Neural Network Based Spectrum Prediction in Land Mobile Radio Bands for IoT deployments

Aim: We seek to assess the performance of time delay neural networks (TDNN), one of the topologies designed for time series prediction, to characterize spectrum occupancy in multiple time horizons in Land Mobile Radio bands. This could lead to dynamic spectrum allocation methods to address potential spectrum shortages facing Internet of Things (IoT) deployments. Background: ANNs are a popular choice for spectrum prediction. Traditionally, ARIMA models have been at the forefront of forecasting and prediction but ANNs that learn from time series have demonstrated good performance using both simulated datasets and real-life data collected in the cellular bands. Methodology: We use three prediction models, a baseline which simply delays the time series, a seasonal ARIMA model and a TDNN. We test their performance on an hourly dataset in LMR bands collected in Ottawa, Canada between the dates of October 2016 and April 2017. Results: We demonstrate that TDNN yields improvements over seasonal ARIMA models in predicting short time horizons. Conclusions: The TDNN based prediction models that are designed to work with time series data provide a better alternative for accurately predicting spectrum occupancy in bands that exhibit similar characteristics to LMR channels, especially as the forecast horizon gets longer.