AoI-aware coexistence scheduling for industrial wireless sensor networks under uncontrollable external networks via deep reinforcement learning

Industrial wireless sensor networks (IWSNs) are increasingly deployed in shared-spectrum environments, where coexistence with external wireless networks is often unavoidable. Existing coexistence scheduling methods usually rely on joint coordination frameworks that assume all coexisting networks are observable and manageable. In practice, however, some external networks are independently operated and uncontrollable, while their internal access rules and future transmission behaviors are unavailable to the target IWSN. This paper investigates an age-of-information (AoI)-aware coexistence scheduling problem for an IWSN sharing spectrum with these external networks. The objective is to improve coexistence efficiency while maintaining information freshness at the gateway. First, the sequential scheduling process is formulated as a Markov decision process (MDP), in which channel access decisions are made according to historical action-observation information and node-wise AoI states. Then, an AoI-oriented reward is developed to account for both the waiting time of the delivered packet and the network-wide AoI variation resulting from a successful transmission. Based on this formulation, a deep reinforcement learning (DRL)-based coexistence scheduling (DRCS) algorithm is developed by combining the reward design with a bidirectional gated recurrent unit (Bi-GRU)-enhanced deep Q-network. Comparative simulation results under coexistence with both time division multiple access (TDMA)-based and q -ALOHA-based external networks show that DRCS effectively utilizes available channel resources and reduces transmission conflicts while maintaining information freshness under different interference conditions.