Research On GPS/BDS Positioning Algorithm By Mobile Phone In Urban Canyon Environment

In recent years,with the continuous improvement of the Global Navigation Satellite System（GNSS）and the increasing popularity of smartphones,smartphone based positioning services have unwittingly become an indispensable part of people’s lives.At the same time,with the continuous development of urbanization,the high-density urban buildings have brought significant challenges to the positioning of smartphones.Urban canyon is a typical scene in an urban environment,characterized by tall buildings on both sides of the street,and the width of the street is relatively narrow.However,when positioning in urban canyon,signals may be diffracted or reflected by the interference of buildings on both sides,resulting in a decline in the quality of signal observation,or even signal reception may be directly blocked,all of which can lead to a decline in positioning performance.At this stage,the aided positioning algorithm based on 3D map data has also become a new research idea by the continuous updating of urban 3D map data.What’s more,likelihood based ranging（LBR）is a typical aided positioning algorithm based on 3D map data.By aiming at the shortcomings of the current algorithm,optimization have been contributed in considering diffraction signal recognition during signal classification,enhancing initial positioning result,and optimizing feature matching processes.Then,the Random Forest Classifier and Likelihood Based Ranging（RF-LBR）algorithm has been proposed,achieving improved accuracy in GPS/BDS dual mode positioning.What’s more,the specific research content mainly includes the following aspects:（1）Researching on the propagation mechanism of diffraction signals in the urban canyon environment.Proposing a scheme for determining diffraction and reflection signals based on 3D map data by combining the occlusion of the Fresnel zone with the geometric conditions for signal diffraction and reflection.Then,based on the results of signal classification,the simulation experimental results revealed that the simulation results were relatively consistent with the observation residual results,verifying the correctness of the proposed scheme.Whereas,in the pseudo-range correction positioning algorithm that considering diffraction signals has a higher accuracy than the algorithm that does not considering diffraction signals in two sets of positioning experiments.Thus,the horizontal positioning accuracy has been improved by 17.88%and 14.53%respectively,indicating the necessity of analyzing diffraction signals for positioning.（2）Implementing a classification model considering diffraction signal recognition.Different from ignoring the type of diffraction signals in existing classification researches,proposing to incorporate diffraction signals into the GNSS signal classification by considering the necessity of diffraction signals for positioning.Owing to the different characteristic of C/N₀ of reflected,diffracted,and direct signals,the C/N₀ fluctuation feature is constructed.The random forest classification model considering diffraction signal recognition is implemented using C/N₀,C/N₀fluctuation,elevation,normalized pseudo-range residual,and pseudo-range variability consistency.Experiments were conducted in two modes:the training set and the test set in the same environment and of which was different by GPS/BDS dual-mode observation.The classification accuracy reached 82.56%and 70.18%,respectively.It was proved to be an effective algorithm.（3）Proposing the RF-LBR algorithm by optimizing the original LBR algorithm.The optimized RF-LBR algorithm increases the recognition of diffracted signals when signal classification as the LBR algorithm only divides signals into LOS and NLOS during signal classification.It avoids correcting the error distribution model for diffracted signals.In addition,the optimization of reweighting the reflected signal and reweighting the diffracted signal is adopted during the initial positioning solution.It improves the positioning accuracy and reduces the radius of the candidates search circle.Comparing to LBR algorithm,the experimental results show that,RF-LBR algorithm improves 17.99%and 29.42%in the along-street direction and 34.93%and21.54%in the cross-street direction respectively in two sets of experiments.