Using Machine Learning for Adaptive Interference Suppression in Wireless Sensor Networks

It is foreseen that billions of Internet of Things devices will be connected to the Internet in the near future. Most of these devices will communicate wirelessly in a limited spectrum which means that a substantial amount of interferences will be generated. To overcome these interferences, a significant increase in (battery) power consumption is needed to re-transmit packages and to provide a transmit power margin. Hence, this challenge calls for agile methods that can overcome the interferences without wasting power. In this paper, the problem is addressed by using a self-adapting machine learning system which uses information from the channel state to predict the transmit power level that is needed to overcome the interferences. This approach predicts the correct transmit power level when a package needs to be sent and thereby avoids wasting power on selecting a wrong transmit power level, i.e., either excess power is wasted or extra power is wasted to retransmit the packet. Extensive simulations based on data from smart homes show that this approach achieves power savings in the range of 42%-82% and a packet receive a ratio of at least 92%.