Research On Robust RSSI-based Indoor Fingerprint Localization Algorithms

With development of the Internet of Things(Io T)and popularization of smart-terminals,location-based services have attracted much attention due to their large commercial and social values.The acquisition of user location information is the basis of LBS applications.The positioning technology based on satellite communication can effectively meet the user's outdoor positioning needs.However,when used indoors,the satellite positioning system is greatly limited.On the contrary,the indoor fingerprint positioning technology based on the received signal strength(RSSI)under the indoor wireless local area network(WLAN)has appeared in the eyes of researchers.It has become a research hot-spot in the field of indoor positioning.The core idea of the fingerprint-based positioning in the WALN environment is to match the RSSI signal vector collected by the online target with the RP's location fingerprint database and use the matched neighbor RPs to estimate the target.With the in-depth study of WLAN fingerprint positioning technology,we found that there are still some problems that need to resolve: Firstly,due to the complex and changeable indoor environment,which is affected by multi-path effects,shadow effects,and so on,there is spike noise in the collection of RSSI signal.The spike noise even makes users cannot detect the offline APs,degrade the indoor fingerprint localization.Secondly,there are many APs in the indoor environment,their settings,location,and other parameters are different,and different APs have different effects on the positioning.Even the status of APs will be changed in the online phase.Therefore,it is crucial to choose APs to construct an offline fingerprint database and filter APs to estimate the target in the online stage.Finally,the fingerprint positioning system needs to deploy RP in the positioning area in advance,and the map of the positioning area is the digital RP map.However,in some indoor positioning areas,the coordinates output by the fingerprint positioning system do not belong to the positioning area.This space outlier estimation has not been studied and resolved.Facing the above problems,the paper seeks to improve indoor fingerprint-based localization's accuracy and robustness.The main research work and innovations of this paper are as follows:Firstly,facing the spike noise in RSSI signal collection,we have proposed an RP torus intersection localization algorithm(TILoc)based on robust principal component analysis(RSPCA).We use RPCA to train offline fingerprint databases in the offline stage,which can reduce the peak and sparse noise in fingerprint collection and improve the offline fingerprint database's consistency;In the online phase,collect the current RSSI signals and set the undetected AP signals as 0,then construct the online RSSI vector after training by RSPCA;Use the constructed online RSSI vector to form RP torus,the neighbor RPs of the target is intersected and matched by RP torus;We estimate the target by the method of online AP weighted neighbor location.We have done many simulations and experiments,and the results show that the proposed algorithm is better than other comparison algorithms under peak noise.In the experiment of UJI,the proposed algorithm reduces the average positioning error by32% compared with the traditional algorithm.Secondly,facing offline AP selection and online APs filtering,we have proposed a fingerprint localization algorithm based on Gaussian detection(AFLoc).We design and offline AP selection method based on nonlinear quantization RSSI information entropy in the offline stage.The AP with larger RSSI information entropy is selected as the offline RSSI signal source and uses the selected APs to construct the RP fingerprint database.We use the Gaussian anomaly detection to filter out the online RSSI signals significantly different from the offline stage in the online phase.We design an AP entropy weighted similarity distance to match the neighbor RPs of the target.Finally,we use the improved AP entropy weighted nearest neighbor positioning algorithm to estimate the target.We have done A large number of simulations and experiments.The simulation and experimental results show that the proposed algorithm is better than other comparison algorithms when the AP has diversity.When two APs are abnormal,the proposed algorithm reduces the average positioning error by 42% compared with the traditional algorithm.Thirdly,we have proposed a fingerprint positioning algorithm based on approximate convex decomposition(ACDLoc),facing the problem that the fingerprint-based localization output the space-outlier coordinate.During the research on fingerprint-based localization,we have found a phenomenon that the localization system locates the target out of the positioning area,and we defined it as an unreasonable estimation.According to the mathematical analysis of fingerprint localization,the fingerprint positioning framework is equivalent to the affine combination of reference points.According to the nature of the convex function,the fundamental reason for space-outlier estimation is the non-convex geometric attribute of the RP map.To reduce space-outlier estimation,we design a feasible method.We divide the RP map into several convex sub-area and decompose the fingerprint database by the RPs in each sub-area.We use a cluster head beam to represent each sub-fingerprint database.In the online phase,locate sub-area by the similarity of online RSSI vector and cluster head beam and locate the target.To verify the algorithm,we carried out a large number of simulations.The simulation results of randomly generated diversified terrain showed that under the random noise power of20 dbm,the proposed algorithm reduced the average number of unreasonable estimates of the traditional algorithm by 75%.