針對物聯網之群集壓縮感知技術

Abstract

壓縮感知對通訊傳輸是一個十分新穎的技術，其技術的特點在於利用大部分的資訊都存在一種稀疏的表示式，經由隨機量測此資訊，即可透過簡單的線性規劃或貪婪式演算法還原此資訊。在新興的物聯網中，如何從眾多的裝置之中迅速且有效率地獲得所需要的資訊為其中之一大課題，本論文著重於針對物聯網內之無線感測網路應用於複雜的物理環境時，提出了一個群集化的壓縮感知技術用於將所收集到部分感測器的資料重建出所有未收集到的感測器的資料以及降低其重建誤差的方法，此方法依據各個感測器所收集的資料以及所在的位置，將其相似性高的資料並且所在位置相互鄰近的感測器分配至同一群集內，再針對各群內的資料進行主成分分析，資料經分析之後可獲得線性轉換矩陣，再配合隨機測量矩陣取得部分感測器的資料，即可完全的重建出全部感測器的資料，除此之外，由於只需要部分感測器傳輸資料，因此這個方法也能夠節省下許多不必要的能量消耗。

Compressive sensing (CS) is an emerging technique for signal processing or image processing. The advantage of compressive sensing is that we can sample a signal of interest below the Nyquist rate and perfectly reconstruct from norm minimization. In this thesis, we apply compressive sensing into wireless sensor network for M2M communications in complex environments. Our proposed methodology is named clustered compressive sensing. Our goal is to recover the signal of unreceived sensor nodes from the signal of received sensor nodes, and furthermore, reduce the reconstruction error by clustering those sensors into clusters according to their data distribution and positions. Next, each clusters use principal component analysis (PCA) to obtain the linear projection matrices which transform the original signal into a sparse representation. Then, choosing active nodes randomly to transmit its data. And finally, recovering the original by norm minimization.