Beam Management Method In Millimeter Wave Intelligent Wireless Communication

Beam management is one of the key technical schemes of millimeter wave communication,which mainly includes beam selection at the initial access of users and beam tracking in the communication process.It has a key impact on maintaining the stability of wireless communication.At present,the traditional beam management scheme still faces some severe challenges.For example,the training overhead required to obtain the millimeter wave channel state information is too large,so we can only find the best beam pair by beam scanning,which brings a certain connection delay; In addition,due to the influence of user movement or link blocking,the communication beam needs to be switched in time in the communication process,and the beam reconstruction process requires a certain time-frequency overhead,which can not meet the needs of users in high mobility or frequent blocking scenarios.Combined with the background of intelligent communication,this thesis makes effective use of the out of band information and visual information of millimeter wave system,and puts forward a series of new solutions to the problems of beam selection,link blocking and beam tracking in beam management.The main work of this thesis is as follows:Firstly,aiming at the problem of downlink beam selection in millimeter wave system,a beam selection algorithm based on sub-6GHz channel and adaptive activated antenna and an algorithm optimization scheme based on meta-learning are proposed.Due to the strong spatial correlation between different frequency signals,and the wide coverage of sub-6GHz signals and the easy acquisition of channels,it is considered to use the upstream channel state information in sub-6ghz band to assist the beam selection in millimeter wave band.In order to compensate for the loss of sub-6GHz channel information in the high resolution Angle domain,a technology scheme of adaptive selection of millimeter-wave activated antenna for information compensation is proposed,and different comparison experiments are designed.At the same time,aiming at the problem of poor generalization of data-driven deep learning model,an optimization scheme based on meta-learning is proposed based on beam selection algorithm,and the optimal initial parameter (?)is obtained to make the model adapt to the new base station environment quickly.Simulation results show that the proposed beam selection algorithm achieves excellent prediction performance under low training cost and different SNR environments,which is significantly improved compared with the baseline model.At the same time,compared to the traditional training plan and random initialization,optimization scheme based on learning algorithm to get the initial model of the migration to the new scene,only need a small amount of the training sample and training rounds can achieve ideal forecast performance,greatly improve the generalization ability of the model and learning ability,reduce the cost of migration model.Secondly,aiming at the blocking problem of downlink line-of-sight links in mm Wave systems,a multimodal blocking prediction algorithm based on visual information assistance is proposed.Current link blocking solutions are mainly based on passive responses.Inspired by multimodal learning,this thesis uses visual information to assist mm Wave systems in active prediction.The algorithm uses deep neural network to extract beam features and image features,performs multi-modal data fusion based on MFB algorithm,and uses Transformer coding to output prediction results.Simulation results show that the proposed multimodal congestion prediction algorithm achieves good prediction performance under various prediction interval lengths.At the same time,experiments compare the performance differences between Transformer coding and GRU coding.The results show that the congestion prediction algorithm based on Transformer coding has higher accuracy and precision,and is more in line with actual communication needs.Finally,aiming at the downlink beam tracking problem of millimeter wave system,a multi-modal beam tracking algorithm based on visual information assistance is proposed.Considering that the existing beam tracking schemes generally have the limitation of high training overhead and are not suitable for blocking situations,the historical beam information and image information are used as training data,and the channel information does not need to be obtained through pilot frequency,which reduces the overhead of beam training.The beams in the future are predicted through the Transformer encoding-decoding structure.The simulation results show that the next moment beam predicted by the proposed algorithm can be close to the ideal optimal transmission rate,which proves the effectiveness and superiority of the algorithm.At the same time,the algorithm predicts the beams at multiple consecutive moments in the future,which provides capability support for the millimeter wave system to carry out long-term planning of future beams.