Prediction of path loss in coastal and vegetative environments with deep learning at 5G sub-6 GHz

Path loss prediction is quite important for the network performance of the wireless sensors, quality of cellular communication-based link budget, and optimization of coverage planning in mobile networks. With the development of 5G technology, even though different log-distance path loss models are generated for these, new-developed methods are required to make models more flexible and accurate for complex environments. In this study, for different coastal terrains (air-dry sand, wet sand, small pebble, big pebble) and various vegetable areas (pine, orange, cherry, and walnut), the principle and procedure of deep learning-based path loss prediction are provided in 3.5 GHz, 3.8 GHz, and 4.2 GHz in the 5G frequency zone, as a novelty. For this, recurrent neural network (RNN) and long short-term memory (LSTM) methods are proposed. The test sample number is 240 since 20% of all datasets (1200) are test data. In general, path loss for coastal terrains is higher than path loss for vegetation areas with an average of 5 dB. For both coastal terrains and vegetation areas, the recurrent neural network method predicts better than the long short-term memory method. Consequently, for both coastal terrains and vegetation areas, RNN models with R² values of 0.9677 and 0.9042, respectively, are preferred.