Re-Planning Versus Learning: Fair Scheduling for Underwater Acoustic Networks Under Model Shift

In this paper, space-time re-planning for multi-user underwater acoustic network (UAN) scheduling under fairness considerations is investigated. As in our previous work, UAN scheduling is conducted in both time and spatial domains such that the resulting policy is a function of the users' location. The existence of mobile users in underwater networks necessitates timely policy updates to combat model-shifts. However, the computationally heavy policy re-planning process should be conducted as infrequently as possible. To this end, a novel re-planning scheme is proposed to optimize the timings of policy updates such that re-planning is only conducted when necessary. Model information is exploited to provide improved learning. Numerical results show that the proposed scheme is able to recover most of the capacity and fairness loss, with much less computation time, compared to current approaches.