Research On Wireless Sensing Technology Using Support Vector Machine And Channel State Information

With the development of the information technology,sensing technique has received much attentions.The position information and activity recognition results can be widely used in many aspects.In order to solve the drawbacks of traditional indoor localization and visual recognition techniques,in this paper,we study the wireless sensing algorithm using the channel state information(CSI)and support vector machine(SVM).The main contributions are as follows:(1)The theoretical knowledge of CSI measurements based wireless sensing are studied.Some knowledge of CSI measurement is introduced at first.Then,the machine learning techniques used in CSI based wireless sensing algorithms are described in detail.At last,the hardware platform and software platform for CSI measurement acquirement are given.(2)A wireless sensing algorithm using dimension reduction and SVM method is proposed.In the off-line phase,after the data normalization and principal component analysis(PCA)preprocessing of the CSI measurements,the problem of activity recognition can formulated as the classification problem.Through the classification learning of SVM method,the activity based classification function can be obtained.For another,for each given activity,the problem of position estimation is formulated as the regression problem.The position based regression function is given by the regression learning with SVM method.In the on-line phase,after the data preprocessing of the received CSI measurements,the activity is estimated by the activity based classification function at first.Then,the position regression function correspond to the estimated activity result is chosen for position estimation.Because the PCA method is used for the proposed algorithm,it can remove the measurement noise,reduce the time cost of off-line phase and improve the learning performance.The experiment results illustrated the efficiency of the proposed algorithm.(3)A wireless sensing algorithm using robust principal component analysis(RPCA),convolution kernel and SVM method is proposed.In the off-line phase,the RPCA method is used to remove the noise of the CSI measurements.Then each CSI measurements is transformed into a RGB image by the image rendering technique.Next,the convolution kernel is proposed to extract the feature of the RGB image.At last,through the classification learning with SVM method,the classification function is obtained for activity recognition and localization.In the on-line phase,after the RPCA process,image rendering and feature extraction of the CSI measurement data set,the activity recognition results and position estimation results can be obtained by the classification.The experiment results illustrated that the performance of the proposed algorithm is better than that of other existing methods.