Structural inversion of radar emitter based on stacked convolutional autoencoder and deep neural network

As various new radar systems are put into use in complex electromagnetic environments, the extraction of only the time-domain parameters of radar signals cannot achieve the accurate cognition of radar emitters. For this reason, a radar emitter structural inversion method is proposed based on a stacked convolutional autoencoder and deep neural network (SCAE-DNN) to complete the two processes of forward modelling and inversion. The method completes the work of modelling from the structure to radar signals via forward calculations and subsequently obtains the structure via structural inversion. The modelling of different radar radiation sources should be realised through device-level simulation to obtain radar signals with radio frequency (RF) structural characteristics. There is a mapping relationship between the RF structural characteristics and the structure of the radar emitter, and this mapping relationship will not be affected by differences in the time, frequency, and spatial domains. Novel feature extraction approaches are then presented, in which SCAE is used to replace the cumbersome calculation in the traditional algorithm to extract the RF structural characteristics. Finally, it is demonstrated that the inversion of the radar emitter structure can be realised by using the RF structural characteristics via DNN. Experimental results show that this method can accurately invert the radar emitter structure and has a strong generalisation ability for multiple modulated radar signals with additive white Gaussian noise with different signal-to-noise ratios (SNRs).