Proactive Dynamic Spectrum Sharing for URLLC Services Under Uncertain Environment via Deep Reinforcement Learning

Abstract

To support the emerging applications with the coming of the Beyond-5G (B5G) era, e.g., Ultra Reliable Low Latency Communications (URLLC) services, our telecommunications networks have witnessed a serious spectrum shortage problem. According to our spectrum measurement campaign, we note that many bands are actually extremely under-utilized, even for the operators’ ones, e.g., LTE spectrums. Thus, it is expected to share the idle spectrums for the B5G services. Nevertheless, how to determine an effective sharing strategy is non-trivial. It is necessary to jointly consider the spectrum requirement of primary networks and the traffic demand of secondary networks when making the sharing decision, which, however, are both uncertain and hardly known precisely in advance. In this paper, taking the uncertain network environment into account, we propose a Proactive Dynamic Spectrum Sharing (PDSS) scheme to employ the under-utilized LTE spectrums for URLLC service provisioning. We take the long-term overall utility as the objective to achieve a trade-off between two networks to avoid the performance degradation of primary networks, while fulfilling as many URLLC services as possible with quality of service (QoS) guarantee. To deal with the environment uncertainty, we develop a model-free deep reinforcement learning (DRL) based solution, which can proactively capture the feature of the uncertain environment and achieve the best sharing decision autonomously. Based on the real spectrum data, simulation results have shown the effectiveness of the proposed DRL based PDSS scheme.