WGAN-Based Oversampling for QoS-Aware M2M Network Power Allocation

Internet of Things (IoT) in the new era of 5G and 6G. Pursuing an autonomous strategy for optimal power allocation remains a significant research topic in M2M communication networks. Recent advances in machine learning offer a viable solution to this challenge. However, the main issue in the power allocation problem when adopting machine learning is the lack of real datasets in the fluctuating network environment. This study delves into the oversampling of the grid-like structure of channel information in M2M communication by leveraging the WGAN. We conducted extensive experiments to assess the benefits of oversampling of the dataset from the simulation environment. Results indicate that the generated dataset is good enough to improve the performance of the machine learning training. Moreover, when comparing the use of solely the original data to a dataset that includes oversampled data, it's evident that the oversampling procedure is beneficial.