Intent-based Radio Scheduler for RAN Slicing: Learning to deal with different network scenarios

TL;DR

This work tackles the challenge of robust radio resource scheduling in RAN slicing under diverse network scenarios by introducing an intent-based RRS powered by multi-agent reinforcement learning. The architecture splits inter-slice scheduling (PPO-based) from intra-slice scheduling (shared-parameter MARL), and aligns decisions with slice intents through an enhanced intent-drift reward. It demonstrates strong protection for high-priority slices and improved overall intent satisfaction across multiple network scenarios, while showing that generalization across unseen scenarios remains difficult and can be aided by transfer learning, which substantially reduces training time. The results suggest that an intent-aware MARL framework, coupled with transfer learning, is a viable path toward production-ready, adaptable RAN slicing in 6G-era networks.

Abstract

The future mobile network has the complex mission of distributing available radio resources among various applications with different requirements. The radio access network slicing enables the creation of different logical networks by isolating and using dedicated resources for each group of applications. In this scenario, the radio resource scheduling (RRS) is responsible for distributing the radio resources available among the slices to fulfill their service-level agreement (SLA) requirements, prioritizing critical slices while minimizing the number of intent violations. Moreover, ensuring that the RRS can deal with a high diversity of network scenarios is essential. Several recent papers present advances in machine learning-based RRS. However, the scenarios and slice variety are restricted, which inhibits solid conclusions about the generalization capabilities of the models after deployment in real networks. This paper proposes an intent-based RRS using multi-agent reinforcement learning in a radio access network (RAN) slicing context. The proposed method protects high-priority slices when the available radio resources cannot fulfill all the slices. It uses transfer learning to reduce the number of training steps required. The proposed method and baselines are evaluated in different network scenarios that comprehend combinations of different slice types, channel trajectories, number of active slices and users' equipment (UEs), and UE characteristics. The proposed method outperformed the baselines in protecting slices with higher priority, obtaining an improvement of 40% and, when considering all the slices, obtaining an improvement of 20% in relation to the baselines. The results show that by using transfer learning, the required number of training steps could be reduced by a factor of eight without hurting performance.

Figures (17)

• Figure 1: Example of RRS executed over the duration of a TTI (TTI) in a RAN slicing scenario with $5$ radio resources (RBGs), and $2$ slices containing $2$ UEs each. The inter-slice scheduler distributes the available radio resources among slices, and then the intra-slice scheduler distributes the respective radio resources among the UE associated with the specific slice.
• Figure 2: An intent-based system with an RRS intent handler responsible for receiving slice intents and network information to generate schedule actions to fulfill the provided intents. The RRS intent handler is based on MARL.
• Figure 3: Proposed intent-based MARL architecture to perform inter- and intra-slice scheduling in different network scenarios. The RL inter-slice scheduler has a dedicated policy, while the RL intra-slice schedulers utilize a shared policy.
• Figure 4: Single base station network scenario possibilities considering a different number of active slices, slice types, and UE channel trajectories. A network scenario is a specific combination of slices, slice types, and different channel conditions.
• Figure 5: Intent-drift example for an effective throughput intent with a requirement of 100 Mbps and an overfulfillment rate of $10$%. In step $n$, the served throughput is equal to or greater than the requested intent requirement. The served throughput decreases in step $n+1$, but the intent is still fulfilled.