Lightweight energy efficient source-level adaptive computing framework for sustainable event-driven data transmission

Wireless Sensor Networks (WSNs) are widely used in Internet of Things (IoT) for various monitoring applications in which sensor nodes are equipped with limited battery power. A large amount of energy is wasted in such applications due to frequent data transmission through repeated switching of the transmission using between ON/OFF states. In order to avoid such energy wastages several existing methodologies like batch-wise transmission and sleep scheduling were developed but they are not aware on the importance of the collected data. Though the switching energy is controlled in such methods, its nature of sending all the data to the base station results in an improved energy consumption. In this paper, an Adaptive Activity-driven Transmission Decision (A-ATD) model is proposed for importance aware and energy efficient data transmission. The proposed approach employs a lightweight statistical mechanism at each sensor node, where the sensing behaviour is learned using sliding window based mean and standard deviation estimations. Therefore, an adaptive threshold value is dynamically computed to identify significant deviations that enables selective transmission of only important data from the sensor nodes by suppressing the redundant readings without any training process. A comprehensive energy model incorporating transmission, transceiver switching and computations is formulated and an extensive simulation is done with MATLAB for evaluations. The result demonstrates that the proposed A-ATD model achieves a high suppression ratio of 0.9971 and that results in reducing unnecessary transmission for providing a better network lifetime of 1,89,013 rounds of operation, and that is better over the 38,428 with decision tree approach and 30,438 with the single hidden layer neural network model. Furthermore, the proposed model attains a better precision of 0.8991 with an F1 score of 0.7816 leads to attain a better overall accuracy of 0.9961 in identifying the important and significant readings. These results indicate that the proposed method achieves improved network-level energy efficiency with a lightweight per-sample computational requirement, that is one of the primary requirement for a sustainable and energy aware next generation WSN applications.