Spectrum Prediction for Protocol-Aware RF Jamming

Neutralizing a drone using a protocol-aware RF jammer requires precise knowledge of the occupied spectrum in the time and frequency domains. This paper aims to develop an automatic spectrum prediction framework utilizing the Convolutional Neural Network (CNN) and Long Short Term Memory (LSTM) models. We generate a synthetic dataset using the commonly used drone signal properties and parameters, and evaluate the prediction performance under several realistic scenarios. Our experiment shows that the CNN-LSTM model can accurately predict future time and frequency sequences by using the spectrogram matrix as the input. We obtained better prediction performance with lower computational costs with our framework compared to the existing frameworks. Furthermore, we show that the CNN-LSTM model can predict future time-frequency sequences of unseen hopping rates and patterns when using transfer learning. The performance validation is also performed using real drone RF signals. Furthermore, we present a two-stage spectrum prediction approach that achieves excellent performance by offering higher frequency resolutions while maintaining a lower computational cost.