On Norm Selection Effect In Energy Efficient Modeling of Correlated Spatial Signals Using Machine Learning in Wireless Sensor Field
Hadi Alasti · 2021
Machine learning (ML) algorithms are proper solutions for identification and tracking of spatial signals. In this paper, the effect of norm selection on performance and cost of modeling and tracking of spatial signals from sensor observations in sensor fields is discussed. The spatial signal is compressed into a number of its contour levels and the sensors with sensor observations within a given margin of these contour levels report their sensor observations to the data fusion center (FC) for spatial signal modeling. A machine learning algorithm that runs on FC determines the contour level’s margin, in an iterative process. It is assumed that the sensors are distributed randomly over the sensor field, and sensor observation is polluted by additive white Gaussian noise. The applied machine learning algorithm in this paper, uses a stochastic gradient method. During the iteration steps of the algorithm, the number of contour levels increases for finer spatial signal modeling. The stochastic gradient uses the iteration error in spatial signal modeling as its convergence measure. The spatial modeling performance and tracking cost of the algorithm with norm-1 and norm-2 are compared in calculation of iteration error.