大规模无线传感器网络中稀疏信号的数据收集策略

The Data Collection Scheme for Sparse Signals in Large-Scale Wireless Sensor Networks

摘要: 在成功部署的无线传感器网络中，从传感器节点采集的感知数据由于时间和空间相关性会存在大量冗余，所以我们需要设计有效的数据收集方案。其中，压缩感知是常用的一种数据收集技术，它包括了三个核心技术：信号的稀疏表征、观测矩阵的设计、信号重构。本文主要对信号的稀疏性进行了研究。首先，从真实信号和合成信号两个方面对其稀疏性进行了分析。在压缩感知中，不同的正交变换使得信号的稀疏性不同。其次，为了更好地研究信号的稀疏性，我们设计了两种正交变换，即Row-trans变换以及Col-trans变换。通过实验发现，傅里叶变换、离散余弦变换、小波变换的稀疏性能比较差，而Row-trans变换、Col-trans变换的稀疏性更好。就重构误差而言，在1800次实验测量中，信号在Row-trans变换、Col-trans变换下重构误差比较稳定，误差值较小，明显优于其他几种正交变换，得到的重构信号更接近于原始信号。

Abstract: In a successfully deployed wireless sensor network, the sensing data taken directly from the sensor nodes are subject to a large amount of redundancy due to their temporal and spatial correlations. Therefore, we need to design an effective data collection scheme. Compressive sensing is a popular data collection technology. It includes three core topics: sparse representation of signals, design of observation matrix, and signal reconstruction. This paper focuses on the sparsity of signals. Firstly, the sparsity performance is analyzed from two aspects: real signals and synthetic signals. In compressive sensing, different orthogonal transforms make the signal different levels of sparsity. In order to study the sparsity of the signal further, we have designed two orthogonal transforms, namely Row-trans transform and Col-trans transform. It is found through experiments that the sparse performance of Fourier transform, discrete cosine transform and wavelet transform is relatively poor. The sparse performance of Row-trans transform and the Col-trans transform is better. As far as the reconstruction error is concerned, in the 1800 experimental measurements, the reconstruction errors of the signal under the Row-trans transform and the Col-trans transform are relatively stable, and the error rate are low, which is obviously superior to other orthogonal transforms. The reconstructed signal is closer to the original signal.