Wireless positioning method, path identification model training method and device

Abstract

The application provides a wireless positioning method, a training method and device of a path identification model, and relates to the field of artificial intelligence. The wireless positioning method comprises the following steps: acquiring a line-of-sight path probability of a signal transmission path between a to-be-positioned device and an access device; acquiring a target line-of-sight path probability greater than or equal to a preset threshold from all line-of-sight path probabilities; when the number of target line-of-sight path probabilities is greater than or equal to three, selecting three access devices from the access devices corresponding to the target line-of-sight path probabilities; acquiring time of arrivals of the to-be-positioned device to the three access devices; and determining the position of the to-be-positioned device according to the time of arrivals of the to-be-positioned device to the three access devices. The wireless positioning method can improve positioning accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of wireless communication, in particular to a wireless positioning method, a path identification model training method and device. BACKGROUND

[0002] There are positioning technologies based on wireless signal measurement, such as positioning technologies based on wireless signal strength, time of arrival, signal arrival angle; there are positioning technologies based on various sensor information, such as positioning technologies based on inertial sensor (accelerometer, gyroscope, geomagnetic meter), geomagnetic, light, camera, sound sensor.

[0003] At present, there is a wireless positioning method as follows: a terminal sends a positioning measurement signal to multiple wireless access points, each wireless access point obtains a time of arrival (TOA) according to the positioning measurement signal, and then calculates the terminal position according to at least three TOAs and the positions of at least three wireless access points.

[0004] In actual application, there may be obstacles between the terminal and one or more wireless access points, so that the measured TOA has a certain error, resulting in low positioning accuracy. SUMMARY

[0005] Therefore, the present application provides a wireless positioning method, a path identification model training method and device, which can improve the positioning accuracy.

[0006] The first aspect provides a wireless positioning method, which comprises: obtaining the line of sight path probability of the signal transmission path between a device to be positioned and an access device, and then obtaining target line of sight path probabilities greater than or equal to a preset threshold from all line of sight path probabilities; when the number of target line of sight path probabilities is greater than or equal to three, selecting three access devices from the access devices corresponding to the target line of sight path probabilities, and then obtaining the time of arrival of the device to be positioned to the three access devices, and determining the position of the device to be positioned according to the time of arrival of the device to be positioned to the three access devices.

[0007] The line of sight path probability refers to the probability that the signal transmission path between the device to be positioned and the access device is a line of sight path. The preset threshold is used to measure whether the line of sight path probability meets the line of sight path requirement, and the specific value can be set according to the actual situation. The time of arrival refers to the transmission time of the signal from the terminal to the access device, or the transmission time of the signal from the access device to the terminal. The time of arrival is also called signal flight time.

[0008] According to the implementation, whether the wireless signal transmission path between the device to be positioned and the access device is a line-of-sight path can be determined according to the line-of-sight path probability. When the signal transmission paths between the device to be positioned and three or more access devices are line-of-sight paths, the device to be positioned is positioned according to the arrival times corresponding to the three line-of-sight paths. In this way, errors caused by non-line-of-sight paths can be reduced, and therefore the positioning accuracy can be improved.

[0009] In a possible implementation, obtaining the line-of-sight path probability of the signal transmission path between the device to be positioned and the access device includes receiving the line-of-sight path probability sent by the access device. In this implementation, the path recognition model is deployed on the access device. After the access device inputs the power-delay profile into the path recognition model to extract features and obtains the line-of-sight path probability, the access device sends the line-of-sight path probability to the wireless positioning apparatus. In this way, a method for obtaining the line-of-sight path probability is provided, and the amount of data of the line-of-sight path probability is small, facilitating transmission, and therefore the wireless positioning apparatus can position a large number of users. The power-delay profile is determined by the access device based on the positioning measurement signal sent by the device to be positioned, and the positioning measurement signal can be but is not limited to a channel sounding reference signal.

[0010] In another possible implementation, obtaining the line-of-sight path probability of the signal transmission path between the device to be positioned and the access device includes: after the access device determines the power-delay profile based on the positioning measurement signal sent by the device to be positioned to the access device, the access device sends the power-delay profile to the wireless positioning apparatus; and after the wireless positioning apparatus receives the power-delay profile sent by the access device, the wireless positioning apparatus inputs the power-delay profile into the trained path recognition model to extract features, to obtain the line-of-sight path probability of the signal transmission path between the device to be positioned and the access device. The positioning measurement signal can be but is not limited to a channel sounding reference signal. In this implementation, the path recognition model is deployed on the wireless positioning apparatus. In this way, another method for obtaining the line-of-sight path probability is provided, and the implementation of the scheme is more flexible.

[0011] In another possible implementation, inputting the power-delay profile into the trained path recognition model to extract features to obtain the line-of-sight path probability of the signal transmission path between the device to be positioned and the access device includes: standardizing the power-delay profile; performing nonlinear transformation on the standardized power-delay profile by using a preset weight matrix and a preset bias vector; and performing product operation on a preset weight vector and a vector obtained through the nonlinear transformation, to obtain the line-of-sight path probability of the signal transmission path between the device to be positioned and the access device. In this way, a specific method for extracting features by using the path recognition model is provided.

[0012] In another possible implementation, determining the position of the device to be positioned according to the time of arrival of the device to be positioned to the three access devices includes: determining the time of arrival difference between the access devices according to the time of arrival of the device to be positioned to the three access devices; and determining the position of the device to be positioned according to the time of arrival difference. In this way, a method for calculating the position of the device to be positioned according to the time of arrival difference is provided, which can reduce the influence of the time difference between the device to be positioned and the access devices on the positioning error, and has good positioning accuracy.

[0013] In another possible implementation, the wireless positioning method further includes: obtaining a reference signal received power of the device to be positioned; and when the number of target line-of-sight path probabilities is less than three, determining the position of the device to be positioned as a position corresponding to the reference signal received power and the time of arrival in a preset fingerprint database. The reference signal received power is determined according to a reference signal sent by the device to be positioned. The preset fingerprint database stores one or more reference signal received powers of the device to be positioned and one or more times of arrival. When the number of target line-of-sight path probabilities is less than three, fingerprint positioning is performed according to the reference signal received power and the time of arrival, thus providing a fingerprint positioning method. Compared with a single channel feature, fingerprint positioning using two channel features has higher accuracy. It should be understood that the present application can also use other types of channel features for fingerprint positioning, or use more channel features for fingerprint positioning. The channel feature can be, but is not limited to, a power delay profile or a channel matrix.

[0014] The second aspect provides a training method of a path identification model, which includes: obtaining a first sample set and a path identification vector; and training the path identification model with the first sample set as input and a model classification error less than or equal to a preset error as a target, to obtain a trained path identification model. Each sample in the first sample set includes a plurality of power delay profile data, and the path identification vector includes a plurality of path identifications, which are in one-to-one correspondence with the samples. Each sample can include the same number of power delay profile data. The model classification error is used to indicate the difference between a model classification result of the path identification model and the path identification vector, and the model classification result is obtained by classifying the line-of-sight path probability output by the path identification model. When the model classification error is less than or equal to the preset error, it indicates that the path identification model can meet the demand, and the weight of the trained path identification model is saved for subsequent use. In this way, the path identification model can be obtained, which can obtain the line-of-sight path probability corresponding to the power delay profile to determine whether the signal transmission path is a line-of-sight path.

[0015] In a possible implementation, the algorithm for training the path identification model is a proximity algorithm, a support vector machine algorithm, a gradient boosting decision tree algorithm, a linear discriminant analysis algorithm, or a random nonlinear discriminant analysis algorithm.

[0016] In another possible implementation, training the path recognition model with the first sample set as input and with a model classification error less than or equal to a preset error as a target includes: step A, standardizing the first sample set; step B, performing nonlinear conversion on the standardized second sample set according to a candidate weight matrix and a candidate bias vector to obtain a data matrix; step C, obtaining an inter-class distance matrix and an intra-class distance matrix of the data matrix; step D, determining a weight vector according to the inter-class distance matrix and the intra-class distance matrix; step E, determining a line-of-sight path probability as a product of the data matrix and the weight vector; step F, determining a classification threshold according to the inter-class distance matrix, the intra-class distance matrix, and the weight vector; step G, determining the model classification error according to the line-of-sight path probability and the classification threshold; step H, when the model classification error is greater than the preset error, updating the candidate weight matrix and the candidate bias vector, and triggering steps B to H; and step I, when the model classification error is less than or equal to the preset error, obtaining the path recognition model corresponding to the candidate weight matrix and the candidate bias vector. This method can train the weight matrix and the bias vector of the path recognition model, thus providing a specific method for training the path recognition model. It can be seen from the model classification error less than or equal to the preset error that this method can train an effective path recognition model according to actual path recognition requirements.

[0017] In a possible implementation, standardizing the first sample set includes: forming a sample center point vector by using average values of columns in the first sample set; forming a sample standard deviation vector by using standard deviations of the columns in the first sample set; and performing operations on the first sample set, the sample center point vector, and the sample standard deviation vector according to a preset standardization formula. This provides a method for standardizing the first sample set. The standardization formula can be, but is not limited to, a z-score standardization formula.

[0018] In another possible implementation, obtaining the first sample set includes: obtaining a third sample set; obtaining invalid samples of the third sample set; and removing the invalid samples in the third sample set to obtain the first sample set. The third sample set can be a raw sample set collected manually or by a machine.

[0019] Optionally, the invalid sample of the third sample set is obtained by: selecting a to-be-processed power-time delay spectrum corresponding to the line-of-sight path identifier from the third sample set; dividing the to-be-processed power-time delay spectrum into multiple intervals; selecting target power-time delay spectrum data from each interval; determining a set of bending angles of the to-be-processed power-time delay spectrum according to the target power-time delay spectrum data of each interval, wherein each bending angle in the set of bending angles is determined according to the target power-time delay spectrum data of three continuous intervals; determining a target angle greater than a preset angle and less than 180 degrees in the set of bending angles; and determining that the to-be-processed power-time delay spectrum is an invalid sample when the number of target angles is greater than or equal to a preset number. When the number of target angles is greater than or equal to the preset number, it indicates that the number of wave crests in the to-be-processed power-time delay spectrum is greater than the preset number, so that the path identifier corresponding to the to-be-processed power-time delay spectrum is probably not the line-of-sight path identifier, and the path identifier of the to-be-processed power-time delay spectrum in the third sample set is the line-of-sight path identifier, so that the path identifier of the to-be-processed power-time delay spectrum in the third sample set is probably marked incorrectly, and thus the to-be-processed power-time delay spectrum is classified as an invalid sample. In this way, the path identification model error caused by the invalid sample can be reduced, and the accuracy of the path identification model can be improved.

[0020] The third aspect provides a wireless positioning device, which comprises a first obtaining module, a second obtaining module, a selecting module, a third obtaining module and a position solving module. The first obtaining module is configured to obtain a line-of-sight path probability of a signal transmission path between a to-be-positioned device and an access device. The second obtaining module is configured to obtain a target line-of-sight path probability greater than or equal to a preset threshold from all line-of-sight path probabilities. The selecting module is configured to select three access devices from access devices corresponding to the target line-of-sight path probability when the number of target line-of-sight path probabilities is greater than or equal to three. The third obtaining module is configured to obtain time of arrivals of the to-be-positioned device to the three access devices. The position solving module is configured to determine a position of the to-be-positioned device according to the time of arrivals of the to-be-positioned device to the three access devices.

[0021] In a possible implementation, the first obtaining module is specifically configured to receive the line-of-sight path probability sent by the access device, and the line-of-sight path probability is obtained by inputting a power-time delay spectrum into a path identification model for feature extraction.

[0022] In another possible implementation, the first obtaining module comprises a receiving unit and a feature extraction unit. The receiving unit is configured to receive a power-time delay spectrum sent by the access device, and the power-time delay spectrum is determined according to a positioning measurement signal sent by the to-be-positioned device to the access device. The feature extraction unit is configured to input the power-time delay spectrum into a trained path identification model for feature extraction, to obtain the line-of-sight path probability of the signal transmission path between the to-be-positioned device and the access device.

[0023] In a possible implementation, the feature extraction unit is specifically configured to normalize the power delay spectrum; perform nonlinear transformation on the normalized power delay spectrum using a preset weight matrix and a preset bias vector; and perform product operation on the preset weight vector and the vector obtained through the nonlinear transformation to obtain the line-of-sight path probability of the signal transmission path between the to-be-positioned device and the access device.

[0024] In a possible implementation, the position calculation module is specifically configured to determine the time-of-arrival difference between the access devices according to the time of arrival of the to-be-positioned device to the three access devices; and determine the position of the to-be-positioned device according to the time-of-arrival difference.

[0025] In a possible implementation, the wireless positioning apparatus further includes a fourth acquisition module and a position calculation module. The fourth acquisition module is configured to acquire the reference signal received power of the to-be-positioned device, the reference signal received power being determined according to the reference signal sent by the to-be-positioned device. The position calculation module is further configured to, when the number of target line-of-sight path probabilities is less than three, determine the position of the to-be-positioned device as the position corresponding to the reference signal received power and the time of arrival in the preset fingerprint database.

[0026] The steps or beneficial effects performed by the modules or units in the wireless positioning apparatus of the third aspect can be referred to the corresponding description in the first aspect.

[0027] The fourth aspect provides a training apparatus of a path identification model. The training apparatus includes an acquisition module and a training module. The acquisition module is configured to acquire a first sample set and acquire a path identification vector. Each sample in the first sample set includes a plurality of power delay spectrum data. The path identification vector includes path identifications corresponding to the samples one by one. The training module is configured to take the first sample set as input, train the path identification model with a model classification error less than or equal to a preset error as a target, and obtain a trained path identification model. The model classification error is used to indicate the difference between the model classification result of the path identification model and the path identification vector. The model classification result is obtained by classifying the line-of-sight path probability output by the path identification model.

[0028] In a possible implementation, the training module is specifically configured to perform the following steps: Step A: normalizing the first sample set; Step B: performing nonlinear transformation on the normalized second sample set according to the candidate weight matrix and the candidate bias vector to obtain a data matrix; Step C: obtaining an inter-class distance matrix and an intra-class distance matrix of the data matrix; Step D: determining the weight vector according to the inter-class distance matrix and the intra-class distance matrix; Step E: determining the line-of-sight path probability as the product of the distance matrix and the weight vector; Step F: determining the classification threshold according to the inter-class distance matrix, the intra-class distance matrix, and the weight vector; Step G: determining the model classification error according to the line-of-sight path probability and the classification threshold; Step H: when the model classification error is greater than a preset error, updating the candidate weight matrix and the candidate bias vector, triggering Steps B to H, until the model classification error is less than or equal to the preset error; and Step I: when the model classification error is less than or equal to the preset error, obtaining the path recognition model corresponding to the candidate weight matrix and the candidate bias vector.

[0029] In another possible implementation, the training module is specifically configured to: compose a sample center point vector by using the average values of the columns in the first sample set; compose a sample standard deviation vector by using the standard deviations of the columns in the first sample set; and perform operations on the first sample set, the sample center point vector, and the sample standard deviation vector according to a preset normalization formula.

[0030] In another possible implementation, the obtaining module includes a first obtaining unit, a second obtaining unit, and a removing unit. The first obtaining unit is configured to obtain the third sample set. The second obtaining unit is configured to obtain invalid samples of the third sample set. The removing unit is configured to remove the invalid samples in the third sample set to obtain the first sample set.

[0031] In a possible implementation, the second obtaining unit is specifically configured to: select a to-be-processed power-time delay spectrum corresponding to the line-of-sight path identifier from the third sample set; divide the to-be-processed power-time delay spectrum into multiple intervals; select target power-time delay spectrum data from each interval; determine a set of bending angles of the to-be-processed power-time delay spectrum according to the target power-time delay spectrum data of the intervals, wherein each bending angle in the set of bending angles is determined according to the target power-time delay spectrum data of three consecutive intervals; determine a target angle that is greater than a preset angle and less than 180 degrees in the set of bending angles; and when the number of target angles is greater than or equal to a preset number, determine that the to-be-processed power-time delay spectrum is an invalid sample.

[0032] The steps or beneficial effects performed by the modules or units in the training device of the path recognition model of the fourth aspect can be referred to the corresponding descriptions in the second aspect.

[0033] The fifth aspect provides a wireless positioning device, comprising a processor and a memory, the memory being used for storing program instructions; the processor is used for implementing the method in any one of the implementation manners of the first aspect by executing the program instructions.

[0034] The sixth aspect provides a path identification model training device, comprising a processor and a memory, the memory being used for storing programs; the processor is used for implementing the method in any one of the implementation manners of the second aspect by executing the programs.

[0035] The seventh aspect provides a computer readable storage medium, the computer readable storage medium storing instructions, when the instructions are executed on a computer, the computer executes the method in any one of the implementation manners of the first aspect or the second aspect.

[0036] The eighth aspect provides a computer program product comprising instructions, when the instructions are executed on a computer, the computer executes the method in any one of the implementation manners of the first aspect or the second aspect.

[0037] The ninth aspect provides a chip, the chip comprising a processor and a data interface, the processor reading instructions stored on a memory through the data interface, and executing the method in any one of the implementation manners of the first aspect or the second aspect.

[0038] Optionally, as an implementation manner, the chip can further comprise a memory, the memory storing instructions, and the processor is used for executing the instructions stored on the memory, when the instructions are executed, the processor is used for executing the method in any one of the implementation manners of the first aspect or the second aspect. The chip can be a field-programmable gate array (FPGA) or an application-specific integrated circuit (ASIC) in particular.

[0039] The tenth aspect provides an electronic device, the electronic device comprising the device in any one of the implementation manners of the third aspect or the fourth aspect. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 An illustration of a communication system of the present application;

[0041] Figure 2 An illustration of the line-of-sight path ratio and the non-line-of-sight path ratio in multiple scenarios;

[0042] Figure 3A An illustration of an existing wireless positioning method;

[0043] Figure 3BAnother schematic diagram of the wireless positioning method in the embodiment of the present application;

[0044] Figure 4 A schematic diagram of the wireless positioning method in the embodiment of the present application;

[0045] Figure 5 A signaling interaction diagram of the wireless positioning method in the embodiment of the present application;

[0046] Figure 6 Another signaling interaction diagram of the wireless positioning method in the embodiment of the present application;

[0047] Figure 7 A schematic diagram of the system architecture for training the path recognition model in the embodiment of the present application;

[0048] Figure 8 Another schematic diagram of the system architecture for training the path recognition model in the embodiment of the present application;

[0049] Figure 9 A flowchart of the training method of the path recognition model in the embodiment of the present application;

[0050] Figure 10 A schematic diagram of the feature extraction based on the path recognition model in the embodiment of the present application;

[0051] Figure 11A A schematic diagram of the power-time delay spectrum of the line-of-sight path in the embodiment of the present application;

[0052] Figure 11B A schematic diagram of the power-time delay spectrum of the non-line-of-sight path in the embodiment of the present application;

[0053] Figure 12 A structural schematic diagram of the wireless positioning device in the embodiment of the present application;

[0054] Figure 13 A structural schematic diagram of the training device of the path recognition model in the embodiment of the present application;

[0055] Figure 14 A structural schematic diagram of the mobile edge computing server in the embodiment of the present application;

[0056] Figure 15 A structural schematic diagram of the base station in the embodiment of the present application. DETAILED DESCRIPTION

[0057] The wireless positioning method of the present application can be applied to shopping mall guide, underground parking lot guide, warehouse logistics, intelligent factory, and scenarios with non line of sight (NLOS) paths. NLOS refers to the blocking of signals between the transmitting and receiving parties during wireless signal propagation, so the signal cannot reach the receiving party in a straight line. Line of sight (LOS) refers to the absence of obstructions between the transmitting and receiving parties during wireless signal propagation, so the signal can reach the receiving party in a straight line.

[0058] The communication system deployed in the above scenarios can be a 2G communication system, a 3G communication system, an LTE communication system, a 5G communication system, or a communication system after 5G. The 2G communication system can be, but is not limited to, the global system for mobile communications (GSM). The 3G communication system can be, but is not limited to, CDMA2000 and WCDMA. The communication system can also be a communication system combining wired communication and wireless communication, such as WIFI, ultrawide band (UWB), etc. The wireless positioning method of the present application can realize the positioning function in the above communication systems.

[0059] Figure 1 is a schematic diagram of the communication system of the present application. Referring to Figure 1 , the communication system includes a terminal 11, an access device 121, an access device 122, an access device 123, an access device 124, and a position calculation device 13. The terminal 11 is connected to the access device 121, the access device 122, the access device 123, and the access device 124 through a wireless link, and the position calculation device 13 is connected to the access device 121, the access device 122, the access device 123, and the access device 124 through a wired or wireless connection.

[0060] The terminal 11 can be a mobile phone, a tablet computer, a notebook computer, a smart wearable device, a vehicle-mounted computer, a virtual reality device, an augmented reality device, an Internet of Things device, etc. The terminal 11 is also called a user equipment, a user terminal, a wireless terminal, etc.

[0061] The access device can be a base station, an indoor base station, a micro base station, a pico base station, a wireless access point, etc.

[0062] The position calculation device 13 is used to calculate the terminal position according to the position related parameters, store the terminal position, etc. The position calculation device 13 can be a mobile edge computing (MEC) server. The terminal position refers to the geographic coordinates of the terminal.

[0063] It should be understood that the number of terminals, the number of access devices or the number of position calculation devices in the communication system of the present application is not limited to the above examples.

[0064] The line-of-sight path and the non-line-of-sight path in some scenarios are described as follows:

[0065] Referring to Figure 2 In an office scenario or a residential scenario, the proportion of scenarios with one line-of-sight path is less than 20%, the proportion of scenarios with two line-of-sight paths is less than 5%, the proportion of scenarios with three line-of-sight paths is less than 2%, and the proportion of mixed scenarios including line-of-sight paths and non-line-of-sight paths is more than 98%.

[0066] In a dense urban scenario, the proportion of scenarios with one line-of-sight path is less than 30%, the proportion of scenarios with two line-of-sight paths is less than 20%, the proportion of scenarios with three line-of-sight paths is less than 20%, and the proportion of mixed scenarios including line-of-sight paths and non-line-of-sight paths is more than 80%.

[0067] In a factory, airport or station scenario, the proportion of scenarios with one line-of-sight path is less than 30%, the proportion of scenarios with two line-of-sight paths is less than 20%, the proportion of scenarios with three line-of-sight paths is less than 20%, and the proportion of mixed scenarios including line-of-sight paths and non-line-of-sight paths is more than 80%.

[0068] In a lecture hall or conference hall scenario, the proportion of scenarios with one line-of-sight path is less than 20%, the proportion of scenarios with two line-of-sight paths is less than 20%, the proportion of scenarios with three line-of-sight paths is more than 60%, and the proportion of mixed scenarios including line-of-sight paths and non-line-of-sight paths is less than 40%.

[0069] It can be seen that in many scenarios, there are non-line-of-sight paths. In the scenario of non-line-of-sight path, there is refraction or reflection phenomenon between the terminal and the access device, so there is an error in the arrival time or other channel characteristics of the non-line-of-sight path. The positioning error in the non-line-of-sight scenario combined with the communication scenario is described as follows:

[0070] Referring to Figure 3AIn one example, the access device is a wireless access point, and there is an obstacle, such as a wall, a door, furniture, etc., between the terminal 11 and the wireless access point 121. The terminal 11 sends a positioning measurement signal to the wireless access point 121, the wireless access point 122, the wireless access point 123, and the wireless access point 124. Each wireless access point can determine the time of arrival of the signal from the terminal 11 to the wireless access point according to the time of reception and the time of transmission of the positioning measurement signal. After the wireless access point 121 transmits the first time of arrival, the wireless access point 122 transmits the second time of arrival, the wireless access point 123 transmits the third time of arrival, and the wireless access point 124 transmits the fourth time of arrival, the position resolving device 13 can perform geometric positioning according to three times of arrival and the preset positions of the wireless access points. If the position resolving device 13 uses the first time of arrival, the second time of arrival, and the third time of arrival, there is a large error in the first time of arrival due to the interference of the obstacle, and the position of the terminal 11 calculated by the position resolving device 13 has a large error.

[0071] Referring to Figure 3B In another example, the access device is a wireless access point, and there is an obstacle between the terminal 11 and the wireless access point 121. The terminal 11 sends a positioning measurement signal to the wireless access point 121, the wireless access point 122, the wireless access point 123, and the wireless access point 124, respectively. The wireless access point 121 obtains the first time of arrival and the first reference signal received power (RSRP) according to the positioning measurement signal, the wireless access point 122 obtains the second time of arrival and the second reference signal received power according to the positioning measurement signal, the wireless access point 123 obtains the third time of arrival and the third reference signal received power according to the positioning measurement signal, and the wireless access point 124 obtains the fourth time of arrival and the fourth reference signal received power according to the positioning measurement signal. Then, each access point sends the RSRP and the TOA to the position resolving device 13. The position resolving device 13 constructs a fingerprint of the terminal 11 according to the RSRP and the TOA sent by each wireless access point, and finds the position corresponding to the fingerprint in the fingerprint database. If there is no position corresponding to the fingerprint in the fingerprint database, the positioning fails. After obtaining the position of the terminal 11, the position of the terminal 11 can be sent to the client device 104.

[0072] The first time of arrival and the first reference signal received power stored in the fingerprint database can be measured in a line of sight path. When positioning, an obstacle is added between the terminal 11 and the wireless access point 121, i.e. the signal transmission path between the terminal 11 and the wireless access point 121 becomes a non-line of sight path. The first time of arrival and the first reference signal received power measured in this way can be different from the first time of arrival and the first reference signal received power in the fingerprint database, which can easily lead to fingerprint positioning failure. Since the actual environment often changes, the robustness of fingerprint positioning is poor, and the accuracy of fingerprint positioning is lower than that of geometric positioning. It should be understood that the obstacle between the terminal and the access device is not limited to the above examples, and the channel characteristics are not limited to TOA and / or RSRP.

[0073] As can be seen from the above, when determining the terminal position according to the TOA and / or RSRP of the non-line of sight path, the positioning accuracy and positioning accuracy can not meet the actual needs. For the above problems, the wireless positioning method of the present application can reduce the channel characteristics of the non-line of sight path, so as to reduce the positioning error caused by the channel characteristics of the non-line of sight path, and thus improve the positioning accuracy and positioning accuracy.

[0074] The wireless positioning method of the present application will be introduced below, please refer to Figure 4 In the wireless positioning method of the present application, the terminal or the access device can obtain the channel information between the terminal and the access device, for example, the access device receives the SRS sent by the terminal, or the terminal receives the downlink measurement signal sent by the access device, etc. Then the line of sight path characteristics are extracted from the channel information, such as RSRP, TOA, power delay spectrum, channel matrix, etc. Optionally, the random subspace method is used to select the channel information, and then the selected channel information is extracted.

[0075] The path recognition model can include one or more of a first path recognition model trained based on a K-nearest neighbor (KNN) classification algorithm, a second path recognition model trained based on a support vector machine (SVM), and a third path recognition model trained based on a gradient boosting decision tree (GBDT). In addition, the path recognition model can also include one or more of a path recognition model trained based on a multilayer perceptron (MLP), a path recognition model trained based on a recurrent neural network (RNN), or a path recognition model trained based on a convolutional neural network (CNN). Different path recognition models in the path recognition model can obtain a line-of-sight path probability of a signal transmission path between the terminal and the access device.

[0076] When the line-of-sight path probability output by the path recognition model is greater than or equal to a preset threshold, it indicates that the signal transmission path between the device to be positioned and the access device is a line-of-sight path, and when the line-of-sight path probability output by the path recognition model is less than the preset threshold, the signal transmission path between the device to be positioned and the access device is a non-line-of-sight path.

[0077] According to the line-of-sight path probability, a shallow fusion is performed, which means that a positioning algorithm is adaptively selected according to the line-of-sight path probability. For example, when there are more than three line-of-sight path probabilities exceeding the preset threshold, a geometric positioning method is selected. Or, when there is one or two line-of-sight path probabilities exceeding the preset threshold, a fingerprint positioning method is selected. In another example, the number of line-of-sight path probabilities exceeding the preset threshold is denoted as N l , and the total number of line-of-sight path probabilities is denoted as N t , when N l ≥ 3 and is greater than or equal to a preset proportion, the geometric positioning method and the fingerprint positioning method are respectively performed, and a terminal position is determined according to a first terminal coordinate calculated by the geometric positioning method and a second terminal coordinate obtained by the fingerprint positioning method.

[0078] In the wireless positioning method of the present application, the execution subject of the line-of-sight path probability obtained by the path recognition model can be an access device or a wireless positioning device. First, the method of obtaining the line-of-sight path probability by the access device through the path recognition model is introduced, referring to Figure 5 One embodiment of the wireless positioning method provided by the present application includes:

[0079] Step 501, the access device receives a positioning measurement signal sent by a device to be positioned.

[0080] The device to be positioned can be, but is not limited to, a terminal. In different communication systems, the specific type of the positioning measurement signal is different. In a cellular network, the positioning measurement signal can be, but is not limited to, a channel sounding reference signal (SRS), which is also called a sounding signal or an uplink sounding signal. In a WIFI network, the positioning measurement signal is a WIFI signal. In a Bluetooth network, the positioning measurement signal is a Bluetooth signal.

[0081] Step 502, the access device determines a power delay profile according to the positioning measurement signal.

[0082] Optionally, when the positioning measurement signal is an SRS, the access device performs channel estimation according to the SRS to obtain a channel matrix in the frequency domain, and inverse fast Fourier transform (IFFT) of the channel matrix can obtain a power delay profile (PDP) in the time domain. The power delay profile is also called a power delay profile (PDP). Another optional example is that when the positioning measurement signal is a WIFI signal, the access device determines multipath information according to the WIFI signal, and determines the power delay profile according to the multipath information. Another optional example is that when the positioning measurement signal is a Bluetooth signal, the multipath information is determined according to the Bluetooth signal, and the power delay profile is determined according to the multipath information. It should be understood that the positioning measurement signal is not limited to the above examples.

[0083] Step 503, the access device inputs the power delay profile into a path identification model for feature extraction.

[0084] In an optional example, step 503 includes: normalizing the power delay profile; performing nonlinear transformation on the normalized power delay profile using a preset weight matrix and a preset bias vector; and performing product operation on the preset weight vector and the vector obtained by the nonlinear transformation to obtain a line-of-sight path probability of a signal transmission path between the device to be positioned and the access device. The line-of-sight path probability can be understood as a feature value of the power delay profile.

[0085] The normalization can be, but is not limited to, z-score normalization. The preset weight matrix, the preset bias vector, and the preset weight vector are model configuration parameters of the path identification model.

[0086] Step 504, the access device sends the line-of-sight path probability extracted by the feature extraction to a wireless positioning device.

[0087] Step 505, the wireless positioning device obtains a target line-of-sight path probability greater than or equal to a preset threshold from all line-of-sight path probabilities.

[0088] After the wireless positioning device obtains the line-of-sight path probabilities sent by the plurality of access devices, the wireless positioning device can obtain target line-of-sight path probabilities greater than or equal to a preset threshold from all the line-of-sight path probabilities. The preset threshold can be set to different values in different scenarios, and can be set according to actual conditions, which is not limited in the present application. For example, in a residential or office scenario, the preset threshold can be set to 0.7. In a factory scenario, the preset threshold can be set to 0.8.

[0089] Step 506, when the number of target line-of-sight path probabilities is greater than or equal to three, selecting three access devices from the access devices corresponding to the target line-of-sight path probabilities.

[0090] The access device corresponding to the target line-of-sight path probability refers to the access device sending the target line-of-sight path probability. When the number of target line-of-sight path probabilities is greater than or equal to three, it indicates that at least three line-of-sight paths can be used for geometric positioning. When the number of target line-of-sight path probabilities is less than three, it indicates that it is difficult to perform geometric positioning, and fingerprint positioning can be used.

[0091] Step 507, obtaining the time of arrival of the device to be positioned to the three access devices.

[0092] Optionally, after the access device receives the positioning measurement signal sent by the device to be positioned, the time of arrival between the device to be positioned and the access device can be determined according to the positioning measurement signal. After the access device obtains the above-mentioned time of arrival, the time of arrival is sent to the wireless positioning device. The wireless positioning device can select the time of arrival sent by the above-mentioned three access devices from the time of arrival sent by the plurality of access devices.

[0093] Another optional, after the device to be positioned receives the downlink measurement signal sent by the access device, the time of arrival is determined according to the downlink measurement signal. Then the device to be positioned sends the time of arrival to the access device, and the access device forwards the time of arrival to the wireless positioning device.

[0094] Step 508, determining the position of the device to be positioned according to the time of arrival of the device to be positioned to the three access devices.

[0095] Optionally, step 508 includes: determining the time difference of arrival between the access devices according to the time of arrival of the device to be positioned to the three access devices; determining the position of the device to be positioned according to the time difference of arrival. Specifically, TDOA refers to the difference between two TOAs, and three TDOAs can be obtained according to three TOAs, then two TDOAs are selected from the three TDOAs, and the position of the device to be positioned is calculated according to the two TDOAs and the preset position of the access device.

[0096] Alternatively, step 508 comprises: calculating the position of the device to be positioned according to the three TOAs and the preset access device positions. Specifically, the TOAs are used to determine the distances from the device to be positioned to the access devices, the preset access device positions and the three distances are substituted into the least square method, and the position of the device to be positioned is calculated.

[0097] In this embodiment, the wireless positioning apparatus can filter the TOAs of the line-of-sight paths from all the TOAs, and then perform geometric positioning according to the TOAs of the line-of-sight paths. Since the TOAs of the non-line-of-sight paths are removed, the error caused by the TOAs of the non-line-of-sight paths can be reduced, and the positioning accuracy can be improved. For example, in a mixed scene including 3 line-of-sight paths and 1 non-line-of-sight path, the geometric positioning accuracy of the present application can be improved by more than 20 cm compared with the existing wireless positioning method. In the mixed scene, the geometric positioning accuracy of the present application is not only higher than that of other geometric positioning methods, but also higher than the fingerprint positioning accuracy. Moreover, in an indoor scene, the line-of-sight path recognition accuracy of the present application is more than 80%, and the present application has good feasibility.

[0098] Secondly, the size of the line-of-sight path probability can be but is not limited to 1 byte, and can be set according to actual conditions. The amount of data sent by the access device to the wireless positioning apparatus is small, so that the access device can send more line-of-sight path probabilities of users to the wireless positioning apparatus, and the wireless positioning apparatus can receive more line-of-sight path probabilities sent by the access device to obtain more position information of users and provide position services for more users.

[0099] In another optional embodiment, the above wireless positioning method further comprises: obtaining a reference signal received power of the device to be positioned, the reference signal received power being determined according to a reference signal sent by the device to be positioned; and when the number of target line-of-sight path probabilities is less than three, determining the position of the device to be positioned as the position corresponding to the reference signal received power and the TOAs in the preset fingerprint database.

[0100] In this embodiment, the access device can determine the reference signal received power according to the reference signal sent by the device to be positioned, and the reference signal can be a channel state information reference signal (CSI-RS). The wireless positioning apparatus can use the reference signal received powers and the TOAs sent by the plurality of access devices as terminal fingerprints, and find the corresponding terminal positions in the preset fingerprint database according to the reference signal received powers and the TOAs sent by the plurality of access devices.

[0101] It should be understood that in addition to the reference signal received power and the time of arrival, other channel characteristics such as a power delay profile or a channel matrix can be obtained by the application, and one or more channel characteristics can be used as the terminal fingerprint. In this way, the terminal fingerprint at the actual positioning time can be compared with the preset fingerprint in the fingerprint library, so that fingerprint positioning is realized. Using multiple channel characteristics for fingerprint positioning has higher accuracy.

[0102] The method of obtaining the line-of-sight path probability by the wireless positioning device through the path identification model will be introduced below. For reference Figure 6 Another embodiment of the wireless positioning method provided by the application includes:

[0103] Step 601, the access device receives the positioning measurement signal sent by the terminal.

[0104] Step 602, the access device determines the power delay profile according to the positioning measurement signal.

[0105] Steps 601 to 602 are similar to steps 501 to 502, and will not be described here.

[0106] Step 603, the access device sends the power delay profile to the wireless positioning device.

[0107] Step 604, the wireless positioning device inputs the power delay profile into the path identification model for feature extraction.

[0108] Optionally, after the wireless positioning device receives the power delay profile sent by the access device, the power delay profile is standardized; the standardized power delay profile is nonlinearly transformed using a preset weight matrix and a preset bias vector; the preset weight vector is multiplied with the vector obtained by the nonlinear transformation to obtain the line-of-sight path probability of the signal transmission path between the device to be positioned and the access device. The standardization can be but is not limited to z-score standardization. The preset weight matrix, the preset bias vector, and the preset weight vector are model configuration parameters of the path identification model.

[0109] Step 605, the wireless positioning device obtains target line-of-sight path probabilities greater than or equal to a preset threshold from all line-of-sight path probabilities.

[0110] Step 606, when the number of target line-of-sight path probabilities is greater than or equal to three, the wireless positioning device selects three access devices from the access devices corresponding to the target line-of-sight path probabilities.

[0111] Step 607, the wireless positioning device obtains the time of arrival of the device to be positioned to the three access devices.

[0112] Step 608, the wireless positioning device determines the position of the device to be positioned according to the time of arrival of the device to be positioned to the three access devices.

[0113] Specifically, steps 605 to 608 are similar to steps 505 to 508, which will not be described here.

[0114] In this embodiment, the wireless positioning device can filter out the arrival times of the line-of-sight paths from all the arrival times, and then perform geometric positioning according to the arrival times of the line-of-sight paths. Since the arrival times of the non-line-of-sight paths are removed, the error caused by the arrival times of the non-line-of-sight paths can be reduced, and the positioning accuracy can be improved.

[0115] Secondly, the wireless positioning device can also obtain the line-of-sight path probability according to the power delay profile sent by the access device, which provides flexibility for the implementation of the scheme.

[0116] In another optional embodiment, the above wireless positioning method further comprises: obtaining a reference signal received power of the device to be positioned, the reference signal received power being determined according to a reference signal sent by the device to be positioned; and when the number of target line-of-sight path probabilities is less than three, determining the position of the device to be positioned as the position corresponding to the reference signal received power and the arrival time in the preset fingerprint database. In this embodiment, the method of fingerprint positioning is used to determine the position of the device to be positioned when the number of target line-of-sight path probabilities is less than three, which is similar to the optional embodiment of the embodiment shown in the description. Figure 5 The optional embodiment of the embodiment shown in the description is similar, which will not be described here.

[0117] The system architecture of the training path recognition model of the present application will be introduced below.

[0118] Figure 7 FIG. 7 is a schematic diagram of the system architecture 700 of the training model of the present application. Referring to FIG. 7, the system architecture 700 of the training model of the present application comprises a training device 720, a database 730, an access device 710, and a client device 740. Figure 7 In one example, the data acquisition device 760 can acquire a plurality of power delay profiles, store the data stream (i.e., the plurality of power delay profiles) in the database 730, and the training device 720 can perform model training on the plurality of power delay profiles to obtain a target model / rule 701, and then deploy the target model / rule 701 in the access device 710.

[0119] The client device 740 sends a positioning measurement signal to the access device 710, and after the access device 710 receives the positioning measurement signal through the transceiver 712, the power delay profile module 713 can obtain a power delay profile according to the positioning measurement signal, and then the calculation module 711 can calculate the power delay profile using the target model / rule 701 to obtain the line-of-sight path probability corresponding to the power delay profile. It should be understood that the calculation module 711 can also calculate the non-line-of-sight path probability using the target model / rule 701 on the power delay profile, and then determine whether a signal transmission path is a line-of-sight path through the non-line-of-sight path probability, because the sum of the line-of-sight path probability and the non-line-of-sight path probability is equal to 1.

[0120] After the line-of-sight probability is acquired by the access device 710, the access device 710 can store the line-of-sight probability in the data storage system 750. The access device 710 can also send the line-of-sight probability to the client device 740.

[0121] Figure 8 FIG. 8 illustrates a system architecture 800 for training a model of the present disclosure. Referring to FIG. 8, in another example, the data collection device 860 can collect a plurality of power delay profiles, store the data stream (i.e., the plurality of power delay profiles) in the database 830, and the training device 820 can train the plurality of power delay profiles to obtain a target model / rule 801, and then deploy the target model / rule 801 in the wireless positioning apparatus 810. The database 830 can also receive the data stream (i.e., the plurality of power delay profiles) from the access device 840.

[0122] The access device 840 sends the power delay profile to the wireless positioning apparatus 810, the wireless positioning apparatus 810 receives the power delay profile through the I / O interface 812, and the computing module 811 calculates the line-of-sight probability corresponding to the power delay profile by using the target model / rule 801. It should be understood that the computing module 811 can also calculate a non-line-of-sight probability by using the target model / rule 801, and then determine whether a signal transmission path is a line-of-sight path by using the non-line-of-sight probability, because the sum of the line-of-sight probability and the non-line-of-sight probability is equal to 1.

[0123] After the line-of-sight probability is acquired by the wireless positioning apparatus 810, the wireless positioning apparatus 810 can store the line-of-sight probability in the data storage system 850. The wireless positioning apparatus 810 can also send the line-of-sight probability to the access device 840.

[0124] Referring to FIG. 8, Figure 9 An embodiment of a method for training a path identification model provided by the present disclosure includes the following steps.

[0125] In step 901, a first sample set is acquired, and each sample in the first sample set includes a plurality of power delay profile data.

[0126] In step 902, a path identification vector is acquired, and the path identification vector includes path identifications corresponding to the samples one by one.

[0127] In this embodiment, the path identification vector includes a plurality of path identifications, i.e., sample labels, and the path identifications correspond to the samples one by one. Each sample is a group of power delay profile data, and each group of power delay profile data can include the same number of power delay profile data. One power delay profile data corresponds to one sampling point of a power delay profile.

[0128] Optionally, when the path identifier is 0, it represents a line-of-sight path, and when the path identifier is 1, it represents a non-line-of-sight path. Alternatively, when the path identifier is 1, it represents a line-of-sight path, and when the path identifier is 0, it represents a non-line-of-sight path. It should be understood that the value of the path identifier is not limited to the above examples, and the correspondence between the value of the path identifier and the path type is also not limited to the above examples.

[0129] Step 903: Training the path identification model with the first sample set as input and the model classification error being less than or equal to a preset error as a target to obtain a trained path identification model.

[0130] The model classification error is used to indicate the difference between the model classification result of the path identification model and the path identifier vector, and the model classification result is obtained by classifying the line-of-sight path probability output by the path identification model.

[0131] The algorithm for training the path identification model can be, but is not limited to, a nearest neighbor algorithm, a support vector machine algorithm, a gradient boosting decision tree algorithm, a linear discriminant analysis algorithm, or a random nonlinear discriminant analysis algorithm.

[0132] The embodiment discloses a training method of a path identification model, which can obtain a line-of-sight path probability corresponding to a power delay spectrum to determine whether a signal transmission path is a line-of-sight path.

[0133] In an optional embodiment, training the path identification model with the first sample set as input and the model classification error being less than or equal to a preset error as a target includes:

[0134] Step A: Standardizing the first sample set.

[0135] Optionally, step A includes: dividing the first sample set into multiple columns of power delay spectrum data, the jth column of power delay spectrum data including the jth power delay spectrum data of each sample; forming a sample center point vector by averaging each column in the first sample set; forming a sample standard deviation vector by calculating the standard deviation of each column in the first sample set; and performing operations on the first sample set, the sample center point vector, and the sample standard deviation vector according to a preset standardization formula.

[0136] The first sample set includes N samples, and each sample is a power delay spectrum. Each power delay spectrum includes M target power delay spectrum data. N and M are positive integers.

[0137] The first sample set can be denoted as:

[0138]

[0139] The average value of each column is obtained in sequence to obtain a sample center point vector μ. The jth value μj in the sample center point vector μ is the average value of the jth column.j The value x of the jth column in the first sample set ij The following formula is satisfied:

[0140]

[0141] The standard deviation of each column is obtained in sequence to obtain a sample standard deviation vector σ. The jth value σ of the sample standard deviation vector σ j The value x of the jth column in the first sample set ij The following formula is satisfied:

[0142]

[0143] The sample center point vector and the sample standard deviation vector are both 1xM vectors.

[0144] The standardization formula can be, but is not limited to, a z-score standardization formula. Specifically, the ith sample x of the first sample set i The ith sample x' of the second sample set i The following formula is satisfied:

[0145]

[0146] According to the formula, each sample of the first sample set can be standardized. i∈[1,N] and i is a positive integer, and j∈[1,M] and j is a positive integer.

[0147] Step B: Nonlinearly converting the second sample set obtained by standardization according to the candidate weight matrix and the candidate bias vector to obtain a data matrix.

[0148] M samples are sampled from an M-dimensional normal distribution N(0, 2sI), each sample including M values. The scaling factor s is an artificially set hyperparameter, and I is an M-dimensional unit matrix. The candidate weight matrix composed of the above m samples can be denoted as W, and W is an m x M matrix.

[0149] M values are sampled from a uniform distribution U(0, 2π) to obtain a candidate bias vector b.

[0150] The candidate weight matrix W, the candidate bias vector b, the second sample set X', and the data matrix Z satisfy the following formula:

[0151] Z = cos(X'W T +b T )

[0152] X' is an N x M matrix, and Z is an N x m matrix. Using the above formula, X' can be nonlinearly converted to obtain Z.

[0153] Step C: obtain the inter-class distance matrix and the intra-class distance matrix of the data matrix.

[0154] The line-of-sight path identifier is taken as an example of 0, and the non-line-of-sight path identifier is taken as an example of 1. A sample set with a path identifier of 0 is taken out from Z, and a sample center vector is calculated according to the sample set, denoted as μ'0. A sample set with a path identifier of 1 is taken out from Z, and a sample center vector calculated according to the sample set is denoted as μ'1. An inter-class distance matrix S B is calculated according to μ'0 and μ'1, and an intra-class distance matrix S W is calculated according to μ'0 and μ'1.

[0155] μ'0, μ'1 and S B satisfy the following formula:

[0156] S B =(μ'0-μ'1)(μ'0-μ'1) T

[0157] μ'0, μ'1 and S W satisfy the following formula:

[0158] S W =(μ'0-Z)(μ'0-Z) T +(μ'1-Z)(μ'1-Z) T .

[0159] Step D: determine the weight vector according to the inter-class distance matrix and the intra-class distance matrix.

[0160] Optionally, a target matrix is determined according to the inter-class distance matrix and the intra-class distance matrix; and the weight vector α is determined as a characteristic vector corresponding to a maximum generalized eigenvalue of the target matrix.

[0161] wherein the inter-class distance matrix S B , the intra-class distance matrix S W and the target matrix S' satisfy the following formula:

[0162]

[0163] Step E: determine the line-of-sight path probability vector as a product of the data matrix and the weight vector.

[0164] The line-of-sight path probability vector F, the data matrix Z and the weight vector α satisfy the following formula:

[0165] F=Zα

[0166] The line-of-sight path probability vector F includes N line-of-sight path probabilities, each corresponding to a sample.

[0167] It should be understood that steps B to E are a feature extraction process based on randomized nonlinear fisher discriminant analysis (RNFDA).

[0168] Step F: determining the classification threshold according to the inter-class distance matrix, the intra-class distance matrix and the weight vector.

[0169] The inter-class distance matrix S B The intra-class distance matrix S W The weight vector α and the classification threshold β satisfy the following formula:

[0170]

[0171] α T μ0 represents the feature value extracted based on the line-of-sight path sample center vector. α T μ1 represents the feature value extracted based on the non-line-of-sight path sample center vector.

[0172] Step G: determining the model classification error according to the line-of-sight path probability vector and the classification threshold.

[0173] Specifically, the line-of-sight path probability vector is compared with the classification threshold to obtain a model classification result of the power delay spectrum; and the model classification error is determined according to the model classification result and the path identifier corresponding to the first sample set. The model classification result includes a value obtained by classifying each line-of-sight path probability in the line-of-sight path probability vector.

[0174] The i-th element of the model classification result The i-th line-of-sight path probability F in the line-of-sight path probability vector F i And the classification threshold β satisfy the following formula:

[0175]

[0176] When F i > β, When F i ≤ β,

[0177] The path identifier vector corresponding to the first sample set is denoted as y, wherein the i-th element is denoted as y i The i-th element of the model classification result The i-th path identifier y of the path identifier vector i And the model classification error mse satisfy the following formula:

[0178]

[0179] It should be understood that a plurality of sample sets can be input into the path recognition model according to the above method, so that a plurality of model classification errors of the path recognition model can be obtained, and an average of the plurality of model classification errors can be taken as the final model classification error. Alternatively, a weighted average of the plurality of errors is taken as the model classification error. The sample set can be a training set or a test set, and the samples and the number of samples included in the sample set can be set according to actual conditions.

[0180] Step H: When the model classification error is greater than the preset error, the candidate weight matrix and the candidate bias vector are updated, and steps B to H are cyclically executed until the model classification error is less than the preset error.

[0181] When the model classification error is greater than the preset error, it indicates that the candidate weight matrix and the candidate bias vector do not meet the actual requirements. The scaling factor S is modified, a new candidate weight matrix is obtained according to the modified S, and m values are resampled from the uniform distribution U(0, 2π) to obtain a new candidate bias vector. When the model classification error is less than or equal to the preset error, the updating of the candidate weight matrix and the candidate bias vector is stopped, and step I is triggered to be executed.

[0182] Step I: When the model classification error is less than or equal to the preset error, a path recognition model corresponding to the candidate weight matrix and the candidate bias vector is obtained.

[0183] When the model classification error is less than or equal to the preset error, it indicates that the candidate weight matrix and the candidate bias vector meet the actual requirements. The candidate weight matrix and the candidate bias vector are taken as the model configuration parameters of the path recognition model. Alternatively, the scaling factor S corresponding to the candidate bias vector and the candidate weight matrix is taken as the model configuration parameter of the path recognition model.

[0184] The embodiment discloses a training method of a path recognition model, which can obtain a path recognition model meeting actual requirements.

[0185] The principle of the nonlinear transformation will be introduced as follows:

[0186] The kernel function can realize the mapping from the low-dimensional space to the high-dimensional space, which can change the two classes of points that are linearly inseparable in the low-dimensional space into the two classes of points that are linearly separable in the high-dimensional space. The kernel function can be but is not limited to a radial basis function (RBF).

[0187] p(w) is the inverse Fourier transform of the kernel function k(x, x i ). When k(x, x i ) is RBF, in k(x, x i ), x iis the center of the radial basis function, x is any point in the low-dimensional space. m weight vectors are randomly sampled from p(w), denoted as w1,...,w m . p(w) is composed of the above m weight vectors.

[0188] m values are sampled from uniform distribution U(0,2π), denoted as b1,...,b m . The sample set includes N samples, denoted as x1,...,x N . Each sample is an M-dimensional vector. The data matrix z(x) includes m samples, denoted as z1,...,z m , z i ∈R N×m , R N×m is the real number of Nxm. The i-th sample z i of the data matrix z(x) includes the following elements:

[0189] In the case of k(x,x i )≤1, p(w), k(x,x i ) and the column vector w in the weight matrix W satisfy the following formula:

[0190]

[0191] where w∈R M , R M is the real number of M.

[0192] From the above formula, it can be deduced that,

[0193]

[0194] that is,

[0195] According to the above formula, when k(x,x i ) is RBF, w i ~N(0,2sI), and

[0196] k(x,x i ), x and x i satisfy the following formula:

[0197]

[0198] s is the scaling factor.

[0199] that is, any deviation value is sampled from uniform distribution U(0,2π), and any weight vector of the weight matrix is sampled from normal distribution N(0,2sI), and according to the formula A vector in z(x) can be obtained, which has a dimension of N x 1. z(x) can be obtained according to m bias values and m weight vectors of the weight matrix. That is, z(x) is a data matrix obtained by performing nonlinear transformation on the sample set, and the nonlinear transformation can enhance the representation ability of the model.

[0200] The processing process of the path recognition model is introduced below, and the following Figure 10 In an example,

[0201] Step 1: standardize the to-be-processed power spectrum x.

[0202] The to-be-processed power spectrum x includes M power delay spectrum data, and the to-be-processed power spectrum x and the standardized to-be-processed power spectrum x' satisfy the following formula:

[0203]

[0204] Step 2: perform nonlinear transformation on x' according to the preset weight matrix W and the preset bias vector b.

[0205] x' and the vector z obtained by nonlinear transformation satisfy the following formula:

[0206] z = cos(x'W T +b T )

[0207] Step 3: multiply the preset weight vector a and z.

[0208] f = z a

[0209] f is the line-of-sight path probability corresponding to the to-be-processed power spectrum x.

[0210] In another optional embodiment, obtaining the first sample set includes: obtaining a third sample set; obtaining invalid samples of the third sample set; removing the invalid samples in the third sample set to obtain the first sample set.

[0211] In this embodiment, the third sample set is data artificially collected and labeled and stored in the electronic device. The third sample set can include valid samples and invalid samples. The invalid sample refers to a sample with an incorrect label. For example, the correct label of a sample is a line-of-sight path identifier, and its label in the third sample set is incorrectly set as a non-line-of-sight path identifier. Or, the correct label of a sample is a non-line-of-sight path identifier, and its label in the third sample set is incorrectly set as a line-of-sight path identifier. Since the invalid sample can reduce the accuracy of the path recognition model, this embodiment can remove the invalid samples in the third sample set, and then use the obtained first sample set to train the model, which can reduce the influence of the invalid sample on the path recognition model.

[0212] The applicant has found that in the line-of-sight path, the signal can directly reach the receiving device from the transmitting device, so the power delay profile is a clean pulse curve with a small number of peaks. In the non-line-of-sight path, the signal reaches the receiving device through reflection or refraction, so it is divided into multiple components to reach the receiving device, and the signal is received successively, so the transmission path of the signal is not the line-of-sight path. Based on this, the present application provides a method for removing invalid samples, which is described as follows:

[0213] In another optional embodiment, the invalid samples in the third sample set include: selecting a power delay profile to be processed corresponding to the line-of-sight path label from the third sample set; dividing the power delay profile to be processed into multiple intervals; selecting target power delay profile data from each interval, determining a set of bending angles of the power delay profile to be processed according to the target power delay profile data in each interval, and each bending angle in the set of bending angles is determined according to the target power delay profile data of three consecutive intervals; determining a target angle greater than a preset angle and less than 180 degrees in the set of bending angles; and determining that the power delay profile to be processed is an invalid sample when the number of target angles is greater than or equal to a preset number.

[0214] In this embodiment, the power delay profile to be processed is divided into r intervals, and r is a positive integer. The value of r can be, but is not limited to, 16. Optionally, a target power delay profile data with the maximum power value is selected from each interval, and then the target power delay profile data of each interval is connected to obtain the profile of the power delay profile. It should be understood that the method of selecting the target power delay profile data is not limited to the above examples, and a target power delay profile data with the second largest power value can also be selected from each interval, or multiple target power delay profile data can be selected from each interval, or the target power delay profile data can be selected in other ways.

[0215] The target power delay profile data of the three consecutive intervals forms a small piece of polyline, and a bending angle is obtained by calculating the target power delay profile data of the three consecutive intervals according to the cosine theorem. The bending angle is the bending angle of the small piece of polyline. The set of bending angles obtained from the target power delay profile data of the r intervals includes r-2 bending angles. The target angle greater than the preset angle and less than 180 degrees is selected from the r-2 bending angles.

[0216] When the number of target angles is greater than or equal to the preset number, it indicates that the power delay profile to be processed has more peaks, and the path label corresponding to the power delay profile to be processed should be the non-line-of-sight path, not the line-of-sight path. When the number of target angles is less than the preset number, it indicates that the number of peaks of the power delay profile to be processed is small, and the path label corresponding to the power delay profile to be processed should be the line-of-sight path. In this way, the invalid samples can be selected from the third sample set, and the first sample set is obtained after removing the invalid samples.

[0217] Figure 11A is a schematic diagram of a power-time delay spectrum of a line-of-sight path. Referring to Figure 11A , the horizontal axis is time, in nanoseconds. The vertical axis is power, in decibel-milliwatts. It can be seen that, of the 14 bending angles, only 2 target angles exist.

[0218] Figure 11B is a schematic diagram of a power-time delay spectrum of a non-line-of-sight path. Referring to Figure 11B , the horizontal axis is time, in nanoseconds. The vertical axis is power, in decibel-milliwatts. Of the 14 bending angles, 7 target angles exist. When the preset number is 3, 4, 5, 6, or 7, the power-time delay spectrum of the non-line-of-sight path can be distinguished.

[0219] The wireless positioning method and the training method of the path identification model of the present application are introduced above, and the wireless positioning device of the present application is introduced below. Referring to Figure 12 , one embodiment of the wireless positioning device 1200 of the present application includes:

[0220] A first acquisition module 1201 is configured to acquire a line-of-sight path probability of a signal transmission path between a to-be-positioned device and an access device;

[0221] A second acquisition module 1202 is configured to acquire, from all line-of-sight path probabilities, a target line-of-sight path probability greater than or equal to a preset threshold;

[0222] A selection module 1203 is configured to, when the number of target line-of-sight path probabilities is greater than or equal to three, select three access devices from the target line-of-sight path probability corresponding access devices;

[0223] A third acquisition module 1204 is configured to acquire a time of arrival of the to-be-positioned device to the three access devices;

[0224] A position solving module 1205 is configured to determine the position of the to-be-positioned device according to the time of arrival of the to-be-positioned device to the three access devices.

[0225] In an optional embodiment, the first acquisition module 1201 is specifically configured to receive a line-of-sight path probability sent by an access device, the line-of-sight path probability being obtained by the access device inputting a power-time delay spectrum into a path identification model for feature extraction.

[0226] In another optional embodiment, the first acquisition module 1201 includes:

[0227] A receiving unit is configured to receive a power-time delay spectrum sent by an access device, the power-time delay spectrum being determined according to a positioning measurement signal sent by a to-be-positioned device to the access device;

[0228] The feature extraction unit is used to input the power delay spectrum into the trained path recognition model for feature extraction, and obtain the line-of-sight path probability of the signal transmission path between the device to be located and the access device.

[0229] In another optional embodiment, the feature extraction unit is specifically used to standardize the power delay spectrum; perform a nonlinear transformation on the standardized power delay spectrum using a preset weight matrix and a preset deviation vector; and multiply the preset weight vector with the vector obtained by the nonlinear transformation to obtain the line-of-sight path probability of the signal transmission path between the device to be located and the access device.

[0230] In another alternative embodiment, the location calculation module 1205 is specifically used to determine the arrival time difference between the access devices based on the arrival time of the device to be located to the three access devices; and to determine the location of the device to be located based on the arrival time difference.

[0231] In another alternative embodiment, the wireless positioning device 1200 further includes:

[0232] The fourth acquisition module is used to acquire the reference signal receiving power of the device to be located. The reference signal receiving power is determined based on the reference signal sent by the device to be located.

[0233] The location calculation module 1205 is also used to determine the location of the device to be located as the location corresponding to the reference signal receiving power and arrival time in the preset fingerprint database when the number of target line-of-sight path probabilities is less than three.

[0234] The wireless positioning device 1200 in this embodiment can perform... Figure 5 The wireless positioning method shown in the embodiment or optional embodiment may be implemented. Figure 6 The wireless positioning method in the illustrated embodiment or optional embodiment can be specifically implemented by the location calculation device 13. The steps performed by each module or unit in the wireless positioning device 1200 and their beneficial effects can be found in [reference needed]. Figure 5 The illustrated embodiment or Figure 6 The corresponding description in the illustrated embodiment.

[0235] See Figure 13 An embodiment of a training device 1300 for a path recognition model provided in this application includes:

[0236] The acquisition module 1301 is used to acquire a first sample set, in which each sample includes multiple power time delay spectrum data;

[0237] The acquisition module 1301 is also used to acquire path identifier vectors, wherein the path identifiers included in the path identifier vectors correspond one-to-one with the samples;

[0238] The training module 1302 is configured to take the first sample set as input, train the path recognition model with a model classification error less than or equal to a preset error as a target, and obtain a trained path recognition model.

[0239] The model classification error is used to indicate the difference between the model classification result of the path recognition model and the path identification vector, and the model classification result is obtained by classifying the line-of-sight path probability output by the path recognition model.

[0240] In an optional embodiment, the training module 1302 is specifically configured to perform the following steps:

[0241] Step A: standardizing the first sample set;

[0242] Step B: performing nonlinear conversion on the standardized second sample set according to the candidate weight matrix and the candidate bias vector to obtain a data matrix;

[0243] Step C: obtaining an inter-class distance matrix and an intra-class distance matrix of the data matrix;

[0244] Step D: determining a weight vector according to the inter-class distance matrix and the intra-class distance matrix;

[0245] Step E: determining the line-of-sight path probability as the product of the distance matrix and the weight vector;

[0246] Step F: determining a classification threshold according to the inter-class distance matrix, the intra-class distance matrix, and the weight vector;

[0247] Step G: determining the model classification error according to the line-of-sight path probability and the classification threshold;

[0248] Step H: when the model classification error is greater than the preset error, updating the candidate weight matrix and the candidate bias vector, triggering steps B to H until the model classification error is less than or equal to the preset error;

[0249] Step I: when the model classification error is less than or equal to the preset error, obtaining the path recognition model corresponding to the candidate weight matrix and the candidate bias vector.

[0250] In an optional embodiment, the training module 1302 is specifically configured to form a sample center point vector by taking the average of each column in the first sample set, form a sample standard deviation vector by taking the standard deviation of each column in the first sample set, and perform operations on the first sample set, the sample center point vector, and the sample standard deviation vector according to a preset standardization formula.

[0251] In another optional embodiment, the obtaining module 1301 includes:

[0252] A first obtaining unit is configured to obtain a third sample set.

[0253] The second obtaining unit is configured to obtain invalid samples in the third sample set;

[0254] The removing unit is configured to remove the invalid samples in the third sample set to obtain the first sample set.

[0255] In another optional embodiment, the second obtaining unit is specifically configured to select, from the third sample set, a to-be-processed power delay profile corresponding to the line-of-sight path identifier; divide the to-be-processed power delay profile into a plurality of intervals; select target power delay profile data from each interval; determine a set of bending angles of the to-be-processed power delay profile according to the target power delay profile data of each interval, wherein each bending angle in the set of bending angles is determined according to target power delay profile data of three continuous intervals; determine a target angle greater than a preset angle and less than 180 degrees in the set of bending angles; and determine that the to-be-processed power delay profile is an invalid sample when the number of target angles is greater than or equal to a preset number.

[0256] The training device 1300 in this embodiment can perform the training method of the path identification model in the embodiments shown in the embodiments or optional embodiments. The steps and beneficial effects performed by each module or unit in the training device 1300 can be referred to the corresponding descriptions in the embodiments shown in the embodiments or optional embodiments. Figure 9 The training method of the path identification model in the embodiments shown in the embodiments or optional embodiments. The steps and beneficial effects performed by each module or unit in the training device 1300 can be referred to the corresponding descriptions in the embodiments shown in the embodiments or optional embodiments. Figure 9 The training method of the path identification model in the embodiments shown in the embodiments or optional embodiments. The steps and beneficial effects performed by each module or unit in the training device 1300 can be referred to the corresponding descriptions in the embodiments shown in the embodiments or optional embodiments.

[0257] Figure 14 is a structural schematic diagram of a mobile edge computing server provided by the present application. Referring to Figure 14 , the mobile edge computing server 1400 can have relatively large differences due to different configurations or performances, and can include one or more central processing units (CPUs) 1422 (for example, one or more processors) and a memory 1432, one or more storage media 1430 (for example, one or more mass storage devices) storing application programs 1442 or data 1444. Among them, the memory 1432 and the storage medium 1430 can be temporary storage or persistent storage. The programs stored in the storage medium 1430 can include one or more modules (not shown in the figure), each of which can include a series of instruction operations in the server. Further, the central processing unit 1422 can be configured to communicate with the storage medium 1430 and execute a series of instruction operations in the storage medium 1430 on the mobile edge computing server 1400.

[0258] The mobile edge computing server 1400 can also include one or more power supplies 1426, one or more wired or wireless network interfaces 1450, one or more input / output interfaces 1458, and / or one or more operating systems 1441, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, and the like.

[0259] The steps performed by the wireless positioning device or the training device of the path identification model in the above embodiments can be based on the Figure 14 The mobile edge computing server structure shown in the above embodiments.

[0260] Figure 15 is a structural diagram of a base station provided by the present application. Referring to Figure 15 In one embodiment, the base station 1500 includes a processor 1501, a memory 1502, a transceiver 1503, an antenna 1504, and a network interface 1505.

[0261] The processor 1501, the memory 1502, the transceiver 1503, and the network interface 1505 can be connected by a bus, and the transceiver 1503 and the antenna 1504 are electrically connected. The number of the processor 1501, the memory 1502, the transceiver 1503, the antenna 1504, and the network interface 1505 can be one or more.

[0262] The processor 1501 can be, but is not limited to, a CPU, a signal processor, an ASIC, or an FPGA.

[0263] The memory 1502 can be, but is not limited to, a static random access memory or a dynamic random access memory. The memory 1502 can include a main memory and a cache. The cache can be integrated in the CPU.

[0264] The network interface 1505 is connected to the core network through a link, or connected to other base stations through a wired or wireless link.

[0265] The steps performed by the access device or the training device of the path identification model in the above embodiments can be based on the Figure 15 The base station structure shown in the above embodiments. When the base station is the training device of the path identification model, the memory 1502 stores computer program code, and the processor 1501 is used to call the computer program code to perform Figure 9 The method in the above embodiments or optional embodiments.

[0266] The application provides a computer readable storage medium, which stores program code for execution by a device, the program code comprising program code for performing the wireless positioning method in the above-described embodiments, or program code for performing the training method of the path identification model in the above-described embodiments.

[0267] In the above-described embodiments, the technical solutions can be fully or partially implemented by using software, hardware, firmware or any combination thereof. When implemented by using software, the technical solutions can be fully or partially implemented in the form of a computer program product.

[0268] The computer program product comprises one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the computer program instructions fully or partially generate the processes or functions described in the application. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another computer readable storage medium, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through a wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) manner. The computer readable storage medium can be any available medium that can be stored by a computer or a data storage device such as a server, data center, etc. integrated with one or more available media. The available media can be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)), etc.

[0269] The above embodiments are only used to illustrate the technical solutions of the present application, but not limit the technical solutions of the present application; although the technical solutions of the present application are described in detail with reference to the above-described embodiments, those skilled in the art should understand that the technical solutions recorded in the above-described embodiments can be modified, or some technical features can be replaced by equivalent features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A wireless positioning method, characterized by, The method comprises: obtaining a line-of-sight path probability of a signal transmission path between a to-be-positioned device and an access device, the line-of-sight path probability being a probability that the signal transmission path between the to-be-positioned device and the access device is a line-of-sight path; obtaining a target line-of-sight path probability greater than or equal to a preset threshold from all line-of-sight path probabilities; when the number of target line-of-sight path probabilities is greater than or equal to three, selecting three access devices from the access devices corresponding to the target line-of-sight path probabilities; obtaining a time of arrival of the to-be-positioned device to the three access devices; determining a position of the to-be-positioned device according to the time of arrival of the to-be-positioned device to the three access devices. The method comprises: receiving a power-time delay spectrum sent by an access device, the power-time delay spectrum being determined according to a positioning measurement signal sent by a to-be-positioned device to the access device; inputting the power-time delay spectrum into a trained path recognition model to perform feature extraction, to obtain a line-of-sight path probability of a signal transmission path between the to-be-positioned device and the access device.

2. The wireless positioning method of claim 1, wherein, The method comprises: receiving a line-of-sight path probability sent by an access device, the line-of-sight path probability being obtained by inputting a power-time delay spectrum into a path recognition model to perform feature extraction.

3. The wireless positioning method of claim 1, wherein, The method comprises: standardizing the power-time delay spectrum; performing nonlinear transformation on the standardized power-time delay spectrum using a preset weight matrix and a preset bias vector; performing product operation on a preset weight vector and a vector obtained through nonlinear transformation, to obtain a line-of-sight path probability of a signal transmission path between the to-be-positioned device and the access device.

4. The wireless positioning method of any one of claims 1 to 3, wherein, The method comprises: determining a time difference of arrival between access devices according to the time of arrival of the to-be-positioned device to the three access devices; determining a position of the to-be-positioned device according to the time difference of arrival.

5. The wireless positioning method of any one of claims 1 to 3, wherein, The method further comprises: obtaining a reference signal received power of the to-be-positioned device, the reference signal received power being determined according to a reference signal sent by the to-be-positioned device; when the number of target line-of-sight path probabilities is less than three, determining that the position of the to-be-positioned device is a position corresponding to the reference signal received power and the time of arrival in a preset fingerprint database. 6.A method for training a path recognition model, the method comprising: The method comprises: obtaining a first sample set, each sample in the first sample set comprising a plurality of power-time delay spectrum data, the power-time delay spectrum data being determined according to a positioning measurement signal sent by a to-be-positioned device to an access device; obtaining a path identification vector, the path identification vector comprising path identifications corresponding to the samples one by one; training a path recognition model with the first sample set as input and with a model classification error less than or equal to a preset error as a target, to obtain a trained path recognition model; The model classification error is used to indicate a difference between a model classification result of the path identification model and the path identification vector, and the model classification result is obtained by classifying a line-of-sight path probability output by the path identification model, and the line-of-sight path probability is a probability that a signal transmission path between the device to be positioned and the access device is a line-of-sight path. 7.The method of Claim 6, wherein, The training of the path identification model by taking the first sample set as input and taking the model classification error less than or equal to a preset error as a target comprises the following steps: Step A: standardizing the first sample set; Step B: performing nonlinear conversion on the second sample set obtained by standardization according to a candidate weight matrix and a candidate bias vector to obtain a data matrix; Step C: obtaining an inter-class distance matrix and an intra-class distance matrix of the data matrix; Step D: determining a weight vector according to the inter-class distance matrix and the intra-class distance matrix; Step E: determining a line-of-sight path probability vector as a product of the data matrix and the weight vector; Step F: determining a classification threshold according to the inter-class distance matrix, the intra-class distance matrix, and the weight vector; Step G: determining a model classification error according to the line-of-sight path probability vector and the classification threshold; Step H: when the model classification error is greater than a preset error, updating the candidate weight matrix and the candidate bias vector, and triggering steps B to H; Step I: when the model classification error is less than or equal to a preset error, obtaining a path identification model corresponding to the candidate weight matrix and the candidate bias vector. 8.The method of claim 7, wherein, The standardization of the first sample set comprises the following steps: dividing the first sample set into multiple columns of power delay profile data, the jth column of power delay profile data comprising jth power delay profile data of each sample, wherein j is a positive integer; composing a sample center point vector by using average values of columns in the first sample set; composing a sample standard deviation vector by using standard deviations of columns in the first sample set; performing operation on the first sample set, the sample center point vector, and the sample standard deviation vector according to a preset standardization formula. 9.The method of Claim 6 or 7, wherein, The obtaining of the first sample set comprises the following steps: obtaining a third sample set; obtaining invalid samples of the third sample set; obtaining the first sample set by removing the invalid samples from the third sample set. 10.The method of Claim 9, wherein, The obtaining of the invalid samples of the third sample set comprises the following steps: selecting a power delay profile to be processed corresponding to a line-of-sight path identification from the third sample set; dividing the power delay profile to be processed into multiple intervals; selecting target power delay profile data from each interval; determining a set of bending angles of the power delay profile to be processed according to the target power delay profile data of each interval, wherein each bending angle in the set of bending angles is determined according to target power delay profile data of three continuous intervals; determining a target angle greater than a preset angle and less than 180 degrees in the set of bending angles; when the number of target angles is greater than or equal to a preset number, determining the power delay profile to be processed as an invalid sample.

11. A wireless positioning device, comprising: The method comprises the following steps: The first obtaining module is configured to obtain a line-of-sight path probability of a signal transmission path between the to-be-positioned device and an access device, the line-of-sight path probability being a probability that the signal transmission path between the to-be-positioned device and the access device is a line-of-sight path. The second obtaining module is configured to obtain, from all the line-of-sight path probabilities, a target line-of-sight path probability greater than or equal to a preset threshold. The selecting module is configured to select three access devices from the access devices corresponding to the target line-of-sight path probabilities when the number of the target line-of-sight path probabilities is greater than or equal to three. The third obtaining module is configured to obtain a time of arrival of the to-be-positioned device to the three access devices. The position resolving module is configured to determine the position of the to-be-positioned device according to the time of arrival of the to-be-positioned device to the three access devices. The first obtaining module includes: The receiving unit is configured to receive a power delay profile sent by an access device, the power delay profile being determined according to a positioning measurement signal sent by the to-be-positioned device to the access device. The feature extraction unit is configured to input the power delay profile into a trained path recognition model to perform feature extraction, to obtain a line-of-sight path probability of a signal transmission path between the to-be-positioned device and the access device.

12. The wireless positioning device of claim 11, wherein, The first obtaining module is specifically configured to receive a line-of-sight path probability sent by an access device, the line-of-sight path probability being obtained by inputting a power delay profile into a path recognition model to perform feature extraction.

13. The wireless positioning device of claim 11, wherein, The feature extraction unit is specifically configured to normalize the power delay profile, perform nonlinear transformation on the normalized power delay profile by using a preset weight matrix and a preset bias vector, and perform product operation on a preset weight vector and a vector obtained through the nonlinear transformation, to obtain the line-of-sight path probability of the signal transmission path between the to-be-positioned device and the access device.

14. The wireless positioning device according to any one of claims 11 to 13, wherein The position resolving module is specifically configured to determine a time difference of arrival between the access devices according to the time of arrival of the to-be-positioned device to the three access devices, and determine the position of the to-be-positioned device according to the time difference of arrival.

15. The wireless location device of any one of claims 11 to 13, wherein, The wireless positioning device further includes: The fourth obtaining module is configured to obtain a reference signal received power of the to-be-positioned device, the reference signal received power being determined according to a reference signal sent by the to-be-positioned device. The position resolving module is further configured to determine the position of the to-be-positioned device as a position corresponding to the reference signal received power and the time of arrival in a preset fingerprint database when the number of the target line-of-sight path probabilities is less than three. 16.A device for training a path recognition model, comprising: The obtaining module is configured to obtain a first sample set, each sample in the first sample set including a plurality of power delay profile data, the power delay profile data being determined according to a positioning measurement signal sent by a to-be-positioned device to an access device. The obtaining module is further configured to obtain a path identification vector, the path identification vector including path identifications corresponding to the samples one by one. ​ The training module is configured to take the first sample set as input, train the path recognition model with a model classification error less than or equal to a preset error as a target, and obtain a trained path recognition model. The model classification error is used to indicate a difference between a model classification result of the path recognition model and the path identification vector, and the model classification result is obtained by classifying a line-of-sight path probability output by the path recognition model, and the line-of-sight path probability is a probability that a signal transmission path between the device to be positioned and the access device is a line-of-sight path.

17. The apparatus for training a path recognition model according to claim 16, wherein, The training module is specifically configured to perform the following steps: Step A: standardizing the first sample set; Step B: performing nonlinear conversion on the second sample set obtained by standardization according to a candidate weight matrix and a candidate bias vector to obtain a data matrix; Step C: obtaining an inter-class distance matrix and an intra-class distance matrix of the data matrix; Step D: determining a weight vector according to the inter-class distance matrix and the intra-class distance matrix; Step E: determining a line-of-sight path probability as a product of the distance matrix and the weight vector; Step F: determining a classification threshold according to the inter-class distance matrix, the intra-class distance matrix, and the weight vector; Step G: determining a model classification error according to the line-of-sight path probability and the classification threshold; Step H: when the model classification error is greater than a preset error, updating the candidate weight matrix and the candidate bias vector, triggering steps B to H until the model classification error is less than or equal to the preset error; Step I: when the model classification error is less than or equal to the preset error, obtaining the path recognition model corresponding to the candidate weight matrix and the candidate bias vector.

18. The path recognition model training apparatus according to claim 17, wherein, The training module is specifically configured to divide the first sample set into multiple columns of power delay profile data, the jth column of power delay profile data including jth power delay profile data of each sample, j being a positive integer; compose a sample center point vector by using average values of columns in the first sample set; compose a sample standard deviation vector by using standard deviations of columns in the first sample set; and perform operations on the first sample set, the sample center point vector, and the sample standard deviation vector according to a preset standardization formula.

19. The path recognition model training apparatus according to claim 16 or 17, characterized by, The obtaining module includes: A first obtaining unit configured to obtain a third sample set; A second obtaining unit configured to obtain invalid samples of the third sample set; A removing unit configured to remove the invalid samples in the third sample set to obtain a first sample set.

20. The path identification model training apparatus according to claim 19, wherein, The second obtaining unit is specifically configured to select a power delay profile to be processed corresponding to a line-of-sight path identification from the third sample set; and divide the power delay profile to be processed into multiple intervals. Target power delay spectrum data is selected from each interval, a set of bending angles of the power delay spectrum to be processed is determined according to the target power delay spectrum data of each interval, each bending angle in the set of bending angles is determined according to the target power delay spectrum data of three continuous intervals, a target angle greater than a preset angle and less than 180 degrees in the set of bending angles is determined, and when the number of the target angles is greater than or equal to a preset number, the power delay spectrum to be processed is determined as an invalid sample.

21. A wireless positioning device, comprising: A device comprising a processor and a memory storing program instructions for execution by the processor to perform the method of any one of claims 1 to 5. 22.A device for training a path recognition model, comprising: A device comprising a processor and a memory storing program instructions for execution by the processor to perform the method of any one of claims 6 to 10.

23. A computer-readable storage medium, characterized in that, The computer readable medium stores program code for execution by a device, the program code comprising program code for performing the method of any one of claims 1 to 10.

Images