Deep Neural Network Based on Long Short-Term Memory for Predictive Learning of Wireless Path Loss Datasets

Abstract

 This paper proposes and evaluates a deep learning approach using Long Short-Term Memory (LSTM) networks for the predictive modeling of wireless path loss. We developed the deep LSTM network architecture trained on measured signal path loss datasets. The model takes relevant environmental and geometrical features as input and predicts the path loss value. We compare the performance of the LSTM model under three prediction optimisation algorithms using standard evaluation metrics like Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). The algorithms include the stochastic gradient descent (SGDM), Root mean square propagation (RMSP) , and Adaptive moment estimation (ADAM). Results demonstrate that the deep LSTM network trained with the RMSP algorithm achieves superior predictive accuracy, effectively capturing complex propagation phenomena and environmental dependencies. The findings imply that an LSTM-based deep learning method trained with the RMSP algorithm offers a robust and potentially adaptive solution for accurate path loss prediction in various wireless environments.