A Temporal Convolutional Network Based on Bayesian Optimization for Frequency Hopping Prediction
Zhe Deng, Ke Lai, Jing Lei · 2022
Frequency hopping (FH) spectrum prediction is challenging because of the discrepancy between the observed FH sequences and true values. Temporal convolutional network (TCN) has excellent sequence modeling ability due to its causal convolution structure and stacked dilated convolutional layers. In this paper, the FH sequence observation process is modeled with the state-space model, and the TCN is applied to FH sequence prediction with Bayesian hyper-parameter optimization. Furthermore, the performance of the proposed TCN model is evaluated in terms of prediction accuracy, training speed, and training stability in different types of FH sequences. Simulation results demonstrate the comprehensive performance of the proposed TCN-based FH prediction, which is suitable for the practical use of FH sequence prediction.