A Convolutional Operation-Based Online Computation Offloading Approach in Wireless Powered Multi-Access Edge Computing Networks

Prompt, uninterrupted, accurate and reliable agricultural information plays an important role for agricultural decision making, which requires stable and efficient data transmission frame. Wireless powered multi-access edge computing (MEC) has recently emerged as a promising paradigm to improve the capability of data transmission with low-power network. Applying wireless powered MEC to agricultural information monitoring will benefit the development of smart agriculture. Management and scheduling of different applications (i.e., computing offloading) is one of the most important influencing factors for the performance of wireless powered MEC network. Of these, the mutual interferences of different WDs are an important factor which should be considered. To address this issue, an online computation offloading method based on convolutional operation is presented in this paper. And a fundamental wireless powered MEC network including one access point (AP) and multiple WDs is constructed to validate the efficacy of this approach. Impacts of three factors (including network size (i.e., the number of WDs), training intervals and memory size) and different application scenarios are studied and their influences on the performance of convolutional operation-based approach are analyzed. Additionally, convolutional operation-based method is compared with the other two offloading approaches (including DROO (a Deep Reinforcement learning-based Online Offloading) and CD (Coordinate Descent) algorithm). The results indicate that the convolutional operation-based approach is more suitable for large-scale wireless powered MEC networks (e.g., the number of WDs is more than 30) with moderate memory size (=512) and training interval (=50).