| Wireless Integrated Network Sensors (WINS) incorporate the latest advances in wireless technology, integrated circuits, networking protocols and computing to provide compact intelligent sensor nodes for many distributed monitoring applications. [5, 13, 14, 36, 85] From battlefield intelligence to personnel monitoring, WINS nodes will play an integral part in national security, the manufacturing industry, transportation and health care. These distributed wireless sensors require signal-processing technology to enable local detection, classification, and identification of events. In this dissertation, a time domain Signal Search Engine (SSE) has been developed for event processing, allowing events to be classified, identified, and communicated with minimum data payload requirements. This SSE was shown to be able to resolve signal time-evolution behavior. This dissertation presents time-domain signal classification and identification algorithms, and the fusion of classified or identified sensor information for distributed and centralized decision-making. One of the highlights of this dissertation includes a modular design using the technique of signal data decomposition into state spaces. The SSE implementation has shown excellent results for acoustic and seismic vehicle signal classification and identification. Furthermore, excellent classification and identification performances are demonstrated for a variety of vehicle signal sources and environments.; The SSE algorithms were benchmarked with two parametric methods, MUSIC and the Pisarenko algorithm, for created test signal models. Further benchmarking with real world signals was done with a wavelet method. It was found that system level processing and functions contained in the signal pre-processing modules is the reason for the high accuracy rates.; The performance of the SSE depends on the assigned template tree and its span. The more complete the span of the tree to sensed signals, the better the accuracy rates and associated confidences measures. The type abstraction hierarchy (TAH) concept is used in building the classification/identification tree. Type abstraction hierarchy is a methodology where similar signals are grouped together within abstract types for selecting a common template. TAH brings a methodical hierarchical classification and identification structure to the SSE. The obtained results are fused for collaborative decision-making with distributed and centralized decision-making architectures. The fusion of these different types (i.e. acoustic, seismic) and state-space sensor results not only enhances the performance and throughput of the SSE or Multi-Signal Search Engines (M-SSE) but also makes it more robust in noisy environments. |
