A Study on Neural Network-Based Millimeter-Wave Path Loss Modeling for Indoor Application Scenarios

Millimeter-wave (mmWave), with its abundant spectral resources, has become a key technology for achieving the high-speed and low-latency goals of 5G and post-5G communications. As the carrier frequency increases, wireless channels exhibit strong frequency dependence in terms of delay, spatial angle, and loss characteristics. Given the decisive impact of channel characteristics on the performance limits of wireless systems, an in-depth exploration of wireless channel characteristics is a necessary prerequisite for the design and deployment of millimeter-wave wireless communication systems [1]. When the carrier frequency rises to the millimeter-wave range, the first to be affected is the loss characteristics. Due to propagation state transitions caused by dense scatterers and multipoint mobility, path loss undergoes rapid variations. Timely and accurate modeling of path loss presents a significant challenge. To address this, this paper investigates a millimeter-wave path loss model for indoor environments based on artificial neural networks. By leveraging the advantages of artificial neural networks in solving nonlinear problems, the path loss factors of large-scale path loss models in various environments are extracted as features. A neural network model was trained to classify line-of-sight (LOS), non-line-of-sight (NLOS), and combined LOS-NLOS areas. Based on predefined thresholds, the Close-In free space reference distance path loss model (CI) or the Floating-Intercept path loss model (FI) is selectively applied to model path loss in LOS and NLOS areas efficiently, enhancing the real-time performance and efficiency of channel measurement and modeling.