考慮多解析度之時間與空間編碼的無線感測網路無失真壓縮

Abstract

在一些要求高精準度量測的無線感測網路(Wireless Sensor Networks)應用中，感測節點必須作長時間感測資料的無失真量測與查詢。然而，感測節點的電力通常有限，為了延長節點的使用時間，如何減少耗電成為許多研究主要考量的重點。 現有方法需要透過無失真資料壓縮，或透過查詢感測區域的概況(rough overview)以降低傳輸資料量。然而，目前並無任何方法有同時考量多解析度之感測資料在時間與空間的相關性作無失真壓縮。因此，本研究針對無線感測網路的省電需求，提出一種考慮多解析度之時間與空間編碼的無失真壓縮，稱之為LMTSC (Lossless Data Compression with Multi-resolution Temporal and Spatial Coding)。 LMTSC將每一點感測資料視為影像的一個畫素，如此可將整體感測資料視為一連串的影像，接著利用畫素在時間以及影像在空間的相關性作無失真資料壓縮，以有效降低感測節點所要傳輸的資料量，進而減少耗電。除此之外，本研究提出動態取樣方法，利用不同取樣率取回資料，並將資料透過無失真壓縮作傳送，提供不同解析度的資料查詢。 本研究利用實際的感測資料評估LMTSC之效能，並將LMTSC與著名的MEC作比較。模擬結果顯示，LMTSC不需像MEC作事前的資料訓練，即可達到較好的資料壓縮率。由於傳輸資料量低，故整體耗電比MEC節省26%。若要維持相同樣取樣點(耗電量)，本研究所提出的動態取樣方法比傳統靜態取樣方法可接受較精準的誤差界限要求；若要維持在相同的誤差界限底下作取樣，動態取樣方法可較傳統靜態取樣方法節省20%以上的電力。由於LMTSC可大幅節省節點的耗電，極適合應用於無線感測網路。

In some WSN (Wireless Sensor Network) applications which require high-accuracy measurements, sensor nodes are used to do long-term lossless measurement and query. However, the limited power makes the power saving become a critical issue of studies. In the existing methods, sensed data can be reduced by lossless data compression or querying rough overview of the sensed area. To the best of our knowledge, since none of previous method takes both temporal and spatial correlation of sensed data into consideration for lossless data compression, they usually cannot obtain a good compression ratio. Therefore, these requirements motivate us to propose the LMTSC (Lossless Data Compression with Multi-resolution Temporal and Spatial Coding) method. LMTSC regards each sensed data as a pixel of an image, and the whole sensed data as sequential images. Using temporal correlation of pixels and spatial correlation of images can reduce the data transmitted and the power consumed efficiently. Besides, we propose a dynamic sampling method which uses various sample rates and lossless data compression to provide different resolution data query. In this paper, we use the real-world sensed data to evaluate LMTSC and make a comparison with MEC. The simulation results reveal that LMTSC has a good compression ratio than MEC without any data training. As the high compression ratio, LMTSC saves 26% power consumption than MEC. For reaching the identical sample point (power consumption), the dynamic sampling method can tolerate a smaller error bound. For reaching the same error bound, the dynamic sampling method can save more than 20% power consumption than static sampling method. Since LMTSC can make a significant power saving, it is very suitable for WSN.