Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Collaborative Optimization of Wireless Communication and Computing Resource Allocation based on Multi-Agent Federated Weighting Deep Reinforcement Learning

Junjie Wu, Xuming Fang

TL;DR

The paper tackles joint optimization of computing and communication resources in privacy-aware wireless networks using a novel framework, MAFWDRL, that fuses multi-agent deep reinforcement learning with Federated Weighting. It defines a dual-privacy system (privacy-sensitive and privacy-insensitive UEs) and develops distinct training schemes, including fully decentralized and semi-centralized MADRL, plus a FedWgt-based global critic ensemble. A tailored exploration-noise function enhances off-policy learning, and the FedWgt weights reduce local–global gradient mismatches, improving model generalization under heterogeneity. Simulation on a NS-3–based platform with mobility and MIMO channels shows improved throughput, reduced latency, and lower energy consumption compared to baselines, validating the approach for real-time, privacy-aware wireless edge networks.

Abstract

As artificial intelligence (AI)-enabled wireless communication systems continue their evolution, distributed learning has gained widespread attention for its ability to offer enhanced data privacy protection, improved resource utilization, and enhanced fault tolerance within wireless communication applications. Federated learning further enhances the ability of resource coordination and model generalization across nodes based on the above foundation, enabling the realization of an AI-driven communication and computing integrated wireless network. This paper proposes a novel wireless communication system to cater to a personalized service needs of both privacy-sensitive and privacy-insensitive users. We design the system based on based on multi-agent federated weighting deep reinforcement learning (MAFWDRL). The system, while fulfilling service requirements for users, facilitates real-time optimization of local communication resources allocation and concurrent decision-making concerning computing resources. Additionally, exploration noise is incorporated to enhance the exploration process of off-policy deep reinforcement learning (DRL) for wireless channels. Federated weighting (FedWgt) effectively compensates for heterogeneous differences in channel status between communication nodes. Extensive simulation experiments demonstrate that the proposed scheme outperforms baseline methods significantly in terms of throughput, calculation latency, and energy consumption improvement.

Collaborative Optimization of Wireless Communication and Computing Resource Allocation based on Multi-Agent Federated Weighting Deep Reinforcement Learning

TL;DR

The paper tackles joint optimization of computing and communication resources in privacy-aware wireless networks using a novel framework, MAFWDRL, that fuses multi-agent deep reinforcement learning with Federated Weighting. It defines a dual-privacy system (privacy-sensitive and privacy-insensitive UEs) and develops distinct training schemes, including fully decentralized and semi-centralized MADRL, plus a FedWgt-based global critic ensemble. A tailored exploration-noise function enhances off-policy learning, and the FedWgt weights reduce local–global gradient mismatches, improving model generalization under heterogeneity. Simulation on a NS-3–based platform with mobility and MIMO channels shows improved throughput, reduced latency, and lower energy consumption compared to baselines, validating the approach for real-time, privacy-aware wireless edge networks.

Abstract

As artificial intelligence (AI)-enabled wireless communication systems continue their evolution, distributed learning has gained widespread attention for its ability to offer enhanced data privacy protection, improved resource utilization, and enhanced fault tolerance within wireless communication applications. Federated learning further enhances the ability of resource coordination and model generalization across nodes based on the above foundation, enabling the realization of an AI-driven communication and computing integrated wireless network. This paper proposes a novel wireless communication system to cater to a personalized service needs of both privacy-sensitive and privacy-insensitive users. We design the system based on based on multi-agent federated weighting deep reinforcement learning (MAFWDRL). The system, while fulfilling service requirements for users, facilitates real-time optimization of local communication resources allocation and concurrent decision-making concerning computing resources. Additionally, exploration noise is incorporated to enhance the exploration process of off-policy deep reinforcement learning (DRL) for wireless channels. Federated weighting (FedWgt) effectively compensates for heterogeneous differences in channel status between communication nodes. Extensive simulation experiments demonstrate that the proposed scheme outperforms baseline methods significantly in terms of throughput, calculation latency, and energy consumption improvement.
Paper Structure (13 sections, 49 equations, 9 figures, 1 table)

This paper contains 13 sections, 49 equations, 9 figures, 1 table.

Table of Contents

  1. Introduction
  2. System Model and Problem Formulation
  3. Channel Propagation Model
  4. Computing Energy Consumption Model
  5. Problem Formulation
  6. resource allocation and energy consumption optimization based on MAFWDRL
  7. Federated Weighting Design
  8. Exploration Noise Design for Off-policy DRL
  9. MAFWDRL algorithm design for distributed wireless system
  10. Simulation results
  11. Experiments on noise design for off-policy MADRL
  12. Experiments on MAFWDRL
  13. Conclusion

Figures (9)

  • Figure 1: Wireless communication scenario based on distributed learning.
  • Figure 2: Resource allocation optimization architecture based on MAFWDRL for distributed wireless networks.
  • Figure 3: An instance of user location and communication status simulated by NS3-simulator.
  • Figure 4: Performance comparison of the explore noise between baseline function $f$ and the proposed function $g$.
  • Figure 5: Performance comparison of MAFADRL and the proposed MAFWDRL in term of the proposed noise function $g$ with different change rate $\eta$.
  • ...and 4 more figures