Mixed Signal Recognition Network Based on FD-MCNN and BiLSTM

Abstract

Mixed-signal recognition in realistic wireless environments is challenging because weak signal components are often masked by stronger ones. To address this issue, this paper proposes a hierarchical recognition framework that combines multi-domain feature disentanglement with temporal dependency modeling. Specifically, the proposed network extracts complementary features from the time, frequency, modulation, and energy domains, enabling more robust representation of mixed signals under complex interference conditions. Based on these features, the framework first identifies the dominant signal component, then enhances the weak component to reduce the masking effect, and finally employs a bidirectional long short-term memory network (BiLSTM) network for temporal modeling and classification. Experiments with signal-to-noise ratio (SNR) ranging from 0 to 30 dB show that the proposed method can effectively recognize both strong and weak signal components while improving overall robustness. These results demonstrate the effectiveness of the proposed framework for mixed-signal recognition in interference-rich wireless environments.