Effective fusion of distorted multi-sensor data

A framework for the detection of bandlimited signals by intelligently fusing the multi-nonlinear sensor data is developed. Though most sensors used are assumed to be linear, none of them individually or in series give the truly linear relationship and errors are inevitable as a result of the assumption of linearity. A new approach, which takes the actual nonlinear characteristics of sensors into account is advocated. Though the fusion of redundant information can reduce overall uncertainty and thus serves to increase the accuracy of the process measurements, identifying the faulty readings and fusing only the reliable data are very difficult and challenging. The main idea of the multi-sensor fusion scheme proposed in this paper is to pick only the reliable data for the fusion and disregard the rest. This is done by assigning a "confident measure' to all available sensor data and picking the ones that lead the list of confidence measures. The result is then used to solve the sensor scheduling problem. The proposed theoretical framework is supported by illustrative examples and simulation data.