Online Detection Of Wi-Fi Fingerprint Alteration Via Deep Learning

Due to the pervasive deployment of Wi-Fi AP(Access Point)and popular portable Wi-Fi devices promotion and use,the research topic of indoor location based on Wi-Fi fingerprint has a broad application prospect.However,some features of the fingerprint will alter with the passage of time and the change of environment,which causes a significant decline in positioning accuracy.In order to maintain the positioning accuracy,the fingerprint database must be constantly updated.The traditional fingerprint update method needs to collect the fingerprints of all areas of the room regularly and update the whole fingerprint database,which is very expensive to maintain.First of all,this paper makes an in-depth experimental study on the alteration of Wi-Fi RSS(Received Signal Strength)fingerprints,and analyzes the features of a large number of measured Wi-Fi fingerprints in different periods.Then a deep learning model ARe AE(Alteration Reducing Auto Encoder)is proposed.ARe AE regards the altered fingerprint as noise.By learning the relationship between the altered fingerprint and the original fingerprint,altered fingerprint is reduced to "pseudo-primitive fingerprint".The pseudo-primitive fingerprint and the original fingerprint have the same data distribution.On the basis of ARe AE,a fingerprint alteration detection algorithm FADet(Fingerprint Alteration Detection)is designed to detect the degree of fingerprint alteration.FADet generates pseudo-original fingerprints from the new Wi-Fi fingerprints collected by ARe AE.The Euclidean distance between the new Wi-Fi fingerprint and the corresponding pseudo-original fingerprint indicates the alteration intensity of the fingerprint.By matching the pseudo-original fingerprint with the original fingerprint database,the location of the new fingerprint can be estimated,and the fingerprint alteration intensity map ASmap(Alternation Strength map)can be constructed.With the increase of new collected fingerprints,the alteration intensity of fingerprints at various locations in the room can be displayed.According to the ASmap,some areas of fingerprint database can be updated selectively,which can effectively reduce the cost of fingerprint update.In this paper,the performance of FADet is evaluated in a real scene.The experimental scene is the corridor of teaching building and shopping center,with a total area of more than 7000 square meters.The experimental results show that FADet can accurately construct the ASmap of the experimental area.In addition,by simulating different degrees of fingerprint alteration,it is verified that FADet can detect not only a wide range of fingerprint alteration,but also small fingerprint alteration.Therefore,FADet can accurately detect the alteration intensity of indoor Wi-Fi fingerprints,which makes it possible to update part of the fingerprint database accurately and greatly reduce the maintenance cost of the fingerprint database.