Research On The Key Technologies Of Compressed Sensing In Wireless Sensor Network

As the core technology of the Internet of things, wireless sensor network attracts more and more attention for its flexibility and the information sensing ability. To en-sure the flexibility, the hardware and energy supply is restricted, and brings great stress to huge amounts of data acquisition, transfer and storage. It has been a problem to address immediate for large-scale high-density network expansion to break through the bottleneck.In recent years, the theory of compressed sensing has been widespread con-cerned and researched. Compressed sensing merges the sampling and compression, discards the "redundant" components of a sparse or compressible signal, and saves the storage and transmission cost. Compressive sensing theory (CS) has innovative ideas on data measuring transmission and reconstruction, and provides a new tech-nical means to solve the wireless sensor network to explore new signal acquisition technology and signal processing.Based on the analysis of the WSN data characteristics, we explored mul-ti-faceted research on the key technologies of CS to apply the theory of CS to WSN:1. Study the sparsity of signal:the natural sparse signals in time domain are rare, but most of them can be sparsely represented in a transform domain. On the basis of in-depth research on the WSN data characteristics, considering the limited hardware, make use of DCT atoms, Haar atoms, Chirplet atoms and Db atoms in generating over complete dictionary, making any sparse signal can find a sparse basal sparse representation in the sparse dictionary. The simulation result shows that:the Chirplet and Db over complete dictionary are more universal sparseness to achieve lower suf-ficiently, higher sparse precision, than the other two, the vast majority of signals are able to get a better sparse effect. In addition, we proposed the multi-parameter over complete learning dictionary, for the signal similar to the prior model, it can make the sparse decomposition faster and occupy less storage space, but also more conducive to the en-hancement of the reconstruction accuracy.2. Study the measurement matrix:based on the condition of satisfies RIP, gener- ate the Gaussian random matrix, Bernoulli random matrix, Toeplitz matrix and Cyclic matrix, and then the advantage and disadvantage. In addition, put forward a kind of Bernoulli pseudo random cyclic matrix, it is easily implemented in hardware, and save the storage space. The simulation shows that when the measurement number M satisfies some conditions, high precision measurement and signal reconstruction can be realized.3. Signal reconstruction:analyzed the existing reconstruction algorithms, com-pare reconstruction speed, accuracy, computational complexity and other aspects of comprehensive optimization and improvement, and put forward a more suitable for large-scale wireless sensor network, data reconstruction algorithm, sparse adaptive threshold iterative orthogonal matching pursuit algorithm, based on the RIP constraint, realized the faster, higher accuracy, high probability reconstruction, greater stability, better noise robustness of the original signal. The result show the algorithm is more suitable for nature of a variety of sparse unknown but compressible signal, and it will speed up the application of the CS theory in wireless sensor network.4. Compressed sensing combine with periodic non-uniform sampling:with the help of multi channel sampling system for signal sampling, periodic non-uniform sampling is an effective method of reducing the sampling frequency and improving the sampling accuracy. Based on the characteristics of periodic non-uniform sampling, combine with the theory of joint subspace, the sampling and reconstruction process can be converted into vector operations. With the help of the CS theory, apply the sparse signal reconstruction algorithm to the periodic non-uniform sampling system to reconstruct the origin signal, the result shows that the system can achieve the signal sampling and reconstruction process, and greatly reduce the sampling frequency.