Ray-Tracing MIMO Channel Dataset for Machine Learning Applied to V2V Communication

Abstract

Machine learning has become a powerful tool for improving vehicle-to-vehicle (V2V) communication systems, and in general requires large datasets for model training and assessment. However, creating large and realistic datasets using field measurements is challenging due to the large bandwidths involved and usage of multiple antennas. Simulations have been widely adopted to circumvent the relative high cost of measurement campaigns. This paper presents the development of a new public dataset for research within V2V scenarios, of machine learning algorithms that require the MIMO channel for simulation or emulation. The adopted methodology relies on realistic simulations of vehicles traffic in 3D virtual worlds. The paper also analyses the influence of key parameters in the ray-tracing simulation with respect to the tradeoff between accuracy and computational cost. Lastly, the paper discusses results of beam-selection for V2V using machine learning and the presented dataset.