Design of An End-to-End Autoencoder for Maritime Communication System towards Internet of Vessels

The Internet of Vessels has drawn more and more attention because the shipping efficiency and shipping safety can be increased, and the development of world transportation can easily be augment. Deep learning (DL) has been recognized as very promising for solving bottlenecks in the evolution of next generation communication systems. To adapt to the particularity of the marine environment, Rician fading channel is selected as the channel model for maritime communication. We propose a Convolutional Neural Network based end-to-end Autoencoder (CNN-AE) for the maritime communication system. A 1-dimensional convolution layers is introduced in AE. We investigate the characteristics of the maritime channel and optimize performance by redesigning the parameters. Compared with the existing schemes, the proposed CNN-AE maritime communication system not only inherits the characteristics of traditional AE system which can optimize all modules jointly, but also has the merits of CNN's local connection which can support more information. Simulation results show that under the same condition of input bit lengths and coding rates, when the values of Ricean K-factor increase, the Block Error Rates (BLER) of the system decrease. When the burst noise is added, the BLER grows as the noise increases. When the Ricean K-factor is the same, the BLER decreases as the coding rate decreases.