| With the large-scale deployment of 5G and the emergence of the Internet of Things,the access of a large number of communication devices has caused communication congestion and interruptions,and traditional wireless communication systems are facing unprecedented challenges.The vigorous development of the marine economy makes it necessary to realize seamless,efficient,reliable,and full-coverage communication which can meet the needs of increasingly frequent maritime communications.However,affected by various complex meteorological environments such as wind and waves,the development of maritime wireless communication is obviously lagging behind the land.How to build a high-speed,reliable,low-latency,and low-cost new maritime wireless communication system is a critical issue to build maritime power in China.The main research contents of this thesis are as follows:Firstly,we systematically analyze the influence of earth curvature,wave motion,and shadow effect on the propagation of electromagnetic waves at sea.The Doppler frequency shift effect which is caused by the relative motion between the transmitting and receiving antennas due to ship motion is also considered.Furthermore,we consider the influence of multipath components formed by the direct path,the specular reflection path,and the diffuse reflection path in the glistening surface on the received signal strength,and construct the three-path fading channel model at sea.Secondly,we propose a multidimensional feature learning-based automatic modulation recognition model named IRLNet.The model aims at solving the confusion among high-order modulation modes,balancing complexity against recognition accuracy,and strengthening the antiinterference ability.By building the dual-channel parallel deep neural network constructed by the improved residual stack and long short-term memory,the model extracts spatiotemporal features from multidimensional In-phase and Quadrature(IQ)information,Amplitude and Phase(AP)information,which enrich and enhance input information,accelerate model convergence,and improve modulation recognition accuracy and generalization capabilities.To verify the utility and generalization ability of the IRLNet model,experiments and comparative analysis are conducted on the Radio ML2016.10 B,Radio ML 2018.01 A,and Hisar Mod 2019.1.The experimental results show that IRLNet has stable performance and low space-time complexity in the training stage.Under the condition of a high signal-to-noise ratio,the recognition accuracy reaches more than93%.By introducing transfer learning,the efficiency of model retraining is improved and the robustness of IRLNet to various modulation methods and channel environments is also verified.Finally,the maritime communication receiver(MCR)is proposed for complicated maritime communication by using a deep neural network to replace all the modules in traditional receivers,such as carrier and symbol synchronization,channel estimation and equalization,demodulation,and decoding.MCR avoids the problem of error accumulation in subsequent processing modules caused by errors in the pre-processing module in the traditional serial processing method and achieves the optimal global performance of the receiver.Furthermore,multiple binary classifiers are also proposed to replace a single multi-classifier,which avoids the problem of high model complexity and slow model training due to too-long information bitstreams.The experimental results show that MCR can effectively recover information in various complex maritime environments and is robust to various complex communication environments. |
