Adaptive Modulation And Coding Algorithm For Underwater Acoustic Communication Based On Multi-source Indicator Fusion

Ocean information technology is crucial for promoting the development of marine affairs,and has a significant impact on national maritime territory security,marine resource utilization,marine climate early warning,as well as the progress of industries such as fisheries and shipbuilding.Underwater acoustic communication technology,as a core component of ocean information technology,is of great importance.Due to the rapid changes in underwater acoustic channels,using fixed modulation and coding methods for communication may lead to wasted resources or communication failure.Adaptive modulation and coding technology can adjust modulation and coding methods in real-time according to channel changes,thereby increasing communication rates when channel conditions are good and decreasing communication rates when channel conditions are poor to ensure communication quality.Therefore,adaptive modulation and coding technology is a key link in underwater acoustic communication technology.However,the complex and ever-changing marine environment poses significant challenges to the development of adaptive modulation and encoding technology for underwater acoustic communication.These challenges specifically include: First,the complex marine environment can severely interfere with the characteristics of underwater acoustic signals,making it difficult to accurately obtain the equivalent signal-to-noise ratio(SNR)of acoustic signals through traditional mathematical calculations.Second,adaptive modulation and coding methods that rely on fixed threshold values for selecting modulation and encoding modes struggle to switch modes promptly when the SNR is at the critical point between different modulation and coding methods.Third,a single feature is insufficient to accurately measure the state of underwater acoustic channels.To address these issues,this paper proposes an underwater adaptive modulation and coding technology based on the fusion of multiple source indicators.This approach effectively extracts channel state information from underwater acoustic signals and further enhances the neural network classification accuracy and system throughput through deep canonical correlation analysis.The main contributions and innovations of this paper include:(1)To address the limitations of traditional methods,such as difficulty in accurately obtaining the equivalent SNR of underwater acoustic signals and the use of fixed threshold modulation and coding methods,this study employs neural network algorithms to learn the features of underwater acoustic signals,achieving automatic selection of Modulation and Coding Schemes(MCS).This overcomes the shortcomings of traditional fixed threshold MCS methods.(2)This paper explores the classification accuracy of different machine learning algorithms based on the extraction of marine environment features and power spectral density features.Through simulation experiments,the advantages of adaptive modulation and coding methods based on machine learning algorithms in terms of system throughput are verified.(3)To address the issue of a single feature being unable to comprehensively reflect the state of underwater acoustic channels,this study proposes a dual-branch neural network based on Deep Neural Networks(DNN)and Long Short-Term Memory(LSTM).This network extracts marine environment features and power spectral density features through the DNN branch,while simultaneously processing time-series signal features using the LSTM network to extract abstract signal features from underwater acoustic signals.By employing multiple source indicators,the network classification accuracy and system throughput are further improved.(4)To tackle the problems of poor generalization ability and nonlinearity of features in the DNN & LSTM network,this paper introduces a deep canonical correlation analysis method to perform nonlinear canonical correlation analysis on the features extracted by the dual-branch neural network.By optimizing the neural network problem,maximization of the correlation coefficient is achieved.The fusion features processed by deep canonical correlation analysis further enhance the classification accuracy of the neural network and improve its generalization ability.