How n00bception/ai-powered-5g-openran-optimizer works
Project Type and Purpose
• This is a Python-based AI/ML optimization system for 5G Open RAN.
• It ingests historical & real-time network data, applies multiple ML models (anomaly detection, predictive planning, dynamic optimization, energy efficiency), and pushes optimization decisions back into the RAN via Open RAN interfaces.
File Structure
Top Level
• src/
– data_preparation/ (data ingestion & cleaning)
– models/ (four distinct optimization models)
– utils/ (API wrappers, config, logging, data loaders, prediction helpers)
– main.py (orchestrator entry point)
• docs/ (architecture write-up, developer & user guides)
• tests/ (unit & integration tests)
• requirements.txt (dependencies)
README for Additional Insights
• Uses supervised/unsupervised ML, deep learning, reinforcement learning.
• Integrates with Open RAN (O-RAN interfaces).
• CLI‐driven workflow: preprocess → train → evaluate → optimize.
Main Components & Their Relationships
A. Data Layer
1. Raw Data Sources
• Historical datasets (traffic logs, signal strength, weather)
• Real-time streaming data from RAN
2. Data Preparation Module (src/data_preparation)
• Extraction → Cleaning → Transformation
• Output: clean feature sets
3. Data Store (could be file system, object storage, or time-series DB)
B. Model Training & Serving
1. Training Subsystem (src/models/*/train.py)
• Four parallel pipelines (each model folder)
• Configurable hyperparameters via utils/config.py
2. Trained Model Repository
• Serialized models (Pickle, ONNX, TensorFlow SavedModel)
3. Prediction/Inference Subsystem (src/models/*/predict.py)
• Loads models from repository
• Wrapped by utils/model_prediction.py
C. Orchestration & API
1. main.py
• CLI arguments route to preprocess, train, evaluate, optimize
2. utils/api.py
• REST/gRPC client to interact with O-RAN southbound interfaces
D. External Services
• Open RAN controllers (via O-RAN API)
• External data providers (weather, geo-location, OSS/BSS)
E. Cross-Cutting Concerns
• Configuration management (utils/config.py)
• Logging (utils/logger.py)
• Testing (tests/)
Architectural Patterns & Design Principles
• Layered architecture: Data → ML pipelines → Orchestration → External interfaces
• Plugin-style model modules: new algorithms slot in under src/models
• Separation of batch (training) vs real-time (inference) pipelines
• Single responsibility: each module has a focused purpose
• Config-driven flows to avoid hard-coding parameters
Essential Architectural Elements
• Distinct separation of concerns (data vs ML vs API)
• Clear depiction of both batch and streaming pathways
• Explicit model management lifecycle (train → store → load → infer)
• Integration touchpoints with external Open RAN interfaces