A Machine Learning Approach for Analysis of Spectrum Availability in KOSOVO based on Experimental Measurements

Establishing the utilization level of spectrum resources is an essential step in evaluating the true availability of the frequency spectrum which could potentially become available for opportunistic use. Such assessments typically require extensive measurement campaigns, in order take into account various factors which may affect the availability of spectrum, such as frequency allocation at national level, and the particulars of the environment and terrain. In this study, we focus on the bands usually reserved for TV broadcasters, since studies show that these bands tend to be least efficiently utilized. In addition, we take advantage of a large set of data obtained from an extensive spectrum measurement campaign conducted over the entire territory of Kosovo, which enables us to attain a comprehensive picture of spectrum availability in TV bands. Spectrum sensing techniques using energy detectors are applied in order to determine whether the spectrum is utilized. However, due to the significant amount of measurement data collected, machine learning techniques can be leveraged to improve the accuracy of the approach, by improving the detection and false alarm rates. As an end result, a highly accurate spectrum availability map is obtained which can be in turn used by cognitive radio network to opportunistically access the spectrum.