This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
Agoran: An Agentic Open Marketplace for 6G RAN Automation
Created by
Haebom
Author
Ilias Chatzistefanidis, Navid Nikaein, Andrea Leone, Ali Maatouk, Leandros Tassiulas, Roberto Morabito, Ioannis Pitsiorlas, Marios Kountouris
Outline
Next-generation mobile networks must coordinate the conflicting goals of multiple service providers. However, current network slice controllers are rigid, policy-bound, and unaware of business context. This paper presents the Agoran Service and Resource Broker (SRB), an agent marketplace that directly engages stakeholders in the operational loop. Inspired by the ancient Greek agora, Agoran decentralizes authority across three autonomous AI departments: the legal department, the executive department, and the judiciary. The legal department uses a search-augmented large-scale language model (LLM) to answer compliance queries; the executive department maintains real-time situational awareness through a database of vectors updated by a watchdog; the judiciary evaluates each agent message with a rule-based trust score; and the arbitration LLM detects malicious behavior and applies real-time incentives to restore trust. Negotiating agents on the stakeholder side and arbitrator agents on the SRB side negotiate feasible, Pareto-optimal proposals generated by a multi-objective optimizer, reaching a consensus in a single round and then distributing them to the Open and AI RAN controllers. Deployed on a private 5G testbed and evaluated using real-world vehicle movement tracking, Agoran achieved significant benefits: (i) a 37% increase in eMBB slice throughput, (ii) a 73% reduction in URLLC slice latency, and (iii) an 8.3% reduction in PRB usage compared to a static baseline. The 1 billion-parameter Llama model, fine-tuned on 100 GPT-4 conversations for 5 minutes, operates within 6 GiB of memory, converges in 1.3 seconds, and recovers approximately 80% of the decision quality of GPT-4.1. These results establish Agoran as a concrete and standards-compliant path toward ultra-flexible and stakeholder-centric 6G networks. A live demo is available at https://www.youtube.com/watch?v=h7vEyMu2f5w\&ab_channel=BubbleRAN .
Takeaways, Limitations
•
Takeaways:
◦
We present a novel network slice management method that efficiently coordinates the needs of various stakeholders through an agent-based market approach.
◦
Experiments using real-world vehicle tracking demonstrate improved eMBB and URLLC slice performance and reduced PRB usage.
◦
Demonstrates the potential for effective decision-making even in resource-constrained environments by leveraging lightweight LLM.
◦
Presenting specific technical directions for ultra-flexible and stakeholder-centric design of 6G networks.
•
Limitations:
◦
Currently evaluated in a private 5G testbed, further validation of performance and scalability in a real-world commercial environment is required.
◦
In-depth consideration and solutions are needed for the reliability and security issues of LLM. Further analysis is needed to determine the actual effectiveness of defense mechanisms against malicious actors.
◦
Further research is needed to determine generalizability across different service types and traffic patterns.
◦
Lack of detailed information on the scale and quality of the data used for LLM fine-tuning.
