Path loss prediction for vehicle-to-infrastructure communications via synesthesia of machines (SoM)

In this paper, a new real-time path loss prediction model based on multi-modal sensory data is proposed to enhance the accuracy of path loss prediction in vehicular communication scenarios. A new multimodal data set containing communication and sensory data is constructed based on simulation platforms. The data set is constructed for intelligent sensing-communication integration in urban vehicular crossroads scenarios. Based on the constructed data set, the mapping relationship between physical space and electromagnetic space is explored. Furthermore, path loss prediction is achieved with environmental information via multi-modal sensory data. Simulation results show that the proposed path loss prediction model is validated, which achieves a mean squared error of 1.9283 × 10⁻⁶. The proposed model improves the accuracy by 2 orders of magnitude over 3GPP TR 38.901 channel models. Compared to the artificial neural network, support vector regression, random forest, and gradient tree boosting, the proposed model achieves the highest accuracy. Finally, the effectiveness of multi-modal sensory data fusion in path loss prediction for vehicular communication scenarios is validated, which shows a 19.8% improvement in accuracy compared to predictions based on uni-modal data.