Satellite Data Transmission Method for Deep Learning-Based AutoEncoders

To improve the accuracy of satellite data transmission, deep learning (DL) are applied to the Ka-band satellite communication system under complex channel conditions, and a satellite data transmission method for deep learning-based auto encoder (AE) is proposed in this paper. Specifically, the transmitter and receiver are integrated with deep neural network (DNN), and an interference layer is used to simulate the complex channels that may occur during the satellite data transmission. Finally, we trained the entire network iteratively. And optimized the entire network performance. The result exposed that the proposed method achieve an order of magnitude higher than traditional BPSK modulation, coherent demodulation, Hamming code (HC) and hard-decision decoding method, and two orders of magnitude higher than Hamming code and maximum likelihood estimation (MLE) methods. It has been verified that the new network structure of the AE can improve the data transmission accuracy, thus providing new avenues for promoting the application of deep learning in wireless transmission, as well as new ideas for satellite data transmission.