Research On Intelligent Modulation Mode Recognition Based On Spatiotemporal Feature Fusion

Marine informatization has great strategic significance in national informatization,and is the primary driving force and strong support for the development of our country's future marine industry.As the core technology of marine information acquisition and transmission,underwater acoustic communication technology needs further research and development.However,the underwater acoustic channel is complex and changeable,and single modulation type cannot adapt to the changeable underwater acoustic channel.The use of adaptive coding technology can select the optimal modulation type according to the channel state and improve the data transmission efficiency,but the underwater noise interference is serious,and it is easy to cause handshake signal error,so it is particularly important to accurately identify the modulation type of the underwater acoustic signal in the complex marine environment.At present,most of the underwater acoustic signal modulation type recognition algorithms are based on the simulated underwater environment.In order to ensure the effectiveness of the algorithm in actual communication,based on the underwater acoustic signal dataset collected by the research group in the South China Sea in 2020,the automatic modulation recognition algorithm based on deep learning is studied,and the existing network model is improved in terms of temporal feature extraction,spatial feature extraction and multi-dimensional feature fusion,and the recognition performance of the deep learning recognition model is improved.The main innovations of this thesis are as follows:(1)Aiming at the problem that the traditional network structure too simple to fit the inherent relationship between signal feature and modulation type,based on the underwater acoustic signal dataset in the South China Sea,this thesis constructs an improved network model for temporal and spatial feature extraction.First of all,based on the perspective of temporal feature,this thesis aims at the problem that in the traditional gated recurrent unit(Gated Recurrent Unit,GRU),the signal can only interact with the update gate and the reset gate,and the information utilization rate is low.The GRU is improved by introducing cross-computation before the input and hidden state enter the update gate and reset gate of the GRU,and cross-computation fuses the information of the input and hidden state,enhancing the information representation between the input and hidden state,and enhancing the information utilization.secondly,this thesis improves the traditional residual network(Residual Network,Res Net)model based on the perspective of spatial characteristics,and designs a Multi-layer residual network to achieve feature extraction at different levels;The output feature of the block are analyzed,and the global information of the signal is fully utilized to improve the expressive ability of the feature and reduce the influence of redundant feature on the signal recognition.The recognition accuracy rate of the improved GRU network model can reach 94.06%,which is 3.12% higher than that of the unimproved GRU network model.The accuracy of the improved Res Net network model can reach 93.75%,which is 5.31% higher than that of the unimproved Res Net network model.The experimental results show that improving the temporal and spatial feature can improve the model accuracy to a certain extent.(2)Aiming at the problem that it is difficult to effectively express the modulation characteristics of underwater acoustic signals by single feature extraction,this thesis proposes a G?R Fused Model based on spatiotemporal feature fusion.According to the correlation and complementarity between temporal and spatial feature,a strategy algorithm based on spatiotemporal feature fusion is designed.The algorithm fully considers the inherent defects of the single feature extraction,and enhances the effectiveness of the feature through the feature fusion strategy,which provides sufficient feature information for the network model to recognize the modulation type of underwater acoustic signals.The experimental results show that the recognition accuracy of G?R fused model reaches 98.43%,which is 7.49% higher than that of the unimproved GRU network model,and 9.99% higher than that of the unimproved Res Net network model.Compared with the unimproved network model,G?R fused model has higher recognition accuracy,and has a great improvement in recognition performance,which is suitable for occasions with high accuracy requirements.To sum up,this thesis conducts relevant research in the field of intelligent recognition of underwater acoustic modulation type,and studies feature extraction methods and efficient feature fusion strategies based on temporal and spatial feature respectively.The G?R fused model based on spatiotemporal feature fusion is proposed,and the performance of the proposed model is verified based on the South China Sea dataset.This model is more suitable for occasions where high recognition accuracy is requried.This thesis provides theoretical support for the development of marine information technology in our country,especially to improve the reliability of underwater wireless data transmission technology,and provide important support for serving the national marine strategy,marine economy,and marine security,and have certain theoretical significance and application value.