Path Loss Model Estimation at Indoor Environment by Using Deep Neural Network and CatBoost for Wireless Application

This paper introduces a novel method for estimating the path loss value in indoor scenarios. It uses a combination of electromagnetic calculation and machine learning. Using electromagnetic software, we design the indoor environment and calculate path loss between the receiver and the transmitter antennas, which are situated randomly with respect to the transmitters for better coverage. The transmitter antenna is modeled by employing the 3GPP standards. The indoor NLOS and LOS scenarios are measured over a 47-m separation distance between the RX and TX antenna locations. As it comes to fitting real measured data that has been acquired via measurement campaigns, the suggested models perform better overall. The achieved data is used to predict path loss using empirical models like FI and CI, and the mean absolute error percentages obtained are 5.87 and 6.61, respectively. After that, deep neural network and CatBoosting methods are employed to predict and estimate the PL precisely in the indoor environment. The mean absolute error percentages obtained are 4.4 and 3.3 for deep neural network and CatBoosting, respectively. Predicting PL, deep neural network, and CatBoosting methods outperform CI and FI by modelling complex, non-linear relationships, incorporating a broader range of features, adapting to diverse environments, and generalizing more effectively from data. These benefits are well-matched for modern wireless communication systems, where accurate estimation of signal loss is essential for maximizing network performance.