Deep Learning Modeling of a WBAN-MIMO Channel in Underground Mine

Abstract

In this paper, an efficient model of the channel matrix is developed for 2 &#x00D7; 2 Wireless Body Area Network Multiple Input Output (WBAN-MIMO) system, based on deep learning algorithms. The model is composed of three deep learning algorithms. Moreover, the model predicts simultaneously the channel matrix <i>H</i> in underground mine and identifies the position of the collected data in both Line of Sight (LoS) and Non-Line of the Sight (NLoS) scenarios. The model is trained and evaluated using the magnitude and phase of the collected data in an underground mine environment within the frequency range of 2.3 GHz &#x2013; 2.5 GHz. These measurements, conducted with different antenna configurations in LoS and NLoS scenarios, constitute an input to the model. The latest predicts the channel matrix <i>H</i> with the position and identifies whether the channel is LoS or NLoS. Finally, the path loss and the channel impulse response models are compared with the measurements-based ones. The modeled channel prediction exhibited lower Root Mean Square Error (RMSE) for channel prediction and high classification accuracy for LoS-NLoS and position identification, respectively. The numerical results reveal that the deep learning MIMO WBAN modeling offers a powerful solution for future wireless systems in underground mine environments.