SDRA'23 - 09 - Stefan Scholl, DC9ST: Radio Signal Identification with Deep Learning in RW Operation
Radio signal identification is the task of detecting the mode or type of an unknown RF signal, e.g. Morse code, SSB voice and RTTY. Recently, deep learning has been investigated as a new promising approach for radio signal identification, where a neural network is trained on large amounts of example signals. However, trained neural networks for RF signals are often only evaluated in academic settings with well-defined synthetic test signals (generated by software simulation) and no data from practical real-world applications. These neural networks may provide good results for synthetic data – but fail, when they face practical operation. This work investigates how training signals need to be designed in order to successfully perform mode identification also in practice. The resulting neural network can identify 20 different shortwave radio signal modes and achieves an accuracy of up to 95% for real-world signals.
Camera & Edit: FurStreaming
Sebastian Kipp, DL5WN
Marc Diensberg, DO1BOL
Torben Hellige