Genetic Algorithm-Holt-Winters Based Minute Spectrum Occupancy Prediction: An Investigation

Nazmat Toyin Surajudeen-Bakinde (Department of Electrical and Electronics Engineering, University of Ilorin, PMB 1515 Tanke, Ilorin, Kwara State, Nigeria)
Frederick Ojiemhende Ehiagwina (Department of Electrical/Electronic Engineering, The Federal Polytechnic Offa, Osogbo-Ajesse-ipo Road, PMB 420, Offa, Kwara State, Nigeria)
Akindele Segun Afolabi (Department of Electrical and Electronics Engineering, University of Ilorin, PMB 1515 Tanke, Ilorin, Kwara State, Nigeria)
Ayinde Mohammed Usman (Department of Electrical & Electronic Engineering, Stellenbosch University, Private Bag X1, Matieland, 7602, Stellenbosch Central, South Africa)

Journal of Engineering and Technological Sciences
Vol. 54 No. 6 (2022)

In this research, the suitability of a genetic algorithm (GA) modified Holt-Winters (HW) exponential model for the prediction of spectrum occupancy data was investigated. Firstly, a description of spectrum measurement that was done during a two-week duration at locations (8.511 °N, 4.594 °E) and (8.487 °N, 4.573 °E) of the 900 MHz and 1800 MHz bands is given. In computing the spectrum duty cycle, different decision thresholds per band link were employed due to differing noise levels. A frequency point with a power spectral density less than the decision threshold was considered unoccupied and was assigned a value of 0, while a frequency point with a power spectral density larger than the decision threshold was considered occupied and was assigned a value of 1. Secondly, the spectrum duty cycle was used in the evaluation of the forecast behavior of the forecasting methods. The HW approach uses exponential smoothing to encode the spectrum data and uses them to forecast typical values in present and future states. The mean square error (MSE) of prediction was minimized using a GA by iteratively adjusting the HW discount factors to improve the forecast accuracy. A decrease in MSE of between 8.33 to 44.6% was observed.