Inference code of the AutoBS framework from our paper "AutoBS: Autonomous Base Station Deployment Framework with Reinforcement Learning and Digital Twin Network".
- We introduce a novel DRL-based framework for single/multi-BS deployment that incorporates PMNet for real-time, site-specific channel predictions.
- AutoBS reduces inference time from hours to milliseconds compared to exhaustive methods, particularly in multi-BS deployments, making it practical for large-scale, real-time optimization.
- The repository includes the checkpoints of our single-BS, multi-BS agent and the PMNet. SionnaRT is used for visualizing the deployment result.
| Model | Download Link |
|---|---|
| Single-BS Agent | Download |
| Multi-BS Agent | Download |
| PMNet | Download |
To evaluate the performance of our AutoBS agent, refer to the following commands to deploy either a single base station or two base stations on a test map. Note that this script would also execute the heuristic and exhaustive methods for comparison.
python inference.py \
--version [single/multi] \
--crop_id [test-map-id] \ # enter a number between 0 and 15
--reward_type [coverage/capacity] \ # for baseline methods
# e.g.,
# python inference.py \
# --version single \
# --crop_id 0 \
# --reward_type coverage \After running the inference script, the output coverage map will be saved in the visulaize/sionna_output/ directory.
We would like to acknowledge the contributions of the following individuals to the framework design and simulations: (1) Arjun Balamwar; (2) Yanqing Lu