Research On AI-aided MIMO-OFDM Receiver

As the mainstream research direction of future communication technology development,intelligent communication has important research value in the accelerated popularization of5 th Generation Mobile Communication(5G)technology.Intelligent communication introduces artificial intelligence(AI)technology to solve communication problems.For example,in the research of channel estimation,signal detection and channel decoding in physical layer,AI aided algorithm design can achieve or even surpass the performance of traditional algorithms,showing great research value.In the application of physical layer,intelligent communication is often combined with multiple input multiple output(MIMO)and orthogonal frequency division multiplexing(OFDM),in order to make full use of the powerful modeling and optimization ability of AI,and give full play to the outstanding advantages of MIMO-OFDM technology in spectrum efficiency and anti-interference ability.The related researches have become a key topic in the field of intelligent communication.In this paper,Artificial Intelligence Aided MIMO-OFDM receiver is studied.Firstly,this paper deeply studies the receiving algorithm of MIMO-OFDM system,focusing on channel estimation and signal detection,introduces the traditional algorithm and AI aided algorithm.In the research of classical receiving algorithm,focusing on channel estimation and signal detection,this paper introduces the linear receiving algorithm in detail,including the implementation process of four receiving algorithms in MIMO-OFDM system,and briefly introduces other receiving algorithms.When introducing machine learning algorithms and common neural network models,the research and application of AI in the field of wireless communication are respectively introduced according to different design concepts where the model driven neural network design has excellent performance in performance and network training cost.On this basis,it focuses on the research cases of AI-assisted channel estimation and signal detection,explains the specific methods of using AI to improve the performance of receiver algorithm.Through the analysis of the above research cases,the great potential of AI in receiver research is demonstrated,which provides a solid theoretical basis for AI assisted MIMO-OFDM receiver design.Secondly,three AI aided channel estimation neural networks are proposed for MIMOOFDM system with sparse orthogonal pilots,and the estimation performance,complexity and robustness of the designed neural networks are deeply explored.After introducing the common interpolation algorithms under the condition of sparse orthogonal pilot,three new channel estimation networks,SI-NET,DI-NET and SDE-NET,are proposed based on the interpolation algorithm and the principle of image superresolution.Si-net is a simple single-layer fully connected neural network based on frequency-domain interpolation algorithm.The goal of this network learning is to obtain the optimal interpolation coefficient.Di-net is based on DFT interpolation algorithm,which is composed of time domain network and frequency domain network in series.The network improves the original DFT interpolation algorithm to improve the channel estimation performance.Based on the principle of image super-resolution,SDE-net analogizes the problem of super-resolution to the problem of channel estimation in OFDM system,and adds additional denoising neural network to improve the performance of channel estimation.Simulation experiment shows that the proposed neural network performance improvement compared to traditional channel estimation methods,then we analyze the complexity of the network so as to explore the feasibility of its deployment in the practical communication systems,then we explore the robustness performance of proposed networks.Simulation results show that the performance of the proposed three new channel estimation networks is improved on the basis of the original algorithm,and SDE-NET shows the best robustness.Finally,this paper designs and implements the receiver prototype verification system for the proposed channel estimation network,including the detailed demonstration of the hardware and software design and implementation details,OTA test in multiple scenarios,and comparative analysis of the experimental results.To achieve an efficient and flexible development process,the software radio equipment used in the prototype system has rich i/o interface,highperformance RF channel and high-precision ADC,which can carry the whole process development and implementation of baseband process.Multi core server is the center of the system and it is deployed with receiver modules including channel estimation neural network,timing synchronization algorithm,signal equalization algorithm and resolution algorithm.On this basis,according to the AI offline and real-time receiver is designed and implemented,detailed introduces the system design and implementation of real-time demodulation process,including UDP transmission scheme design,the server side multi-threaded design and the realization of demodulation algorithm,by many scenes of hollow measurement performance of the proposed channel estimation of the neural network has carried on the inspection,the advance of the proposed algorithm is verified.