LTE Primary User Modeling Using a Hybrid ARIMA/NARX Neural Network Model in CR

In this paper, we study cognitive radio and propose mathematical and behavioral modeling of the LTE downlink signal of primary users in order to facilitate spectrum decisions of the secondary users. Auto-regressive integrated moving average (ARIMA) time series model, non-linear autoregressive exogenous input (NARX) neural network (NN) are investigated and a hybrid system is proposed. Using the proposed model instead of the ARIMA model results in a decrease in the average mean square error (MSE) by 23.23% in addition to a decrease of 9.68% in the number of the classifier's misclassification. The hybrid model yields a 29.25% improvement in the average MSE compared to the random NN, it also offers an advantage of 15.15% decrease in the number of misclassifications. Results show that using predictive models and previous instances of sensed data, the behavior of the primary user could be captured and the secondary user could decide about usable parts of the spectrum accordingly.