6GSmartRRM is an AI/ML-driven project addressing the challenges of joint sub-band allocation and power control in hyper-dense 6G In-Factory subnetworks (InF-S). These subnetworks form the backbone of industrial automation systems, such as robotics, production modules, and conveyors, and face significant interference and resource management issues.
This project focuses on optimizing spectral efficiency, mitigating interference, and adapting to dynamic industrial environments using advanced AI techniques like deep learning and graph neural networks (GNNs).
This project is part of the Radio Resource Management (RRM) for 6G in-X Subnetworks Challenge from the ITU AI/ML in 5G Challenge, which focuses on:
- Tackling interference and resource allocation in hyper-dense industrial networks.
- Developing AI-powered solutions for dynamic and mobile environments.
- Ensuring high spectral efficiency and reliability in constrained radio frequency environments.
The project addresses the limitations of traditional radio resource management (RRM) methods and explores adaptive, scalable, and computationally efficient AI-based solutions for 6G In-Factory scenarios.
The proposed model (in the future) responds to a dynamic that we could not find a suitable solution in the literature. After examining a series of publications, we were able to determine possible actionable insights. We have summarize the findings in bibliography documentation
The first approach we are considering is the definition of a two-stage control system. The first stage will use a model to estimate the optimal allocation while the second will set the power settings to reduce interference and increase the output.
In the following table, you can see the current results (they may not be updated, check the wandb link for updated results) for the first stage models / algoriths with the maximum power allocated (not yet done).
Subband Allocation Results:
| Algorithm / Model | Bit Rate (Mbps) | Spectral Efficency | Jain Fairness | Proportional Factor Loss (%) |
|---|---|---|---|---|
| SISA Algorithm (Baseline) | 44.04659 | 4.404659 | 0.9330245 | 94.2641 % |
| Rate Confirming (Min Rate Loss) | 44.66844 | 4.466844 | 0.9417886 | 88.3355 % |
For comparison and procedure, the results were obtained from the mean of 40.000 simulations of the networks. All metrics are itself the mean bit rate and metrics from each simulation between all subnetworks.