5G RAN slicing makes it possible to simultaneously support enhanced mobile broadband (eMBB) and ultra-reliable low-latency communications (URLLC) services characterized by different numerologies over a shared radio physical layer. However, the inter-numerology interference (INI) generated by the multiplexing of spectrum slices having heterogeneous sub-carrier spacing can hinder the service provisioning performance of both slices. In this context, we propose a spectrum allocation scheme capable of meeting the eMBB and URLLC service requirements while also mitigating the INI affecting each user. To overcome the computational complexity of the optimization problem, we design an INI-aware agent, based on Branching Dueling Q-networks (BDQ), which allocates a suitable number of spectrum resources to each user in order to satisfy the service requirements. In addition, we boost the agent learning efficiency by designing an action masking module that removes unfeasible actions. We compare the agent performance to state-of-the-art resource allocation algorithms that do not account for the INI. Results reveal that the proposed agent outperforms the considered benchmark schemes by ensuring a higher eMBB user data rate and, at the same time, a lower URLLC user delay.
Mixed-Numerology Interference-Aware Spectrum Allocation for eMBB and URLLC Network Slices
Zambianco, Marco;Verticale, Giacomo
2021-01-01
Abstract
5G RAN slicing makes it possible to simultaneously support enhanced mobile broadband (eMBB) and ultra-reliable low-latency communications (URLLC) services characterized by different numerologies over a shared radio physical layer. However, the inter-numerology interference (INI) generated by the multiplexing of spectrum slices having heterogeneous sub-carrier spacing can hinder the service provisioning performance of both slices. In this context, we propose a spectrum allocation scheme capable of meeting the eMBB and URLLC service requirements while also mitigating the INI affecting each user. To overcome the computational complexity of the optimization problem, we design an INI-aware agent, based on Branching Dueling Q-networks (BDQ), which allocates a suitable number of spectrum resources to each user in order to satisfy the service requirements. In addition, we boost the agent learning efficiency by designing an action masking module that removes unfeasible actions. We compare the agent performance to state-of-the-art resource allocation algorithms that do not account for the INI. Results reveal that the proposed agent outperforms the considered benchmark schemes by ensuring a higher eMBB user data rate and, at the same time, a lower URLLC user delay.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
