ML-supported Spectrum Sensing for Underlay Vehicular D2D Communications in Urban Environments

Device-to-device communications (D2D) have been an important feature of cellular networks since they were first introduced in LTE Release 12. D2D enables more efficient use of spectrum resources and better data rates through offloading of cellular traffic onto short-range D2D links. Notably, D2D has been key in supporting vehicle-to-vehicle (V2V) and vehicle-to-everything (V2X) cellular communication in vehicular networks. Underlay D2D communications in which devices access cellular spectrum in an opportunistic manner, face several challenges such as efficient spectrum sensing access and interference. In this paper we address the problem of identifying potential cellular spectrum resources which can be used opportunistically to conduct D2D communications between vehicles in an urban vehicular network. To achieve this, we adopt a machine-learning approach to cooperatively sense the spectrum by using the observations of a cluster of vehicles to identify which spectrum resources can be used to establish V2V links. The performance of the solution is evaluated in a single cell network operating in an urban environment, with realistic modeling of vehicular mobility, using numerical simulations. The performance of our approach is evaluated in terms of spectrum efficiency and network throughput.