Artificial Intelligence-Based Classification of Multipath Types for Vehicular Localization in Dense Environments

Abstract

Multipath-geometry is the most promising approach for vehicular localization in line of sight (LOS) and non-line of sight (NLOS) scenarios. In such approach, identifying the type of the propagated multipath (MP) is an important pre-required process. However, identifying the type of the MP in dense multipath environments is challenging. The previous works proposed iterative methods for this task. The iterative methods have their limitations such as required more in-depth analysis and high complexity of computation. However, leveraging artificial intelligence advantages, a lower complexity identification method is proposed in this work. We utilized supervised learning algorithms to distinguish the direct link, first-order, and higher-order MPs of millimeter-Wave Vehicle-to-Infrastructure communication. In particular, four models namely KNN, and SVM, MLP, and LSTM have been applied. The characteristics of the received signal paths including received signal strength and elevation and azimuth angle of arrival are considered as features of the training dataset. The results showed that the accuracy rates of the classification are ranged between 96.70% and 84.0%. The best accuracy rate was 96.70% obtained by LSTM, followed by 94.47 % obtained by MLP. Whereas, 93.67% and 84.0% accuracy rats were achieved by KNN and SVM respectively.