Implementation of Computation-Efficient Sensor Network for Kalman Filter-based Intelligent Position-Aware Application

Most edge nodes operate under limited power conditions for reasons of unstable power supply or small battery capacity. Therefore, computations running on edge nodes need to be lightweight. There are extensive usages of the Kalman filter, this paper presents the implementation of a computation-efficient IoT sensor network which tracking the moving target with the Kalman filter. Rather than improving the Kalman filter itself, the Kalman filter operation performed at each node varies performance based on the information being sensed. The nodes are decreasing their data receiving interval according to the distance between the target and each node. Also, we use an arithmetic approximation with changing precision when receiving the target information. The presented computation-efficient sensor network evaluated under MATLAB simulation environment to prove its lower computation amount. Evaluation showed that the computation time at each node was reduced by 17.7% on average compared to the conventional sensor network followed by nodes with a fixed behavior. Besides, despite the cost of arithmetic approximation, the edge node estimates the velocity of the target with reasonable accuracy. Based on this result, this paper suggests that decreasing the computation load of whole sensor network by varying each node's calculation with a macro perspective.