Machine learning-enhanced 3GPP channel modeling for 5G networks: A vendor-calibrated framework with cross-scenario...

Area of Science:

• Telecommunications Engineering
• Wireless Communication Systems
• Machine Learning Applications

Background:

• Accurate channel characterization is crucial for 5G network planning.
• Existing machine learning methods often lack integration with standardized 3GPP frameworks and vendor-specific equipment parameters.

Purpose of the Study:

• To present a regression-based framework for channel characterization in diverse 5G propagation environments.
• To integrate standardized 3GPP channel models with vendor-specific equipment parameters using supervised learning algorithms.
• To evaluate the performance and generalization capabilities of different regression models.

Main Methods:

• Developed a framework combining 3GPP TR 38.901 channel models with five supervised learning algorithms: linear regression, polynomial regression, SVR, decision tree, and ANN.
• Trained models on 10,000 deterministic samples across Urban Macro (UMa), Urban Micro (UMi), Rural Macro (RMa), and Indoor Hotspot (InH) scenarios at 0.7-60 GHz.
• Utilized vendor-calibrated parameters from Nokia, Huawei, and ZTE equipment for link budget simulations.

Main Results:

• ANN and decision tree achieved highest accuracy for throughput prediction (R2=0.998, RMSE ≤ 24 Mbps).
• All models showed near-perfect fit for path loss estimation under Urban Micro NLOS conditions (R2≈1.0).
• Mixed-scenario training improved generalization across environments (R2 > 0.75), while single-scenario training failed.
• Distance and frequency were identified as dominant predictors, with frequency importance increasing at millimeter-wave bands.

Conclusions:

• The proposed framework provides accurate performance estimates for 5G network planning by integrating standardized models with vendor equipment data.
• ANN and decision tree models are recommended for throughput prediction due to their high accuracy.
• Mixed-scenario training is critical for achieving robust cross-environment generalization in channel modeling.