| Summary: | 碩士 === 國立臺灣大學 === 電信工程學研究所 === 97 === Sensor networks to collect various information from environments enable deployment and
application of many intelligent devices and systems, such as robots, intelligent vehicles, and even
biomedical instruments. Observing traditional approach separately executing information fusion
from sensor networks, decision, and later control functions, we propose a novel intelligent decision
framework to allow thorough system modeling of such devices, and thus further enhancement
beyond traditional approach. Intelligent decision framework improves traditional estimation theory
by separating the mapping from event to observation into two mappings, the mapping from
observed physical quantity to sensor observation and the mapping from target event to physical
quantity. The mathematical formulation is constructed and applied in the firefighting robot
navigation scenario to illustrate its effectiveness. We further shows that the intelligent decision
framework can be degenerated to traditional decision schemes under special conditions. More
importantly, we can extend the framework to fuse observations from multiple kinds of physical
quantities and derive the optimal decision, beyond traditional statistical decision mechanisms. For
the decision with limited knowledge of the correlations among physical quantities, we propose
Observation Selection and derive the equality condition with optimal decision. While fuzzy logic of
less strict-sense mathematic structure is commonly employed to resolve this application scenario,
we can demonstrate that Observation Selection derived from well-defined decision theory can be degenerated to fuzzy logic of multiple kinds of observations. Finally, simulation results show that
the proposed intelligent decision framework indeed improves the accuracy of the decision and
enhances system performance. In addition to sensor network, this framework can also be applied in
various intelligent system or cognitive systems. We propose a novel cognitive radio spectrum
sensing scheme, Dual-way Time-Division Spectrum Sensing, derived under intelligent decision
framework to demonstrate the application of this general framework other than sensor network.
This scheme mitigates the hidden terminal problem by only one node taking multiple observations
from independent sensing channel, while cooperative spectrum sensing needs multiple nodes to
perform multiple observation. Moreover, this scheme takes the path-loss due to geographical
separation into consideration to improve the sensing performance. Analytical and simulation result
shows that the proposed spectrum sensing scheme significantly improves the performance of
traditional spectrum sensing.
Keywords: Sensor network, information
|
|---|
