Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

NetMind: Adaptive RAN Baseband Function Placement by GCN Encoding and Maze-solving DRL

Haiyuan Li, Peizheng Li, Karcius Day Assis, Adnan Aijaz, Sen Shen, Reza Nejabati, Shuangyi Yan, Dimitra Simeonidou

TL;DR

NetMind tackles efficient joint placement of baseband functions and UPF in disaggregated RANs with MEC to minimize power under latency and capacity constraints, formulated as a non-convex problem $P$. It introduces a maze-solving MDP where a Graph DQN operates on a unified state representation produced by a GCN encoder, enabling a single policy to generalize across heterogeneous network topologies. The key contributions include (i) casting placement as maze solving, (ii) a GCN-based encoder/decoder for cross-network generalization, and (iii) substantial training-cost reductions (~70%), with NetMind delivering ~32.76% additional power savings and ~41.67% improvement in service stability versus baselines, approaching MILP performance. The approach supports near real-time control via Near-RT RIC and offers scalability to other 3GPP split options, with future work on incremental learning and distribution drift analysis.$

Abstract

The disaggregated and hierarchical architecture of advanced RAN presents significant challenges in efficiently placing baseband functions and user plane functions in conjunction with Multi-Access Edge Computing (MEC) to accommodate diverse 5G services. Therefore, this paper proposes a novel approach NetMind, which leverages Deep Reinforcement Learning (DRL) to determine the function placement strategies in RANs with diverse topologies, aiming at minimizing power consumption. NetMind formulates the function placement problem as a maze-solving task, enabling a Markov Decision Process with standardized action space scales across different networks. Additionally, a Graph Convolutional Network (GCN) based encoding mechanism is introduced, allowing features from different networks to be aggregated into a single RL agent. That facilitates the RL agent's generalization capability and minimizes the negative impact of retraining on power consumption. In an example with three sub-networks, NetMind achieves comparable performance to traditional methods that require a dedicated DRL agent for each network, resulting in a 70% reduction in training costs. Furthermore, it demonstrates a substantial 32.76% improvement in power savings and a 41.67% increase in service stability compared to benchmarks from the existing literature.

NetMind: Adaptive RAN Baseband Function Placement by GCN Encoding and Maze-solving DRL

TL;DR

NetMind tackles efficient joint placement of baseband functions and UPF in disaggregated RANs with MEC to minimize power under latency and capacity constraints, formulated as a non-convex problem . It introduces a maze-solving MDP where a Graph DQN operates on a unified state representation produced by a GCN encoder, enabling a single policy to generalize across heterogeneous network topologies. The key contributions include (i) casting placement as maze solving, (ii) a GCN-based encoder/decoder for cross-network generalization, and (iii) substantial training-cost reductions (~70%), with NetMind delivering ~32.76% additional power savings and ~41.67% improvement in service stability versus baselines, approaching MILP performance. The approach supports near real-time control via Near-RT RIC and offers scalability to other 3GPP split options, with future work on incremental learning and distribution drift analysis.$

Abstract

The disaggregated and hierarchical architecture of advanced RAN presents significant challenges in efficiently placing baseband functions and user plane functions in conjunction with Multi-Access Edge Computing (MEC) to accommodate diverse 5G services. Therefore, this paper proposes a novel approach NetMind, which leverages Deep Reinforcement Learning (DRL) to determine the function placement strategies in RANs with diverse topologies, aiming at minimizing power consumption. NetMind formulates the function placement problem as a maze-solving task, enabling a Markov Decision Process with standardized action space scales across different networks. Additionally, a Graph Convolutional Network (GCN) based encoding mechanism is introduced, allowing features from different networks to be aggregated into a single RL agent. That facilitates the RL agent's generalization capability and minimizes the negative impact of retraining on power consumption. In an example with three sub-networks, NetMind achieves comparable performance to traditional methods that require a dedicated DRL agent for each network, resulting in a 70% reduction in training costs. Furthermore, it demonstrates a substantial 32.76% improvement in power savings and a 41.67% increase in service stability compared to benchmarks from the existing literature.
Paper Structure (11 sections, 6 equations, 7 figures, 1 table)

This paper contains 11 sections, 6 equations, 7 figures, 1 table.

Table of Contents

  1. Introduction
  2. Baseband Function Deployment Scenario in O-RAN and problem formulation
  3. NetMind: Graph DQN-based adaptive baseband function chain placement
  4. Maze-solving Environment
  5. NetMind for Baseband Function Placement
  6. System Modelling and Simulation
  7. Network Modelling
  8. Training of NetMind Components
  9. Numerical Result
  10. Discussion and Future Works
  11. Conclusion

Figures (7)

  • Figure 1: O-RAN in edge computing networks. The network is divided into multiple sub-networks with various network structures based on MEC distributions.
  • Figure 2: Maze solving scenario.
  • Figure 3: NetMind composition and mechanism.
  • Figure 4: Simulation network consists of three sub-networks with a shared 5GC and various network structures.
  • Figure 5: (a) GCN encoder training performance for three sub-networks; (b) DQN training convergence for different scenarios. Notice the 95-percentile confidence intervals marked over the mean performance curves.
  • ...and 2 more figures