Research And Application On Ranging Model Based On Multi-channel BLE RSSI

With the rapid development of wireless communication technology and the popularity of mobile Internet,Location Based Service(LBS)has been widely used in various fields,such as navigation,advertising,social networking,and commerce.Among them,new technologies and methods have emerged in the field of ranging technology that supports LBS.Ranging technologies using wireless signals such as Bluetooth,ultrasound,and infrared include empirical ranging models as well as ranging models based on machine learning and ranging models based on deep learning.The accuracy of the models is largely influenced by the input features and algorithms,so the design of optimal input features for the ranging models needs to take into account the representativeness,adequacy,interpretability,and noise immunity of the features.In the existing ranging methods based on Bluetooth Received Signal Strength Indication(RSSI),the RSSI value which is used as the feature has the problem of not distinguishing the channel.The lack of representativeness and adequacy of the feature value of the composite channel affects the accuracy of the model.To address this problem,this thesis proposes a multi-channel signal characterization method to clarify the degree of channel separation and correlation,solving the problem of signal frequency dependence.More accurate understanding and processing of input data can reduce model errors.Ranging in large indoor spaces is a challenging problem in the field of range.Due to the complexity of space with the presence of a large number of obstacles and signal fading,it is difficult to obtain high accuracy results with traditional ranging methods,so it is also important to improve the representativeness and adequacy of features to optimize the ranging model.In addition,this thesis proposes a gridding ranging model based on transfer learning,which extracts features and adjusts model parameters to the referenceable environmental information in space adaptively through transfer learning to adapt to the ranging task in large indoor spaces better,thus,improving the ranging accuracy.The main contributions of this thesis are as follows:(1)The thesis proposes a multi-channel Bluetooth Low Energy(BLE)signal feature method,optimizing the empirical ranging model,ranging model based on machine learning,and ranging model based on deep learning respectively by multi-channel signal features.Improving the accuracy and generalization of the model by optimizing the feature representativeness and adequacy of the ranging model,making it more reliable and stable in practical applications.(2)The thesis proposes a gridding ranging model based on multi-channel RSSI,which defines three types of grids and combines with transfer learning to optimize the feature representativeness of the model input data based on the three types of grids.Transferring the feature extraction and model structure information of the pre-trained models in the three types of grids.The gridding ranging model achieves customized design for specific spatial scenarios and combines with transfer learning to realize model reuse organically.The method solves the problems of limited data volume,mismatched sample distribution,and complex environment of large-scale indoor space effectively,thus,improving the accuracy and efficiency of the ranging model.(3)The thesis constructs a BLE multi-channel RSSI dataset and conducts experiments on this dataset to verify the effectiveness of the proposed method.A prototype system of ranging tracing is designed and implemented,and the ranging model is practically applied to hospital personnel contact tracing.