Channel Characteristics Analysis And Feature Mining Based On Measurement Data

With the application and development of multi-antennas,the number of antennas in communication systems is increasing.At the same time,the propagation characteristics of radio waves show complex and diverse characteristics with the change of scene,frequency band and bandwidth.Moreover,these characteristics and laws are hidden in a large amount of channel measurement data.The analysis method is difficult to obtain reliable and accurate variation rules efficiently,only through traditional data processing and segmentation.Therefore,channel characteristic analysis and feature mining based on the idea of big data and machine learning method are of great value and significance to the analysis and regularity summary of channel measured data.Based on channel measurement,this thesis studies the influence of transmission power on channel measurement results,and provides guidance value for channel measurement.Besides,this thesis plans to collect the channel data of rural Macro-cellular scenario,which is one of the important deployment scenarios of 5G.The characteristics of rural Macro-cellular scenario are summarized by using traditional data analysis methods,and the relationship between the dynamic range of channel impulse response and channel characteristics is analyzed from the perspective of correlation.Finally,based on the idea of data mining and machine learning method,this thesis extracts the effective features of channel data,analyses the potential rules of channel data,and proposes a scene similarity calculation method based on neural network.The main contents of this thesis include:1.The influence of transmission power on channel characteristicanalysis in channel measurementIn the acquisition of scene channel data,due to the limited transmission power of the measurement system and the large scope of the scene,some multipath information will be submerged in the noise,which will have a certain impact on the analysis of the multipath characteristics of the channel.In this thesis,a channel measurement scheme for indoor hot spots is designed and planned.The channel impulse response under different transmission power is acquired and processed by using 3D-MIMO channel measurement platform.By comparing the channel characteristics under different transmission power from three hands of square sum error,delay expansion and channel mutual information,it is found that only when the dynamic range of the received response is above 20 dB in the measurement process,the sufficient multipath signal can be collected and the accurate channel impulse response can be obtained.2.Channel Measurement and Channel Characteristic Analysis in Rural Macro SceneryRural Macro scenario is one of the important deployment scenarios of 5G,but there are still some deficiencies in the measurement and research of the three-dimensional spatial channel of this scene.In this thesis,the 3D-MIMO channel characteristics of 3.5GHz rural scenes are analyzed by channel measurement.For outdoor visual distance,outdoor non-visual distance and outdoor covering indoor environment,channel analysis results including large-scale,small-scale and correlation between parameters are given.In addition,due to the strong diffraction in Rural Macro scenarios,this thesis analyses the changes of channel characteristics based on the dynamic range of channel im pulse response,and finds that there is a strong relationship between them.3.Research on Feature Mining Based on Machine LearningIn order to analyze the distribution and characteristics of channel impulse response in different channel environments,this thesis firstly uses principal component analysis(PCA)method to reduce the dimension of channel impulse response delay and extract effective features.Secondly,the channel impulse response is clustered by using effective features after dimensionality reduction,and the first two important dimensions are used to visualize the channel impulse response distribution,and the channel characteristics in different channel environments are analyzed and summarized.Finally,based on the neural network model,this thesis proposes a method of calculating the similarity of channel scenarios,and uses the rural Macro-cellular scenario data for experimental testing.