Parameter estimation of linear frequency modulated signals based on a Wigner–Ville distribution complex-valued convolutional neural network

Our work aims to address the problem of estimating the parameters of constant-amplitude, time-unsynchronized linear frequency-modulated (LFM) signals that have single or multiple components, which is a crucial task in electronic countermeasure techniques. A method for estimating the parameters, center frequency f0, and chirp rate μ of an LFM signal is proposed; the method is referred to as the Wigner–Ville distribution complex-valued convolutional neural network (WVD-CV-CNN). The method can be regarded as an application of neural networks for extracting parameter features from the signal spectrogram, wherein the CV-CNN is the main body of the network, which takes a complex-valued WVD matrix as the input and outputs several sets of estimated parameters. A performance analysis shows that the estimation accuracy and computational efficiency of the proposed method are significantly improved compared with those of the conventional methods. Further, the proposed method shows strong robustness to changes in modulation parameters. We apply the CV-CNN to other spectrograms and prove compatibility of the WVD and CV-CNN by comparison. We also demonstrate that the estimation accuracy of the proposed method is robust against cross interference on the WVD. Our study shows the advantages of using deep learning systems in signal parameter estimation.