Maximizing Decision-Making Time Overhead-Aware Computation Rate in Wireless-Powered Mobile Edge Computing

Wireless powered mobile edge computing (WPMEC) has emerged as a promising paradigm to mitigate the energy and computation limitations of small-scale Internet-of-Things (IoT) devices. Typical WPMEC operates in a discrete-time manner: at the beginning of each time block, a resource optimization algorithm determines system configuration parameters, and the remainder time block is exploited to perform computation and communication tasks based on optimized settings. Although various decision-making algorithms have been proposed for WPMECs, the associated computational time overhead is often overlooked in existing studies. However, the execution time of decision-making algorithms occupies a non-negligible portion of time in practice, particularly in fast fading channels. In view of this, we define a unified performance evaluation metric termed decision-making time overhead-aware computation rate (DTO-ComRate), which jointly accounts for decision accuracy and algorithmic complexity. Then, we first formulate an optimization problem to maximize the weighted sum DTO-ComRate by jointly optimizing offloading and channel resources. To efficiently address this problem, we further propose a candidate solution set (CSS)-guided deep learning (CSS-DL) algorithm, in which linear relaxation is leveraged to construct the CSS, and a deep neural network (DNN) is trained to assign channel resources in an online self-learning manner. Experiments show that CSS-DL significantly outperforms baseline algorithms and maintains robust performance as the number of IoT devices increases, demonstrating its scalability to large-scale wireless systems.