LSTM-Based Spectrum Occupancy Prediction Performance with Thin Plate Spline Model

Prediction-based Dynamic Spectrum Access (DSA) in Cognitive Radio (CR) is a promising approach for efficient frequency spectrum utilisation for fast and reliable next generation of wireless communication. This led us to meticulously examine the relation of spectrum occupancy characteristics on prediction performance to accurately predict spectral opportunities in a dynamic radio environment. Our research involves utilising a Long-Short-Term Memory (LSTM)-based deep learning model to predict binary spectrum occupancy, which is characterised by the discrete-time Markov process. The LSTM prediction error is analysed and modelled using a Thin-Plate Spline (TPS) model for arbitrary binary Markovian model parameters. The proposed TPS model for prediction error is presented here along with the model verification results, which show a clear correlation between prediction performance and channel occupancy patterns and gives superior accuracy for conveniently computing the prediction error under the proposed settings.