling method based on cyclic reshuffle and data recovery, the non-parametric Bayesian data interpolation method under missing data compressive sampling. The main research works are as follows:1) The existing compressed sensing based compressive sampling method in wireless sensor networks usually assume that the sensed data are sparse or compressible. However, the sparsity of raw sensed data in some case is not straightforward. In this paper, we present compressive sampling method based on cyclic reshuffle to achieve energy efficiency in WSNs.By incorporating CS into the cluster protocol, the method is able to reduce the energy consumption and support larger networks. Moreover, the sparsity of raw sensed data can be greatly improved by reshuffling pretreatment. A theoretical analysis to energy consumption of cluster head is performed, and the cost of the pretreatment is small enough to be neglected.Based on these natures, the raw sensed data can be recovered from fewer samples. Also,considering the sensed data to be of excellent temporal stability in a short time, we reshuffle them just one time in this stable period to further reduce the energy consumption of WSNs. In addition, we have demonstrated the theoretical analysis of the energy consumption and delay in detail when adopting the method we proposed. We carry out simulations on real sensor datasets. The results show that the method we proposed can effectively compress the data transmission and decrease energy consumption of WSNs while ensuring the reconstruction accuracy.2) The existing data interpolation method based on sparse representation usually assume that the signal are sparse under the orthogonal basis or over-complete dictionary. However,due to the dictionary which have fixed structure lack of adaptability, it is hard to get the optimal representation of signal which sparse is dynamic change. In this paper, we proposed a data interpolation method based on non-parametric Bayesian. On the basis of combina with sparse representation theory and nonparametric bayesian learning, adopting non-parametricBayesian learning to achieve the optimal sparse representation of signal which sparse signal is dynamic change. On the basis of optimal sparse representation and considering the data interpolation can be equivalent to a sparse sampling problem, a non-parametric Bayesian model with observation, missing data and prior is set up. We can get all model parameters including the optimal dictionary and sparse coefficients by model learning. Using the optimal dictionary and sparse coefficients, the missing data can be recovered. We carry out simulations on real soil moisture data and temperature data, and the results show that the data interpolation method we proposed can recovery the data accurately.Based on the problem of energy constrained and energy consumption imbalance in wireless sensor networks(WSNs), we resolve the problem of energy limited in WSNs from the aspects of compression sampling. Improve the signal sparse by sorting to periodically. By sorting data periodically, the sparsity of sensor data can be greatly improved. Thus it can obtain an efficient compression sampling method which can reduce and balance network energy consumption and improve the network life cycle. However, compressed sampling will produce information loss inevitably.Using non-parametric Bayesian learning without needing to set the the specific distribution of signal in advance, it can learn the change of signal sparsity adaptively and realize the optimal sparse representation. Due to the data recovery and the sparse representation is a dual process, using the non-parametric Bayesian data interpolation method can recover the lost data and solve the contradiction between energy consumption and high data accuracy.