An Adaptive Prediction Framework for Energy Efficient Wireless Sensor Networks
Muruganantham Arunraja, Veluchamy Malathi and Erulappan Sakthivel
In a Wireless Sensor Network (WSN), communication consumes significant portion of energy. Numerous data reduction strategies use adaptive filters that exploit the temporal correlation between sensor data to predict the future values. Thus, only a subset of sensory data is communicated to the sink, the rest is predicted. Prediction filters suffers from inefficiencies, when the filter parameters are inappropriate for the specific data context. Prediction efficiency and data accuracy can be substantially improved by adaptively selecting filter parameters based on data context. In the proposed adaptive step size and adaptive length based nLMS (ASAL-nLMS) filter, we defined the adaptation methods for the prediction filter in WSN scenario. The proposed adaptive filter adapts its filter length based on data dynamics and adapts its step size based on the convergence state of data prediction. The proposed work outperforms previous works in terms of energy efficiency. ASAL-nLMS achieves up to 95% data reduction, with the maximum tolerance of 0.5°C on a real world temperature dataset.
Keywords: Wireless sensor networks, Data reduction, Dual prediction, Adaptive filter, Auto correlation, steepest descent.