An Empirical Study of Interference Features in Licensed and Unlicensed Bands for Intelligent Spectrum Management

Abstract

In this paper, we first present the results of an empirical study of the comparative statistics of fourteen interference features in licensed and unlicensed bands in a selected route in downtown Worcester, MA. Then, we benefit from these features to train a machine learning algorithm to predict the availability of the channels for intelligent spectrum access for a vehicle. The main component of the vehicular interference monitoring system is an ultra wideband programmable 26GHz Agilent E4407 spectrum analyzer, interfaced with a GPS device to record the location of measurements, and a laptop to store the results in a centralized database. With this measurement system loaded in a car we drive in a selected path to monitor the interference in 1.9-2.5 GHz frequency band, which includes high traffic density neighboring licensed and unlicensed bands. With the empirical monitored interference, we study the statistical behavior of fourteen interference features logically categorized into four classes: interference intensity, correlation properties, spectrum occupancy, and Doppler spectrum in licensed and unlicensed bands. Finally, We benefit from these features to train a machine learning algorithm to predict the availability of the licensed and unlicensed bands for vehicular network access to the fixed backbone network infrastructure.