Radio Access Network (RAN) slicing is a shared resource allocation problem: each slice seeks to satisfy its Service Level Agreement (SLA) with minimal radio-resource consumption, while all slices compete for a limited resource budget. This creates both an intra-slice trade-off between SLA assurance and resource efficiency and an inter-slice conflict over shared resources, calling for specialized yet coordinated control. In this work, we present CoSlice, a coordinated multi-agent reinforcement learning framework for inter-slice physical resource block allocation in Open Radio Access Network (O-RAN). CoSlice uses preference-conditioned slice agents for fast resource allocation and an LLM-based coordinator to adapt their objectives using global network observations at a slower timescale. We formulate the problem as a multi-objective partially observable stochastic game (mo-POSG), establish the existence of a Preference Correlated Equilibrium, and derive a sublinear regret bound for the learning procedure. We implement the algorithm by using a two-stage training procedure that consists of multi-objective pretraining and joint optimization. We build a network slicing evaluation environment using ORANSlice and its radio-frequency simulator (RFsim), supporting realistic protocol-stack execution and application traffic across multiple slices. Experimental results show that CoSlice reduces both average and tail SLA violations compared with other centralized and multi-agent baselines under diverse slice workloads.
Figures & tables
Fig. 1 : System overview of CoSlice. From top to bottom, the architecture consists of an LLM coordinator in the Non-RT RIC, slice agents in the Near-RT RIC, and a PRB scheduler in the O-DU. Our design focuses on the first two layers.
Fig. 2 : Periodic decisions and delayed control propagation across the rApp, xApp, and MAC layers. Each controller makes a decision from the collected information, but the decision affects the next layer only after inference and message delivery, making low end-to-end latency essential.
Fig. 3 : Pareto front of each slice approximated by MORL pretraining. Each point corresponds to an operating point under a different SLA-versus-PRB preference. A larger SLA weight leads to lower SLA violation but generally requires more PRBs, whereas a larger PRB-saving weight favors more conservative requests. The resulting policy can represent multiple Pareto-optimal solutions, providing a suitable initialization for subsequent multi-agent training.
Fig. 4 : Measured throughput and RAN latency trace data in the three- and six-slice settings. The traces show distinct service dynamics: eMBB exhibits high-rate bursty or sustained transmissions, whereas URLLC shows intermittent latency-sensitive activity and mMTC presents lower-rate aggregated demand. The six-slice setting further exposes differentiated application-level patterns within these service classes.
Fig. 5 : Comparison of all algorithms in the three- and six-slice settings. We report the macro-averaged mean and tail (95th- and 99th-percentile) relative SLA violations across all slices; lower values are better. CoSlice achieves the lowest values for all three statistics in both settings. Compared with the baseline with best performance, CoSlice reduces the mean/P95/P99 violations by 19.4%/30.2%/0.2% in the three-slice setting and 37.9%/37.4%/12.7% in the six-slice setting.
Fig. 6 : Per-slice SLA performance in the three-slice setting. CoSlice leads consistently on eMBB, with moderate tradeoffs on URLLC and mMTC; the mMTC violations remain small.
Fig. 7 : Per-slice SLA performance in the six-slice setting. CoSlice leads on eMBB Live, eMBB Bulk, URLLC Control, and aggregated mMTC, while trading performance on eMBB Video and URLLC Gaming for the strongest overall macro-average result.
Fig. 8 : Case study of one LLM decision step.
Fig. 9 : Ablation study of CoSlice in the three- and six-slice settings. The bars show the macro-averaged mean and tail (95th- and 99th-percentile) relative SLA violations across slices; lower values are better. Compared to the base version, adding the LLM coordinator reduces the macro-averaged mean and 95th-/99th-percentile violations by 7.6%/17.2%/8.9% in the three-slice setting and 1.0%/11.9%/7.2% in the six-slice setting. Further adding MORL pretraining yields additional reductions of 17.6%/18.4%/29.6% and 47.0%/49.9%/12.8%, respectively.
xSlice
MLP subagents
LLM agent
End-to-end latency (ms)
8.4±0.8
6.5±0.8
508.2±8.8
TABLE I : Control latency across different components. All components remain well below their control periods. Local CoSlice agents achieve lower and more stable inference latency than the single xSlice agent due to their simpler input spaces. The LLM coordinator incurs higher latency from autoregressive decoding, but still averages below 1,s. MLP results are averaged over 1000 samples, and LLM results over 30 warm-up-excluded samples.
The recent advances toward sixth-generation (6G) and beyond-6G networks have accelerated the need for intelligent resource management mechanisms capable of supporting heterogeneous services under shared infrastructures in network slicing. However, resource allocation in network slicing naturally forms a resource-coupled cooperative optimization problem with competing slice demands. Slices compete for limited resources to minimize individual latencies while coordinating to avoid conflicts and underutilization. Although multi-agent reinforcement learning (MARL) has shown promising performance in such settings, existing online formulations remain costly, unsafe, and difficult to deploy due to their reliance on environmental interactions and communication among agents. In this work, we present explainable artificial intelligence (XAI)-guided conservative decentralized execution (X-CODE). X-CODE is an explainable offline MARL that operates offline without environmental interaction, nor inter-agent communication. It exploits explainability-aware reward shaping to modify the relative preference among joint offline transitions during centralized training to improve decentralized resource-allocation behavior. In deployment, the agents operate independently without signaling exchange among the agents. Simulation results demonstrate that the proposed approach achieves zero observed resource-conflict events in the evaluated test episodes while minimizing per-slice latencies. Moreover, the proposed framework exhibits lower signaling overhead and reduces effective inference latency by 88 % under the considered communication-delay model compared to the online baselines. Source codes and datasets are available through: https://github.com/Eslam211/xcode-ran-slicing.
Eslam Eldeeb, Hatim Chergui, Merouane Debbah
Centre for Wireless Communications (CWC), University of Oulu, Finland · i2CAT Foundation, Spain · Research Institute for Digital Future, Khalifa University, 127788 Abu Dhabi, UAE
Next-generation (NextG) cellular networks are designed to support emerging applications with diverse data rate and latency requirements, such as immersive multimedia services and large-scale Internet of Things deployments. A key enabling mechanism is radio access network (RAN) slicing, which dynamically partitions radio resources into virtual resource blocks to efficiently serve heterogeneous traffic classes, including enhanced mobile broadband (eMBB), massive machine-type communications (mMTC), and ultra-reliable low-latency communications (URLLC). In this paper, we study the impact of adversarial attacks on AI-driven RAN slicing decisions, where a budget-constrained adversary selectively jams slice transmissions to bias deep reinforcement learning (DRL)-based resource allocation, and quantify the resulting service level agreement (SLA) violations and post-attack recovery behavior. Our results indicate that budget-constrained adversarial jamming can induce severe and slice-dependent steady-state SLA violations. Moreover, the DRL agent's reward converges toward the clean baseline only after a non-negligible recovery period.
Deemah H. Tashman, Soumaya Cherkaoui
Department of Computer and Software Engineering, Polytechnique Montreal, Montreal, Canada
Open RAN (O-RAN) slicing xApps must adapt resource allocations to changing channel conditions and traffic demands while meeting service-level agreements (SLAs). Deep reinforcement learning can produce adaptive policies, but their allocation rules remain encoded in neural-network parameters. Our goal is to retain this adaptability while making the controller's decision logic directly inspectable and editable by operators. We use a large language model (LLM) to evolve slicing controllers as compact Python programs whose decision logic remains readable and editable after optimization. The LLM proposes and revises candidates offline, while a calibrated simulator scores them, and the selected decision module runs unchanged in the O-RAN control path. On the NSF POWDER 5G testbed, the evolved controller releases resources from a guaranteed slice whose throughput target becomes unattainable under a sustained channel fade, improving best-effort throughput from 158.2 to 228.6 Mbps, a 44.5% gain over the best static allocation. Since the controllers are readable source code, their behavior can be predicted from their equations, defects can be diagnosed by reading the code, and calibration errors can be corrected with one-line edits, reducing SLA misses from 79.9% to 2.2% in one case and more than doubling fitness in another. In a four-slice trace-driven simulation calibrated to the same testbed, evolutionary search achieves higher average evaluation scores than independent prompting at a matched proposal budget, with mean normalized gains on held-out traces of 16.3% for prompting alone, 32.1% for evolution from scratch, and 51.0% for evolution from a starting program.
Faezeh Dehghan Tarzjani, Bhaskar Krishnamachari
Department of Electrical and Computer Engineering, University of Southern California