CN-122002213-A - Bluetooth AOA signal outlier eliminating method and system based on deep learning

CN122002213ACN 122002213 ACN122002213 ACN 122002213ACN-122002213-A

Abstract

The invention relates to a deep learning-based Bluetooth AOA signal outlier eliminating method and a deep learning-based Bluetooth AOA signal outlier eliminating system, wherein the method comprises the steps of initializing system configuration and then cooperatively acquiring Bluetooth signals by utilizing a plurality of Bluetooth base stations; acquiring IQ data in a plurality of Bluetooth base stations, preprocessing the data, calculating to obtain a preliminary positioning result of the Bluetooth terminal according to the IQ data, constructing an input data set of a deep learning model, designing a network structure of the deep learning model, training the deep learning model according to the input data set, judging the training precision of the deep learning model, inputting the IQ data acquired in real time into the trained deep learning model to obtain a real-time positioning result of the Bluetooth terminal if the training precision meets the requirement, otherwise, returning to adjust the network structure, and optimizing the real-time positioning result by using a Kalman filtering algorithm to obtain a final positioning result. Compared with the prior art, the method has the advantages of high positioning accuracy and strong robustness.

Inventors

HU XINGYU
YIN CHENGLIANG
WU YUEPENG
GAO LIANGQUAN

Assignees

上智联(深圳)智能科技有限公司

Dates

Publication Date: 20260508
Application Date: 20251222

Claims (10)

1. A Bluetooth AOA signal outlier eliminating method based on deep learning is characterized by comprising the following steps of Initializing system configuration, then utilizing a plurality of Bluetooth base stations to cooperatively acquire Bluetooth AOA signals, acquiring IQ data in the plurality of Bluetooth base stations, preprocessing the data, calculating to obtain a preliminary positioning result of a Bluetooth terminal according to the IQ data, constructing an input data set of a deep learning model according to the preprocessed data and the preliminary positioning result, and designing a network structure of the deep learning model; judging the training precision of the deep learning model, if the training precision meets the requirement, inputting the IQ data acquired in real time into the trained deep learning model to obtain the real-time positioning result of the Bluetooth terminal, otherwise, returning to adjust the network structure; and thirdly, optimizing the real-time positioning result by using a Kalman filtering algorithm to obtain a final positioning result.
2. The deep learning-based bluetooth AOA signal outlier rejection method according to claim 1, wherein the initializing system configuration specifically includes: The method comprises the steps of configuring basic parameters of a plurality of Bluetooth base stations, including center frequency, bandwidth and modulation mode, initializing state vectors of a Kalman filter, and setting training parameters of the deep learning model, including learning rate, batch size and maximum iteration times.
3. The method for removing the wild value of the Bluetooth AOA signal based on deep learning according to claim 1, wherein the acquisition process of the Bluetooth AOA signal specifically comprises the following steps: And the plurality of Bluetooth base stations cooperatively collect the AOA signals sent by the Bluetooth terminal and set a main node Bluetooth base station and a passive node Bluetooth base station, wherein the main node Bluetooth base station and the passive node Bluetooth base station adopt different antenna array structures.
4. The method for removing the wild value of the bluetooth AOA signal based on deep learning according to claim 1, wherein the first step further includes a processing procedure of IQ data, specifically including: and performing time synchronization processing on IQ data acquired from a plurality of Bluetooth base stations by adopting a master-slave base station time synchronization mode or a GPS synchronization mode.
5. The deep learning-based bluetooth AOA signal outlier rejection method of claim 1, wherein the data preprocessing specifically includes: adopting a threshold value eliminating method based on median and standard deviation or a threshold value eliminating method based on PCA to carry out noise reduction treatment on the IQ data, and eliminating abnormal values; and calculating the phase difference of the IQ data, and eliminating the multipath effect in the IQ data by adopting a cross-correlation method or a subspace method.
6. The method for removing the wild value of the Bluetooth AOA signal based on deep learning according to claim 1, wherein the network structure for designing the deep learning model specifically comprises: setting the input layer, the hidden layer and the output layer of the network structure, and respectively designing the node number of the input layer, the node number of the hidden layer and the node number of the output layer of the network structure of the deep learning model.
7. The method for removing the wild value of the bluetooth AOA signal based on deep learning according to claim 1, wherein the process of training the deep learning model further includes: When the input data set is constructed, the input data set is divided into a training sample and a test sample, and the number of the training sample and the test sample is respectively determined; After the IQ data acquired in real time is input into the trained deep learning model, the deep learning model performs feature extraction on the IQ data to generate feature vectors and sends the feature vectors to a multi-layer perceptron, the multi-layer perceptron analyzes the feature vectors, outputs fractions representing whether the IQ data is a outlier and sends the fractions to a Softmax function for processing, and finally the Softmax function gives out probability distribution of whether the IQ data is the outlier.
8. The method for removing the wild value of the bluetooth AOA signal based on deep learning according to claim 1, wherein the determining the training accuracy of the deep learning model specifically includes: Setting a training precision threshold, returning and adjusting parameters of the network structure if the training precision of the deep learning model is smaller than or equal to the training precision threshold, and inputting the IQ data acquired in real time into the trained deep learning model if the precision is larger than the training precision threshold.
9. The method for removing the wild value of the bluetooth AOA signal based on deep learning according to claim 1, wherein the second step further comprises: After the acquired IQ data is input into the trained deep learning model, the deep learning model outputs real-time positioning coordinates of the Bluetooth terminal, including x, y and z-axis coordinates; And setting a specific filtering time constant when the real-time positioning result is optimized by using a Kalman filtering algorithm.
10. A deep learning based bluetooth AOA signal outlier rejection system, comprising a memory and a processor, the memory storing a computer program, the processor invoking the computer program to perform the steps of the method according to any of claims 1 to 9.

Description

Bluetooth AOA signal outlier eliminating method and system based on deep learning Technical Field The invention relates to the field of Bluetooth indoor positioning, in particular to a Bluetooth AOA signal outlier eliminating method and system based on deep learning. Background With the rapid development of the internet of things technology, location-based services play an increasingly important role in the fields of intelligent logistics, intelligent home, automatic driving and the like. The Bluetooth technology is widely applied in the field of indoor positioning due to the advantages of low power consumption, easy deployment, simple integration and the like. Among them, bluetooth angle arrival (AOA) technology is widely used as a common positioning method due to its characteristics of high accuracy, strong interference resistance, etc. However, the existing bluetooth AOA technology still faces many challenges in practical application, such as multipath propagation, noise and phase shift, which results in insufficient positioning accuracy and lack of robustness. In addition, the traditional positioning technology is difficult to meet the positioning requirements of high precision and strong real-time in complex and changeable indoor environments such as hospitals, markets, large office buildings and the like. To solve these problems, researchers have begun to explore new solutions. The deep learning-based method is an emerging indoor positioning technology because the deep learning-based method can directly learn features from original data and reduce manual intervention. However, the existing deep learning method still has some limitations in dealing with the problem of multi-channel voice separation, such as difficulty in fully utilizing spatial information, lack of adaptability to environmental changes, and the like. Therefore, how to apply the deep learning technology to the bluetooth AOA signal processing to improve the positioning accuracy and robustness becomes a hot spot and a difficulty of the current research. Under the background, the Bluetooth AOA signal outlier eliminating method based on embedded deep learning is generated. The method aims at removing the interference of the wild value by carrying out the advanced processing on the Bluetooth AOA signal, improving the positioning precision and providing a new solution for the indoor positioning technology. Regarding to solving the problem of the indoor positioning precision of Bluetooth, a plurality of invention patents are available at present. For example: CN115840190a discloses a high-precision positioning method based on fusion of bluetooth AOA and deep learning. According to the method, phase data of original sampling values I and Q of a Bluetooth terminal are obtained through an arrival angle positioning AOA main node Bluetooth base station and a plurality of passive node Bluetooth base stations, and the obtained IQ phase data are sent to a PC processing terminal. And the PC processing terminal inputs the acquired IQ phase data into a trained neural network model to acquire a real-time positioning result of the Bluetooth terminal. The method comprises the steps of judging whether the motion state of a Bluetooth terminal is dynamic or static according to the real-time positioning result of the Bluetooth terminal, optimizing the positioning result based on an extended Kalman filtering algorithm if the motion state is dynamic, and optimizing the positioning result based on a multipoint average filtering algorithm if the motion state is static. However, the method may be faced with problems such as signal reflection interference, antenna switching time delay, and multipath effects caused by various shields and reflectors in indoor environments in practical application. CN112040394a proposes a bluetooth positioning method based on AI deep learning. The working principle of the method is that an AOA positioning base station obtains phase data of a first signal sent by an AOA signal source, the AOA positioning base station sends the obtained phase data to an AI server, and the AI server determines the position of the AOA signal source for transmitting the first signal according to a trained neural network model and the obtained phase data. The method further abstracts the phase difference by introducing an AI technology into the AOA positioning field and sampling the original phase value of a positioning signal source received by multiple antennas, then converts the phase difference into angles, and trains the related phase data to the space coordinates of a known network format by the angle combination of different antennas. However, this method still faces the problem of how to further optimize the neural network model to improve the accuracy and speed of positioning in practical applications. The prior art has the following disadvantages: 1. The traditional Bluetooth AOA positioning technology is easily affected by signal reflection interfe