THDeg-Bench approved as first one6G-endorsed testbed
On August 27, 2025, the one6G Board officially approved the very first one6G-endorsed testbed: THDeg-Bench from the Deggendorf Institute of Technology (TH Deggendorf). This groundbreaking endorsement gives all one6G members immediate access to a powerful, standardized tool for evaluating AI solutions critical to the future of 6G.
The new one6G-endorsed testbed program, a key initiative of one6G Working Group 4, is designed to provide high-quality, jointly accessible testbeds to the one6G community. This collaborative approach ensures that one6G members can leverage shared, validated infrastructure to accelerate research and development, avoiding the need to build complex testing environments from scratch.
THDeg-Bench is a Kubernetes-native framework engineered for the automated benchmarking of AI-based time series forecasting tasks. Time series forecasting is a cornerstone of future 6G applications, including real-time traffic prediction, mobility forecasting, and signal-quality estimation. However, the deployment of such models at scale has been hampered by a lack of tools that offer container-native scheduling, dynamic resource allocation, and, crucially, reproducibility.
This is where THDeg-Bench fills the gap. Built on PyTorch Lightning and Hydra, the framework synthesizes a complete workflow, from data ingestion to model training, hyperparameter tuning, and monitoring, eliminating the need for custom orchestration. Experimenters simply upload their datasets and provide a configuration file, allowing for scalable, reproducible evaluation of various forecasting methods.
Key features of the THDeg-Bench framework include:
- Kubernetes-native architecture: Enables scalable, one-click deployment with integrated telemetry from Prometheus, Grafana, and Kepler, allowing for comprehensive monitoring of experiments.
- YAML-defined experiments: Supports scalable Optuna sweeps and AIM tracking for efficient and reproducible hyperparameter tuning.
- GPU-aware scheduling: Optimizes resource allocation for AI workloads, ensuring efficient use of computational resources.
- Energy-aware monitoring: Tracks the energy consumption of model training and inference tasks, a vital metric for developing sustainable 6G networks.
- Spatiotemporal forecasting support: Includes spatial kernels to address the complexity of spatiotemporal forecasting, a crucial capability for network-wide predictions.
By providing a streamlined, automated, and reproducible benchmarking process, THDeg-Bench will enable one6G members to rigorously compare AI models across accuracy, latency, and energy efficiency. This is particularly valuable for operators who need to select models that strike the right balance between performance and operational constraints, such as favoring faster, lower-power models for edge computing applications.
The approval of THDeg-Bench marks a significant step forward for the one6G community, solidifying its commitment to fostering a collaborative ecosystem for 6G innovation. The availability of this powerful tool will accelerate the development and deployment of AI-native solutions, paving the way for the intelligent and efficient networks of the future.
one6G members can now contact TH Deggendorf to arrange access and begin leveraging this state-of-the-art testbed in their research and development efforts.
For more information, contact:
- Prof. Andreas Kassler <andreas(dot)kassler@th-deg.de>
- one6G WG4 leadership team