Online Interference Mitigation via Learning in Dynamic IoT Environments

A key challenge for ensuring self-organization capabilities in the Internet of things (IoT) is that wireless devices must be able to adapt to the network's unpredictable dynamics. In the lower layers of network design, this means the deployment of highly adaptive protocols capable of supporting large numbers of wireless “things” via intelligent interference mitigation and online power control. In view of this, we propose an exponential learning policy for throughput maximization in time-varying, dynamic IoT environments where interference must be kept at very low levels. The proposed policy is provably capable of adapting quickly and efficiently to changes in the network and relies only on locally available and strictly causal information. Specifically, if the transmission horizon T of a device is known ahead of time, the algorithm under study matches the performance of the best possible fixed policy in hindsight within an error margin of O(T
^-1/2
); otherwise, if the horizon is not known in advance, the algorithm still achieves a O(T
^-1/2
log T) worst-case margin. In practice, our numerical results show that the interference induced by the connected devices can be mitigated effectively and - more importantly - in a highly adaptive, distributed way.