Distributed deep learning-based signal classification for time-frequency synchronization in wireless networks

In this paper, we propose a novel distributed deep learning (DL) network for signal classification to achieve accurate time-frequency synchronization in wireless communication networks. Restricted by the non-random signal generated by the transmitter, the existing time-frequency synchronization algorithms cannot obtain reliable synchronization results. With the development of DL, feature-based signal classification has become an effective way to improve the precision of time-frequency synchronization. Considering the multipath effect in the practical propagation environments, a distributed DL network for wireless communication is proposed, where the received signal is classified by the distributed DL network. Specifically, preprocessed images are obtained by applying the discrete Fourier transform (DFT) and short-time Fourier transform (STFT) to facilitate the signal feature extraction. Convolutional neural network (CNN) architecture is designed for feature extraction on the preprocessed images. Meanwhile, a feature fusion module based on attention mechanisms is proposed to fuse data adaptively and obtain the final signal classification results. Simulation results show that the proposed distributed DL network can achieve higher classification accuracy than traditional DL networks under multiple modulation schemes. More importantly, reliable signal classification can effectively improve the performance of time-frequency synchronization in wireless communication networks.