Entropic Sensing for Energy Efficiency

We present a novel energy-efficient approach to wireless real-time sensing. For a sensor node (SN) transmitting samples of a discrete time series in real-time, its lifetime depends largely on its battery capacity. With most of the energy consumed in wireless transmission, we present an energy efficient scheme that can significantly reduce the number of transmitted samples, while maintaining a low mean absolute error between the original and the recovered signals. We introduce the concept of instantaneous entropy and we derive a computationally efficient iterative formula for computing Shannon’s entropy. The SN evaluates the information content in each sample and decide whether to transmit or omit the sample. At the sink, we use incremental machine learning to recover the omitted samples in real-time. Our approach showed an average of 60% reduction in energy consumption by the SN with less than 2% mean absolute error in the recovered signal.