Deep Learning for Path Loss Prediction in the 3.5 GHz CBRS Spectrum Band

Abstract

In the 3.5 GHz Citizens Broadband Radio Service (CBRS) band, accurate path loss prediction is very important to protect the incumbent from harmful interference caused by the lower tier users. The current CBRS standards developed by the Wireless Innovation Forum use the irregular terrain model (ITM), also known as the Longley-Rice model, for path loss calculation. However, the model does not include clutter data, and thus, it underestimates the path loss. This paper utilizes a model-aided deep learning (DL) technique with satellite images to improve path loss prediction. Numerical study shows that the proposed approach can achieve a 4.23 dB root mean square error (RMSE) and outperform the Longley-Rice model and some tuned or fitted propagation models.