Research on Communication Signal Modulation Recognition Method based on Optimized VGG16 Network

Abstract

With the advancement of communication technology, Multicarrier Modulation (MCM) has emerged as a crucial technique for mitigating interference and achieving high-rate data transmission. The first step in demodulating signals during communication is to identify the modulation scheme. Cooperative communication utilizes pilot sequences for this purpose; however, non-cooperative environments rely on techniques such as modulation scheme recognition, channel estimation, and carrier frequency estimation. Leveraging the powerful feature extraction capabilities of deep learning, modulation recognition technology based on deep learning can automatically extract signal features to determine the modulation scheme. By employing the VGG16 convolutional neural network and utilizing a constructed dataset for the recognition of OFDM signals, a recognition accuracy of 91.9% was achieved on the dataset presented in this paper. An optimized version of the VGG16 neural network, referred to as VGG16R, is proposed. This network architecture treats convolutional layers as convolutional groups, adds a Batch Normalization (BN) layer after each convolutional layer, and introduces branches that sum the input image with the results of the convolutional groups, thereby forming a new feature set.