Data Reduction in Wireless Sensor Networks: A Hierarchical LMS Prediction Approach

In wireless sensor networks (WSNs), due to the restriction of scarce energy, it remains an open challenge how to schedule the data communications between the sensor nodes and the sink to reduce power usage with the aim of maximizing the network lifetime. To face this challenge, this paper proposes a workable data communication scheme utilizing the hierarchical Least-Mean-Square (HLMS) adaptive filter. The HLMS predicting techniques are explored that predict the measured values both at the source and at the sink, sensor nodes are subsequently required only to send those readings that deviate from the prediction by an error budget. Such data reduction strategy achieves significant power savings by reducing the amount of data sent by each node. We discuss the working mechanism of HLMS in the purpose of data reduction in WSNs, analyze the mean-squared error in the two level HLMS, and design the interactive HLMS prediction algorithm implemented at sink and sensor node and the transmission protocol between them. To elaborate on our theoretical proposal, the HLMS algorithms and protocols are then evaluated by simulation. Simulation results show that our proposed scheme achieves major improvement in convergence speed compared with previous approaches, and achieves up to 95% communication reduction for the temperature measurements acquired at Intel Berkeley lab while maintaining a minimal accuracy of 0.3 °C.