Crossterm-Free Time-Frequency Representation Exploiting Deep Convolutional Neural Network

Bilinear time-frequency representations (TFRs) provide high-resolution time-varying frequency characteristics of nonstationary signals. However, they suffer from crossterms due to the bilinear nature. Existing crossterm-reduced TFRs focus on optimized kernel design which amounts to low-pass weighting or masking in the ambiguity function domain. Optimization of fixed and adaptive kernels are difficult, particularly for complicated signals whose autoterms and crossterms overlap in the ambiguity function. In this letter, we develop a new method to offer high-resolution TFRs of nonstationary signals with crossterms effectively suppressed. The proposed method exploits a deep convolutional neural network which is trained to construct crossterm-free TFRs. The effectiveness of the proposed method is verified by simulation results which clearly show desirable autoterm preservation and crossterm mitigation capabilities. The proposed technique significantly outperforms state-of-the-art time-frequency analysis algorithms based on adaptive kernels and compressive sensing techniques.