Identification of Modulation Type Based on Time-Fequency Analysis and Neural Networks
Weinan Wang, Youyun Xu, Jian Chen · 2022
Modulation type recognition of signals belongs to the category of signal recognition technology, which can provide a basis for radio monitoring and band resource management, and plays an important role in applications such as signal detection, communication countermeasures and cognitive radio. Compared to traditional modulation recognition methods, deep learning based modulation recognition can automatically learn the characteristics of the modulated signals and improve the recognition performance in complex environments. Therefore, this paper proposes an identification method based on SPWVD transform and deep residual network for the identification of modulation types. The proposed method first uses the SPWVD transform to extract the features of the signals and generate a grey-scale map. Secondly, the grey-scale map is compressed and transformed into a network. ResNet is used to learn the time-frequency characteristics of the signals to identify 2ASK, 2FSK, 2DPSK, LFM, VFM and FH six modulation types. Compared with existing algorithms, this method can effectively reduce the influence of noise interference and cross terms; increase the depth of the network by using the characteristics of residual blocks can improve the recognition performance of the network. Experimental results demonstrate the identification accuracy of 94% or better was acquired at SNR of 0 dB.