FedCritic: Serverless Federated Critic Learning-based Resource Allocation for Multi-Cell OFDMA in 6G

2026-05-20 · Source: Takara TLDR - Daily AI Papers · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation, Cloud Computing & IT Infrastructure · Depth: Expert, medium

Summary

FedCritic introduces a serverless federated multi-agent actor-critic framework for distributed downlink resource management in 6G ultra-dense networks. This framework addresses the challenge of amplified inter-cell interference (ICI) in multi-cell orthogonal frequency-division multiple access (OFDMA) by jointly optimizing subcarrier scheduling and power allocation under interference coupling and long-term per-user quality-of-service (QoS) minimum-rate constraints. Unlike centralized training with decentralized execution (CTDE) methods, FedCritic federates the critic learning through lightweight gossip-based parameter averaging across the interference graph, enabling stable value estimation without a central coordinator while keeping policies local. Simulations in an interference-rich reuse-1 setting demonstrate that FedCritic improves mean signal-to-interference-plus-noise ratio (SINR) and cell-edge rate, increases network-wide average sum-rate and fairness, and achieves more stable training with lower coordination overhead compared to non-coordinated and CTDE baselines.

Key takeaway

For network architects designing resource allocation strategies in future 6G ultra-dense networks, FedCritic offers a compelling decentralized approach. You should consider its serverless federated critic learning to manage inter-cell interference and enforce QoS without a central coordinator. This method can significantly improve network-wide sum-rate, fairness, and cell-edge rates while reducing coordination overhead, making it a robust option for scalable and efficient 6G deployments.

Key insights

FedCritic enables decentralized 6G resource allocation by federating critic learning via gossip, avoiding central coordination.

Principles

• Decentralized critic learning enhances scalability.
• Gossip-based averaging stabilizes value estimation.
• Virtual-queue deficit weights enforce QoS.

Method

FedCritic uses a serverless federated multi-agent actor-critic framework. It performs joint subcarrier scheduling and power allocation, enforcing long-term QoS via virtual-queue deficit weights. Critic learning is federated through gossip-based parameter averaging over the interference graph.

In practice

• Improve SINR and cell-edge rates.
• Increase network-wide sum-rate and fairness.
• Reduce coordination overhead in 6G.

Topics

• 6G Networks
• OFDMA Resource Allocation
• Federated Learning
• Actor-Critic Reinforcement Learning
• Inter-Cell Interference
• Serverless Architecture

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• FedCritic: Serverless Federated Critic Learning-based Resource Allocation for Multi-Cell OFDMA in 6G — FedCritic is a serverless federated actor-critic framework for 6G resource allocation, using gossip-based critic averaging to manage interference.
• Federated Multi Agent Deep Learning and Neural Networks for Advanced Distributed Sensing in Wireless Networks — MADL unifies decision-making and inference in tightly coupled wireless sensing, communication, and computing systems.
• Optimization Algorithms for Joint OFDM Waveform Design and RIS Configuration in 6G Networks: From Convex Relaxation to Foundation Models — ML-based optimization offers significant speedups and N-invariance for 6G OFDM-RIS, despite current benchmarking challenges.
• AI-powered 6G core network reaches autonomous control with 40% efficiency gain — AI-powered 6G core networks can achieve autonomous control and significant efficiency gains, as demonstrated by a Korean research team.
• Diffusion Offline Reinforcement Learning for Fair and Energy-Efficient UAV-Assisted Wireless Networks — Diffusion-SAC combines offline RL with diffusion models for robust, data-efficient policy learning in dynamic wireless networks.
• A Reinforcement Learning Based Universal Sequence Design for Polar Codes — A reinforcement learning framework designs universal Polar code sequences, scaling to 2048 code lengths with performance gains.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.