Deep Learning for Path Loss Prediction at 7 GHz in Urban Environment.

Area of Science:

• Wireless communication
• Radio propagation modeling
• Machine learning applications

Background:

• The 6 GHz band enables spectrum sharing between licensed and unlicensed devices.
• Automated Frequency Coordination (AFC) systems manage unlicensed devices to prevent interference.
• Accurate radio propagation models are essential for interference analysis in shared spectrum bands.

Purpose of the Study:

• To develop and evaluate a novel model-aided deep learning technique for path loss prediction at 7 GHz in urban settings.
• To assess the performance of a hybrid model integrating physics-based knowledge and neural network capabilities.
• To investigate the generalization capability of the proposed model to new, unseen locations.

Main Methods:

• A hybrid deep learning model was developed, combining a physics-based channel model with a neural network.
• The model was trained and validated using extensive, high-quality propagation measurement data from four urban locations.
• Path loss predictions were compared against established empirical models.

Main Results:

• The hybrid deep learning model demonstrated superior prediction performance compared to most empirical models.
• The model achieved accurate path loss predictions at 7 GHz in a complex urban environment.
• Fine-tuning the model showed feasibility for generalization to new geographical areas.

Conclusions:

• Model-aided deep learning offers a promising approach for accurate path loss prediction in the 6 GHz band.
• Hybrid models can effectively leverage domain knowledge and data-driven learning for radio propagation.
• The developed technique supports efficient spectrum management and interference mitigation in shared frequency bands.