In order to use the EKF for the DRNN to estimate the optimal weights, V was regarded as the state of a dynamic system.
The dynamic equations of (7) for the perfect DRNN are
The training of the DRNN was carried out from the initial guess [caret.
We carried out the experiment for the WiFi channel in order to validate the spectrum prediction based on the DRNN.
predictor-2], predicted the multiple time slot spectrum with the DRNN and selected the channel with the largest number of idle slots.
Increasing the number of hidden layers of the DRNN reduces the prediction error.