Development of Neural Network-Based Spectrum Prediction Schemes for Cognitive Wireless Communication: A Case Study of Ilorin, North Central, Nigeria

The concept of spectrum prediction has become necessary as a result of the cost and time intensiveness of continuous spectrum measurements essential to the deployment of cognitive radio network riding on 5G and beyond. Spectrum prediction forecasts future channel states based on previously observed data from spectrum sensing. In this study, long-short term memory (LSTM), Artificial Neural Network-LSTM (ANN-LSTM), and ANN techniques are used for the prediction of spectrum channel duration metric in the 900/1800 spectrum bands, hitherto allocated for GSM services. Ilorin, an urban area in North central region of Nigeria was used as a case study. The root mean square errors (RMSEs) for the artificial neural networks (ANN), LSTM and ANN-LSTM for the GSM 900 spectrum were 3.126, 3.119 and 2.964 respectively with a regression coefficient of 0.9271 at a 60-minute observation time. The RMSEs for the ANN, LSTM, and ANN-LSTM for the GSM 1800 spectrum data were 3.577, 3.0428 and 2.791 respectively. The regression coefficient utilizing the ANN-LSTM with the real data set is 0.9147 at a 60-minute observation time. The results showed that ANN-LSTM technique performed better than either ANN or LSTM standing alone.