AoI Optimal Dynamic Power Control for IoT networks: A DRDPG Approach

Information freshness is crucial for time-critical IoT applications, since stale information will significantly affect the reliability and accuracy of decisions. This paper investigates the dynamic power control policy for IoT networks to optimize the information freshness, in terms of age of information (AoI), where the status update arrivals at source nodes (SNs) are stochastic. To reduce signaling overheads, the BS can observe the information freshness of the update packets stored at SNs only when the packets have been successfully delivered. To this end, we first formulate the AoI optimal dynamic power control problem as a Partially Observable Markov Decision Process (POMDP) with the continuous action space, where neither the instantaneous AoI at SNs nor channel power gains are known to the BS at the time of decision making. Then, we have proposed a deep recurrent deterministic policy gradient (DRDPG)-based power control (DRPC) algorithm to minimize the weighted average AoI at the BS. Finally, simulations are conducted to validate the effectiveness of our proposed algorithm by comparing it with baseline algorithms.