Hybrid Clustering and Deep Learning-based Spatio Temporal Analysis of Spectrum Utilization

No Thumbnail Available



Journal Title

Journal ISSN

Volume Title


Addis Ababa University


Radio spectrum is a finite resource, while the demand for wireless systems is increasing at an exponential rate. To meet this demand, new generations of cellular networks were introduced.Spectrum utilization of cellular bands is analyzed widely using spectrum measurements. Knowledge of spectrum utilization will help operators like Ethio telecom to understand and plan band usage. In this thesis, using the K-means algorithm and Deep learning algorithms, namely Convolutional Neural Network (CNN) and Long Short Term Memory (LSTM), downlink Global System for Mobile Communication (GSM) 900 spectrum utilization is analyzed and modeled to know the spectrum utilization of Ethio telecom. The data is collected from Addis Ababa 639 GSM base stations. Spectrum utilization is modeled using CNN and LSTM algorithms for clustered and non-clustered data. Because of the differences in base station behavior, clustering base stations is done and model the spectrum utilization of the base stations in each cluster. Our results show that the GSM 900 downlink spectrum is not utilized optimally. The highest observed average spectrum utilization was 71%, with the lowest observed average spectrum utilization being 1.4%. The model developed for the cluster data using the CNN algorithm can model spectrum utilization with an RMSE value of 0.58 and this model can predict the next twenty-four-hour base station spectrum utilization with an RMSE value of 1.04.



Spectrum Utilization, GSM900, Downlink, K-means, LSTM, CNN