Feature Mining And Modeling Research Based On Cheanal Data

Nowadays,many countries have carried out extensive and in-depth researches on 5G mobile communication.In the key technologies for 5G,three-dimensional MIMO and massive MIMO have attracted much attention.Besides,the high-speed train scenario has became an important scenario in 5G communication.Therefore,the channel data in 5G communication shows the qualities of high volume,high diversity,high speed and high value,which indicates that the wireless communication has entered the era of Big Data.Faced with the complex and various data,traditional methods for channel analysis and modeling can not accurately analyze and model the correlation between the sub-channels and characteristics in time domain simultaneously.In this paper,methods of machine learning and data mining are applied to analyze the channel data collected in Indoor Hotspot scenario.Afterwards,the analysis results are applied to model the channel.Finally,the performances of the proposed wireless channel model are presented.The concrete contents of this paper are:(1)3D channel measurement for Indoor Hotspot scenario and data preprocessing.In order to explore new analysis methods for channel data,we proposed a 3D channel measurement in Indoor Hotspot scenario applying wideband 3D MIMO measurement platform and acquire the Channel Impulse Response(CIR)data in different geometric position in this scenario.In addition,in the preprocessing of the measurement data,neural network is applied to denoise the measured CIR data.This method can denoise the data intelligently and accurately.(2)Feature mining for amplitude and phase parts of the CIR data.Most of the existing channel analysis methods just focus on path loss,shadow fading,delay,angle of arrival,angle of departure and doppler shift.However,this paper proposes analysis methods which regard the channel impulse response just as data without any physical meaning.The CIR data are complex matrices,so we analyze the phase and amplitude of CIR matrices respectively.Besides,in order to analyze the correlation among the sub-channels as well as channel capacity,the CIR should be transformed to frequency domain and then analyzed independently as amplitude and phase part.On one hand,principal component analysis is applied to analyze the amplitude matrices,acquiring the principal components and the projections of the amplitude data on the principal component vectors.On the other hand,the sub-channels are clustered according to correlation among the sub-channels.Then,the relationships between the phase vectors of sub-channels are figured out.(3)Model the Indoor Hotspot wireless channel based on the analysis results of data mining.After acquiring the data mining results of the measured data,we apply the results of Principal Component Analysis(PCA),e.g.,the principal components and the projections of the data on the principal component vectors,to reconstruct the amplitude part.Then,the relationships between the phase vectors are used to reconstruct the phase matrices of the CIR data.Finally,the phase and amplitude parts are combined to acquire the CIR of the channel.This paper applies methods of data mining and machine learning to gain the features and structures of channel data.Based on these features and structures,a channel model which describes channel data from amplitude and phase perspectives is proposed.By comparison with the traditional model and measured data,the proposed model is more consistent with the measured result than the traditional model.Besides,this paper applied neural network to denoise channel data.Compared with the tradition denoising method,this method can recognize the noise signal more accurately.Moreover,the workload of denoising is decreased remarkably due to this approach.