DANN-Based Radar Signal Waveform Recognition Under Scenario Variation Conditions

A radar waveform recognition method based on domain adversarial neural network (DANN) is proposed. The method aims to address the challenges of the low recognition rate of radar waveforms under the condition of scene change, the difficulty of resetting the dataset, and the large cost of re-training by traditional recognition methods. First, the signal time-domain waveform is converted into a two-dimensional TFI by the Choi-Williams distribution (CWD) to efficiently extract the characteristics of radar waveforms, and subsequently a DANN model for radar waveform recognition for scene transformation is proposed. Simulation experiments were conducted to classify and recognize six radar waveforms in various scenarios, thereby verifying the effectiveness of the model. The experimental outcomes demonstrate that the overall recognition rate of the signals is enhanced by 11 to 20 percentage points under conditions of scene transformation, the three types of signals, namely BPSK, Frank, and LFM, the recognition rate is increased by 29 to 46 percentage points. When the signal-to-noise ratio (SNR) is -6 dB, the waveform recognition rate reaches 91.4%, thereby improving the recognition rate of radar signal waveform under scene transformation conditions.