Probabilistic Caching Strategy and TinyML-Based Trajectory Planning in UAV-Assisted Cellular IoT System

Unmanned aerial vehicles (UAVs) deployed as an aerial assisted base station has the characteristics of flexibility and mobility. As an effective way to reduce the communication pressure of network center, content edge caching combined with UAV-assisted network is a promising solution to release the surge of network data traffic pressure. This paper studies the probabilistic caching strategy in UAV-assisted IoT system which supports device-to-device (D2D) communication and edge caching. Firstly, a three-tier heterogeneous model including user devices (UDs), ground small base stations (SBSs) and UAV is proposed. Considering the random characteristics of user movement and the interference characteristics between different nodes, the cache hit probability and successful transmission probability under different content transmission modes are derived by using stochastic geometry. On this basis, the total offloading probability is derived. The joint caching strategy of UD, SBS and UAV is solved with the goal of maximizing cache hit probability and successful offloading probability, respectively. For the mobile deployment of UAV, considering the limited computing resources and battery endurance of UAV, to enable the UAV to provide services to requesting UDs as soon as possible, this paper first uses tiny machine learning (TinyML) to predict the requesting probability of UDs, and then designs a UAV path planning algorithm to cover all users with high requesting probability in the shortest time. Through simulation analysis, we compared the performance of the two proposed caching strategies and found that the strategy of maximizing successful offloading probability has more advantages.