Research On The Key Technology Of Automatic Recommendation Of Communication Waveform

Traditionally,waveform design for wireless communications has relied on waveform knowledge and experience held by experts.However,as the application environment and requirements become more and more complex and diverse,the accumulated waveform knowledge becomes more and more abundant,and the manual waveform design method based on the designer's knowledge base and experience is not only inefficient,but also difficult to ensure the optimization of the designed waveform.Therefore,it is necessary to improve the automation and intelligence of waveform design by structurally characterizing,storing and utilizing the existing communication waveform knowledge.Knowledge Graph(KG)technology,which has emerged in recent years,can build a structured knowledge base to characterize the relationships among entities and attributes,and has been successfully applied in many fields.Adopting knowledge graph to represent,store,and manage complex communication waveform knowledge,and making autonomous recommendation of communication waveforms based on the knowledge graph of communication waveforms,and then realizing waveform reconstruction and generation,provides a new technical direction for communication waveform design.The thesis studies the knowledge graph representation of communication waveforms and the key technology of waveform recommendation based on knowledge graph,and the main research contents and contributions are as follows.First,to address the problem of structured and unified representation of communication waveform knowledge,the thesis constructs a knowledge base model of communication waveform based on knowledge mapping and proposes a method for vector mapping of communication waveform knowledge by a representation model of joint semantic and numerical relations.Then the thesis defines the entities,relations and attributes of the communication waveform knowledge graph,and constructs the knowledge map by extracting knowledge from the existing communication waveform performance results under multiple transmission environments.Based on this,the thesis establishes a performance representation model of communication waveform knowledge,maps structured communication waveform triad information into low-dimensional vectors by Trans D model,extracts waveform semantic information and maps it into vectors by using Word2 Vec model,and constructs numerical feature vectors by customizing dictionaries and parameter taking values,and finally splices the three features to obtain the results of waveform knowledge representation learning.Taking Link-16 data chain waveform and OFDM waveform as examples,the thesis gives specific examples of communication waveform knowledge graph construction and verifies the feasibility of the adopted model and method.Next,the thesis designs the Communication Waveform Recommendation Network(CWRN)to realize the embedding enhancement and matching recommendation of communication waveform and user features.The CWRN uses a multi-layer perceptron model with three layers of fully connected neural networks to calculate the matching degree between users and waveforms to achieve waveform recommendation.Optimization is performed by minimizing the Trans D model distance function and the matching recommendation minimum mean square error(MSE)loss function to learn the user's probability rating for each waveform,and finally the network recommendation accuracy is measured by the Hit@1 value.The simulation results show that the Hit@1 of CRWN is about 0.99,which has reached a high recommendation success rate.Finally,the thesis develops a complete communication waveform knowledge graph display and recommendation simulation system using MATLAB and Python for typical application scenarios.The system has the functions of communication waveform knowledge mapping and link search display and management,and can make predictive recommendations for user input transmission environment and requirements based on the trained model,which can provide a basis for further development of more practical intelligent waveform design systems.