Research On Underwater Acoustic OFDM Signal Receiver Based On Model And Data Drive

The ocean occupies the vast majority of the earth’s area,and the expansive ocean contains a wealth of various resources.Our country has a long coastline and a large sea area,so deepening ocean research is an important development strategy for our country.Underwater communication technology can solve the most basic information transmission problems in ocean research.Sound waves have lower attenuation and scattering effects underwater,so underwater long-distance communication mainly uses Underwater Acoustic(UWA)communication technology.UWA channel has various noise impacts,Inter-Symbol Interference(ISI)caused by multipath effects,and severe Doppler and delay spread phenomena.Orthogonal Frequency Division Multiplexing(OFDM)technology can resist the influence of ISI,and is often used in UWA communication.Scholars usually transfer the channel models in wireless communication to underwater communication and improve it in combination with the characteristics of UWA channel,but these approaches will lead to a large deviation between the results of many field experiments and the simulation results during modeling.The method based on deep learning can be combined with the classic wireless communication model,and the UWA channel data measured in the field can be used for training to obtain a model that is more suitable for the UWA channel characteristics.In this paper,a channel estimation and a signal equalization module are designed based on deep learning technology,and a receiver suitable for UWA-OFDM communication system is built.The main contributions of this paper can be summarized in the following four aspects:1.In order to solve the problems of low estimation accuracy in traditional channel estimation algorithms or the need to rely on prior information that is difficult to obtain in practical application,this paper builds a super-resolution network called UESNet based on the deep learning method for channel estimation.The main improvements of the network can be summarized as follows:(1)The architecture of the encoder and decoder is used to design the network,so that the network can fuse the information of different scales of the channel image.(2)Use the residual structure of "batch normalization layer + convolution layer + activation function layer" and the PRe LU activation function to optimize the network performance.2.In order to solve the problem that the large Doppler frequency shift in the UWA channel causes serious Inter-Carrier Interference(ICI)in the OFDM signal,which leads to the obvious degradation of the performance of the traditional equalization algorithm,this paper proposes a novel network structure named UDNet,which for frequency domain equalization of UWA-OFDM systems.UDNet uses a deep network to iteratively expand the Maximum Likelihood(ML)equalization algorithm,and adds trainable parameters to improve the performance of the classical equalization algorithm.The improvement of the network focuses on three aspects:(1)Considering that the constellation map identification problem in signal equalization is a classification problem,one-hot encoding and classification structure are applied to UDNet,and the minimum Kullback-Leibler(KL)divergence criterion is used to Train the network such that the soft outputs of the network are in probabilistic form.(2)Add more residual connection structures across network layers to optimize network performance.(3) Introduce a sliding structure based on the banded approximation of the channel matrix to reduce network computation and adapt to signals of different lengths.3.In order to enable the constructed deep learning network to learn the characteristics of the actual UWA channel,this paper uses the measured UWA channel data set as much as possible for training,and uses the channel data set with different characteristics to evaluate the performance of the proposed network.The experimental results show that both UESNet and UDNet designed based on deep learning technology achieve better performance than traditional algorithms in different UWA channel scenarios.4.In order to test the performance of the receiver composed of UESNet and UDNet,this paper builds a pool experiment system to conduct the UWA communication measurement experiment.The experimental results show that the receiver can adapt to the real underwater environment and has excellent information recovery effect.