Reseach On RF Fingerprinting Localization Based On Machine Learning

With the rapid development of mobile Internet, Internet of Things and smart mobile devices, the demand for location-based services is growing. In addition, because of the randomness of crowd and complexity of positioning environment, how to achieve localization in complex environments accurately and effectively has great significance. Though the RF fingerprinting technology receives more and more attention because of its superior positioning performance, it is not mature at present and there are still many problems and challenges.Based on current research, the main problems in existing RF fingerprinting location methods are analyzed and a comprehensive study of RF fingerprinting localization methods based on machine learning is made. This study aims to solve the existing problems more or less. This paper mainly includes the following contents:Firstly, the statistical characteristics and influencing factors to RSS fingerprints are analyzed based on experiments. Some key factors of the RF fingerprinting positioning are deep studied and the relationships between these key factors and positioning performance are mined, which provide a good theoretical support for the later optimization and implementation of algorithms.Secondly, an analytical fingerprinting method is proposed in this paper to solve the problem of building the fingerprints database manually. This method can build the database automatically which reduces the workload effectively. The simulation results prove that the proposed method has better performance and robustness than other methods in the environment with low SNR. Furthermore, based on the analyse of the spatial variability of the algorithm, some machine learning methods, include genetic algorithm and BP neural network, are applied in the analytical fingerprinting positioning method. This method reduces the influence of spatial variability and improves the positioning accuracy and performance effectively.Finally, a RSS fingerprinting localization method based on nonlinear support vector machine is proposed in this paper. This method introduces the concept of meachine learning into the RSS fingerprinting localization process and realized positioning by the learning of non-linear relationship between the signal space and the physical space. To analyze the algorithm parameters and verify the fingerprinting positioning method based on machine learning, an experimental platform in actual indoor environment is built. The whole process of the fingerprinting positioning is realized in this platform and the experiment results show that the proposed method can reduce the size of the database, amount of calculation and workload of building database in keeping unchanged positioning accuracy effectively which has important significance in the applications of indoor positioning.