Research on UAV Remote Control Signal Recognition Based on Time-frequency Map

The actual collected UAV (Unmanned Aerial Vehicle) signal is converted into CWD (Choi-Williams Distribution), and the maximum time-frequency spectrum of each part of the signal is spliced to make the characteristics of the frequency-hopping signal more obvious, and the adaptive method based on energy statistics is used to set the denoising threshold. According to the difference of correlation between remote control signal and image transmission signal in time and frequency domain, the image transmission signal and remote control signal in the original signal are innovatively separated into low-rank matrix and sparse matrix by using low-rank matrix recovery method, so as to achieve effective extraction of time-frequency characteristics of remote control signal. The convolution neural network CNN (Convolutional Neural Network) is used to input the time-frequency spectrum obtained by CWD of the collected signal into the trained CNN network to complete the sorting of UAV models. This method realizes blind classification of UAV signals without acquiring the target signal parameter characteristics, and the accuracy of test set reaches 89.19%; Further use the LSTM (Long and Short Term Memory) network to convert the remote control signal into a time series, and express the time series in a five-element form, which can make the mode representation of the data closer to the trend change of the original data, and the accuracy of the final test set reaches 92.48%.