An Efficient Accelerated Learning Algorithm For Tracking Of Unknown, Spatially Correlated Signals In Ad-Hoc Wireless Sensor Networks

An efficient accelerated learning algorithm is proposed and discussed for tracking of spatially correlated signals in ad-hoc wireless sensor networks. The proposed algorithm is low-cost and computationally efficient. It models an unknown, spatially correlated signal using a number of its contour lines at equally spaced levels. In the proposed algorithm, each sensor is modeled as one neuron in a neural network. The accelerated learning's agent is implemented at the fusion center (FC). The algorithm is performed in two phases of spatial modeling and spatial tracking. In spatial modeling phase that accelerated learning is implemented, the algorithm discovers the model parameters. In spatial tracking phase, the model parameters are updated to track the varying, unknown spatial signal. Those sensors (neurons) that their observation are in a given margin of at least one of the contour levels, report their filtered observations to the FC. The FC updates the model parameters based on the reported observations and returns the model features to the sensor network for the next iteration step. The performance evaluation results show that the proposed accelerated learning is low cost and converges faster than single layer machine learning approach. The modeling performance, convergence speed and the cost of the proposed algorithm are compared with those of single layer machine learning algorithm. The algorithm is proposed for environmental monitoring.