Tối ưu phân bổ tài nguyên và Gia tăng độ tin cậy trong mạng tích hợp Sub6GHz/mmWave bằng Học sâu tăng cường
English below
Dự án này tập trung vào việc tối ưu phân bổ tài nguyên và tăng cường độ tin cậy trong mạng không dây tích hợp Sub6GHz/mmWave bằng phương pháp Học tăng cường sâu (DRL). Chúng tôi đề xuất một phương pháp mới mang tên SACPA (Soft Actor Critic with Power Allocation), giúp đồng thời học được giao diện truyền tối ưu (Sub6GHz hoặc mmWave) và công suất truyền cho từng thiết bị trong mạng.
- Mô phỏng mạng tích hợp Sub6GHz/mmWave theo API Gymnasium và StableBaselines3
- Phân bổ tài nguyên dựa trên học tăng cường sâu
- Cơ chế tăng cường độ tin cậy
- Tải mã nguồn:
git clone https://github.com/ductaingn/Power-Allocation- Cài đặt thư viện:
cd Power-Allocation
pip install -r requirements.txt- Sinh tập dữ liệu:
- Sinh vị trí thiết bị (hỗ trợ tối đa 15 thiết bị với vị trí định sẵn): Sử dụng
generate_devices_positions()trongPower-Allocation/environment/Environment.py. - Sinh
$\tilde{h}_{kn}^v$ : Sử dụnggenerate_h_tilde()trongPower-Allocation/environment/Environment.py.
- Sinh vị trí thiết bị (hỗ trợ tối đa 15 thiết bị với vị trí định sẵn): Sử dụng
- Huấn luyện với một thuật toán:
- Chỉnh sửa cấu hình trong
train_config.yaml - Chạy:
- Chỉnh sửa cấu hình trong
cd Power-Allocation
python train.py- Huấn luyện nhiều thuật toán (benchmark):
- Chỉnh sửa cấu hình trong
train_config.yaml - Chạy:
- Chỉnh sửa cấu hình trong
cd Power-Allocation
python benchmark.py- Sử dụng các hàm trong
Power-Allocation/helper.pyđể trích xuất và xử lý dữ liệu từ Wandb.
Các kết quả thí nghiệm cho thấy SACPA vượt trội hơn rõ rệt so với các thuật toán cơ sở về phần thưởng và độ tin cậy. Đặc biệt, SACPA đạt tỷ lệ thành công trung bình cao hơn trong nhiều điều kiện mạng khác nhau.
Resource Allocation Optimization and Reliability Enhancement in Sub6GHz/mmWave Integrated Network with Deep Reinforcement Learning
This project focuses on optimizing resource allocation and improving reliability in an integrated Sub6GHz/mmWave wireless network using Deep Reinforcement Learning. We propose a novel approach called SACPA (Soft Actor Critic with Power Allocation), which simultaneously learns both the optimal interface (Sub6GHz or mmWave) and transmission power for each device in the network.
- Sub6GHz/mmWave integrated network simulation with Gymnasium and StableBaselines3 API
- Deep reinforcement learning-based resource allocation
- Reliability enhancement mechanisms
- Clone this repository:
git clone https://github.com/ductaingn/Power-Allocation
- Install dependencies:
cd Power-Allocation
pip install -r requirements.txt - Generate dataset
- Generate device positions (currently only supports upto 15 devices with predefined position): Use
generate_devices_positions()function inPower-Allocation/environment/Environment.py - Generate
$\tilde{h}_{kn}^v$ : Usegenerate_h_tilde()function inPower-Allocation/environment/Environment.py
- Generate device positions (currently only supports upto 15 devices with predefined position): Use
- Training with one algorithm:
- Change algorithm configurations in
Power-Allocation\train_config.yaml - Run
- Change algorithm configurations in
cd Power-Allocation
python train.py- Training multiple algorithm:
- Change algorithm configurations in
Power-Allocation\train_config.yaml - Run
- Change algorithm configurations in
cd Power-Allocation
python benchmark.py- Use helper functions in
Power-Allocation/helper.pyto get the data from Wandb logs and process data with functions inPower-Allocation/helper.pyto get the results.
Our experiments demonstrate that SACPA (LearnInterfaceAndPower) significantly outperforms baseline algorithms in both reward and reliability metrics. Particularly, SACPA achieves better average success rates under varying network conditions.
For any questions or contributions, please contact: Duc Tai Nguyen - [email protected]