Learning in vehicular dynamic spectrum access networks: Opportunities and challenges

Abstract

In this paper, we discuss various ways of incorporating machine learning techniques in the area of dynamic spectrum access, especially in vehicle communications, how learning can benefit multiple applications in vehicle communications, and the specific requirements on the implementation in this field. In particular, we describe an architecture for optimizing the overall performance of vehicular dynamic spectrum access (VDSA) networks. Due to the high level of mobility for vehicles operating under highway conditions, coupled with spatially variant spectrum allocation across a large geographical region, we envision that future vehicular communications will employ a form of dynamic spectrum access (DSA) in order to facilitate wireless transmissions between vehicles and with roadside infrastructure. Moreover, the VDSA concept will be realized via a combination of software-defined radio (SDR) technology, spectral occupancy databases, and machine learning techniques for enabling network automation. A vehicular networking scenario possesses the potential to be substantially different relative to a generic mobile scenario with respect to the particular type of mobility involved, the predictable trajectories of the vehicular traffic, and the overall scale of the network range. Consequently, this architecture is designed to enable VDSA in a more flexible wireless spectrum environment by leveraging the cognitive radio concept and existing wireless spectrum databases actively being developed while simultaneously being compatible with current spectrum regulations.