Deep learning based mobile data offloading in mobile edge computing systems

Abstract

Abstract Mobile Edge Computing (MEC) has been regarded as a key technology of the future communication systems in the industry due to its capability to satisfy a wide range of requirements of the emerging wireless terminals (virtual reality devices, augmented reality, and Intelligent Vehicles), such as high data rate, low latency, and huge computation. Besides, difficulties in the lack of resources in the licensed band have prompted researches on mobile data offloading. Owing to the cheap and effective characteristics of WiFi AP, it is utilized to offload some devices from small base stations (SBS) in this paper. Furthermore, a multi-Long Short Term Memory (LSTM) based deep-learning model is constructed to predict the real-time traffic of SBS, which may help us perform the offloading process accurately. According to the prediction results, an mobile data offloading strategy based on cross entropy (CE) method has been proposed. The presented results based on actual dataset provide strong proofs of the applicability of the prediction and offloading scheme we proposed.