Research On Multipath Classification Method Of CSI System Based On Improved Random Forest Algorithm

With the rapid development of mobile Internet technology,people have an urgent need for accurate location information.Wi Fi indoor positioning technology,which is economical,convenient,fast,and easy to deploy,stands out among many indoor positioning technologies and becomes a research hotspot.Positioning accuracy is a fundamental index for evaluating positioning performance.The complexity and variability of indoor environments make positioning accuracy often subject to the factors such as multiple-access interference,multipath propagation,and non-line of sight(NLOS)propagation.Whether the multi-path classification can be completed based on the superimposed signal at the receiver has become one of the core issues of indoor Wi Fi positioning.In the indoor positioning system based on Wi Fi,the traditional multipath classification method usually uses the Received Signal Strength Indicator(RSSI)as the observational measurement.However,RSSI is prone to cause violent fluctuation in multipath propagation environment,affecting classification performance.In addition,the existing multipath classification methods usually need to set a feature discrimination threshold,but it is difficult to obtain a highly accurate and widely applicable threshold.Motivated by this fact,this paper selects Channel State Information(CSI)as the observation measurement to study the multipath classification problem.The key points include:1.In the existing classification methods,the distribution characteristics of typical statistical characteristics(such as mean,standard deviation,etc.)are usually used for classification,but their accuracy can be further improved.For this problem,this paper proposes a KLOS factor based on Kmeans clustering algorithm by analyzing the propagation characteristics of wireless signals in line-of-sight(LOS)and non-line-of-sight environments.Simulation and actual measurement prove that the proposed features are beneficial to improve classification performance.2.Aiming at the problem that the existing multipath classification methods are usually restricted by the feature discrimination threshold,this paper proposes a multipath classification method based on an improved random forest algorithm.(1)This paper establishes a random forest classification architecture based on the C4.5 algorithm,calculates the information gain rate of different statistical features,selects the statistical features with outstanding information gain rate as the root node to construct multiple decision trees.Form the multiple trees into a random forest and use a simple majority voting mechanism to obtain classification results.Simulation and actual measurement prove that the proposed method is beneficial to improve classification performance.(2)In view of the different classification performance of different feature combinations,this paper proposes a feature selection method based on Fisher criterion for the purpose of obtaining the optimal feature combination.Feature calculation is performed by calculating the inter-class scattering and intra-class scattering of the different classes,so that the inter-class scattering of each feature is as large as possible and the intra-class scattering is as small as possible,which effectively reduces the calculation amount and running time of the algorithm,and can quickly find the feature combinations with higher classification performance.The proposed Fisher Criterion is verified by simulation and actual measurement.