품질 조절이 가능한 센서 데이터의 스케일러블 부호화 분석

Abstract

Machine-generated data such as sensor data now comprise major portion of available information. This thesis addresses two important problems: storing of massive sensor data collection and efficient sensing. We first propose a quality-adjustable sensor data archiving, which compresses entire collection of sensor data efficiently without compromising key features. Considering the data aging aspect of sensor data, we make our archiving scheme capable of controlling data fidelity to exploit less frequent data access of user. This flexibility on quality adjustability leads to more efficient usage of storage space. In order to store data from various sensor types in cost-effective way, we study the optimal storage configuration strategy using analytical models that capture characteristics of our scheme. This strategy helps storing sensor data blocks with the optimal configurations that maximizes data fidelity of various sensor data under given storage space. Next, we consider efficient sensing schemes and propose a quality-adjustable sensing scheme. We adopt compressive sensing (CS) that is well suited for resource-limited sensors because of its low computational complexity. We enhance quality adjustability intrinsic to CS with quantization and especially temporal downsampling. Our sensing architecture provides more rate-distortion operating points than previous schemes, which enables sensors to adapt data quality in more efficient way considering overall performance. Moreover, the proposed temporal downsampling improves coding efficiency that is a drawback of CS. At the same time, the downsampling further reduces computational complexity of sensing devices, along with sparse random matrix. As a result, our quality-adjustable sensing can deliver gains to a wide variety of resource-constrained sensing techniques.