Spatiotemporal Forecasting of Wireless Coverage and Frequency Availability with Sparse Geo-location Spectrum Databases

Abstract

The television spectrum can contain unused frequencies or channels called white spaces. The white spaces can be managed to provide internet access in coordination with surrounding TV channels to avoid interference. Geo-location databases are helpful for dynamically shared frequencies when updated and complete. In real life, the spectrum availability for a secondary user lacks numerous information; hence, it is sparse. This paper aims to forecast wireless coverage and frequency availability in such sparse geolocation databases. Spatiotemporal models are formulated in this paper to forecast wireless coverage and frequency availability in sparse geo-location spectrum databases. Eight channels are explored by this study, and the data used are gathered from TV program guides sourced online and to the best knowledge of the researcher. The forecasting models are evaluated using accuracy, precision, recall, and F1 score. Spatiotemporal logistic models had a decent accuracy of at least 84%. The linear VAR models have a decent accuracy except for predicting time. The formulated spatiotemporal logistic VAR model attained the highest accuracy of at least 94%.