Abstract: In distributed IoT paradigm, we often have cases where multiple nodes need to keep a common server updated with their status information. However, status being dependent on time, it requires a periodic flow of data-packets (containing fresh status information) from nodes to the server. But due to the decentralized nature of the system, nodes lack de nite information about the network and are unable to coordinate with each other. This results in multiple nodes transmitting data-packets simultaneously, which leads to data-packet collision, thereby incurring unnecessary transmission cost and disrupting timely status updates. In this paper, we consider Age of Information (AoI) as the metric for timeliness of status updates, and propose a learning algorithm based transmission policy for each node to avoid data-packet collision and achieve low time-averaged AoI and average transmission cost. The learning algorithm works at each node independently and assumes no network information, except that upon each transmission, a node gets to know if its data-packet is received by the server or dropped due to collision. Based on the history of transmissions and their corresponding result (transmission success/collision), the learning algorithm suggests a transmission probability to the node for the next time-slot. We prove that the transmission probability so obtained, converges to a unique fixed point, and further analyse its properties using numerical simulations.
The work has been jointly done with Rahul Vaze, TIFR.
