A hierarchical fractional LMS prediction method for data reduction in a wireless sensor network

A network of sensor nodes forms WSN, where the nodes observe the environment and transfer the sensed data to the sink node. Constraints on various resources, like energy, bandwidth, and memory, are usual in WSN, which the researchers attempt to solve. This paper presents a transmission technique with data reduction using Hierarchical Fractional Least-Mean-Square (HFLMS), in WSN. The proposed HFLMS filter is a prediction method that attempts to predict the sensed data based on an error estimate. The filter design of HFLMS extends Hierarchical Least-Mean-Square (HLMS) by modifying its weight update using Fractional Calculus (FC). The proposed adaptive filter reduces energy constraints in WSN by allowing the sensor nodes to transmit only the required data to the sink. Thus, HFLMS with integrated FC prolongs the lifespan of the network preserving the energy. Two evaluation parameters, energy and prediction error, are utilized to measure the performance of the algorithm. The experimental results performed using two datasets from UCI machine learning, show that HFLMS has better results than the existing without prediction, LMS, and HLMS techniques, with the energy of 0.1202 at 500th round and minimum prediction error of 0.0253. (C) 2018 Published by Elsevier B.V.