Modeling GSM Spectrum Occupancy Using Time Series Analysis: The case of Ethio telecom
No Thumbnail Available
Date
2021-09
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Spectrum occupancy models can help you make better use of the radio spectrum. It has
also been extensively researched in recent decades as it is critical for developing new
regulations for spectrum allocation for future technologies as well as monitoring the activities
that take place on that spectrum. Understanding the amount of available spectrum
is critical for future wireless technologies that want to address the so-called spectrum
scarcity issue. Spectrum occupancy measurements provide critical data for frequency
planning and optimization, as well as assist in smart decision-making. The goal of time
series modeling is to collect and thoroughly examine previous data from a time series
in order to construct an appropriate model that accurately captures the series’ intrinsic
structure. This thesis examines three types of time series analysis methods: Holt-winters,
Seasonal Auto Regressive Integrated Moving Average (SARIMA), and SARIMA eXogenous
regresses (SARIMAX) based models, as well as their inherent prediction strengths
and weaknesses. Time series modeling principles such as trend, stationarity, seasonality,
residual, and so on have also been covered. To assess the accuracy rate, we fitted multiple
models to a time series using five primary metrics. Among the methods used are
mean square error, mean absolute error (MAE), root-mean-square error, mean absolute
percentage error, and R-squared. At 1800MHz, the maximum spectrum occupancy is
60.35%, and at 900MHz, it is 44.71%. For 900MHz MAE, the SARIMAX model produced
better predictions (35.34% and 50.1% lower than the SARIMA and Holt-Winter
models, respectively), while for 1800MHz, the SARIMAX model produced 42.6% and
52.6% lower than the SARIMA and Holt-Winter models, respectively
Description
Keywords
Spectrum Occupancy, Holt-Winters, SARIMA, SARIMAX