부산대학교 도서관

In industrial edge caching networks, to ensure long-term accurate decision making of industrial applications, it is critical to obtain fresh sensing contents with low sensor energy consumption. The acquisition of sensing contents consists of cache update and content delivery, jointly determining the Age of Information (AoI) of applications. However, cache update suffers from the large sensor energy consumption and the mismatch between content offerings and demands. Content delivery suffers from the limited fronthaul capacity. Furthermore, contents from multiple sensors typically need to be aggregated, allowing the AoI of applications to be determined by the co-AoI of all correlated sensors. It is challenging to make the tradeoff between the energy efficiency of each sensor and the co-AoI performance of all correlated sensors. In our work, the weighted sum of application AoI and sensor energy consumption is minimized by jointly optimizing cache update and content delivery, which is formulated as a long-term stochastic optimization problem. Next, two caching schemes, access point centric scheme (APCS) and request adaptive caching scheme (RACS), are presented. In APCS, we fully decouple cache update and content delivery by applying statistical probability of application requests to control update. In RACS, cached contents are updated along with content delivery according to real-time requests. Thus, we introduce the concept of decision reward to transform the stochastic problem into the per-time slot reward maximization problem and propose online algorithms to solve it. Simulation results show that proposed schemes can reduce the sensor energy consumption by 40% while guaranteeing the application AoI.