A method and system for fingerprint positioning without access point based on multipath information

Abstract

The present invention discloses a fingerprint positioning method and system without access points based on multipath information. The method comprises the following steps: (1) setting a positioning signal transceiver and an error signal transceiver in a positioning area, and recording coordinate matrices respectively; (2) constructing a positioning multipath fingerprint set consisting of a positioning label set and corresponding multipath information; (3) constructing an error multipath fingerprint set consisting of an error label set and corresponding multipath information; (4) establishing an 8-layer convolutional neural network for a fingerprint positioning scenario; (5) testing and training the multi-layer convolutional neural network to simulate actual positioning accuracy; (6) collecting corresponding online multipath fingerprints, inputting the trained multi-layer convolutional neural network to obtain corresponding predicted two-dimensional coordinates, and completing positioning. The present invention improves the applicability and reliability of the system in complex environments.

Description

Technical Field

[0001] The present invention relates to the field of information communication technology, and in particular to a method and system for non-access point fingerprint positioning based on multipath information. Background Art

[0002] With the development of the digital economy, data has become increasingly important as one of the main driving forces for social progress. In the era of the Internet of Things, the demand for location information has also received increasing attention. The 6G white paper clearly points out the indicators and challenges related to wireless positioning technology. In outdoor positioning scenarios, mature positioning and navigation systems (GPS, GLONASS, Galileo, BDS) are often relied on to complete the positioning of the target. This is often a feasible and effective method in open outdoor scenes, but in environments where the positioning and navigation system is limited, such as: indoor environments, underground environments, scenes where the positioning and navigation system is shielded, complex urban environments, etc., due to multipath effects, shadows, fading and delay distortion, it is impossible to complete high-precision positioning in complex scenes through the positioning and navigation system. Although fingerprint positioning can be an effective technical means to effectively solve the positioning tasks in the above scenarios, in many open scenarios, stable and fixed-position access point (Ap) distribution is difficult to achieve, which poses a challenge to fingerprint positioning in wireless environments. Therefore, how to achieve high-precision positioning in an environment where the positioning and navigation system is limited and effective access points cannot be provided in the environment has become an important problem that needs to be solved urgently. Summary of the invention

[0003] Purpose of the invention: The purpose of the present invention is to provide a fingerprint positioning method and system without access points based on multipath information, so as to solve the problem that the satellite positioning and navigation system cannot complete accurate positioning under the condition of no direct line of sight (LOS), and the fixed access points required for fingerprint positioning cannot be provided in the open environment.

[0004] Technical solution: The present invention provides a method for fingerprint positioning without access points based on multipath information, comprising the following steps:

[0005] Offline stage:

[0006] (1) Positioning signal transceivers and error signal transceivers are set at equal intervals in the positioning area, and the coordinate matrices are recorded respectively;

[0007] (2) The positioning signal transceivers are grouped into three and receive and transmit signals in a rolling manner. The receiving and transmitting order of any three positioning transceivers is recorded to form a positioning rolling receiving and transmitting order set. The positioning tag set corresponding to the positioning rolling receiving and transmitting order set and the positioning multipath fingerprint set composed of the corresponding multipath information are determined.

[0008] (3) The error signal transceivers transmit and receive signals in a rolling manner in groups of three. The transmission and reception order of any three positioning transceivers is recorded to form an error rolling transmission and reception order set. The error label set corresponding to the error rolling transmission and reception order set and the error multipath fingerprint set composed of the corresponding multipath information are determined.

[0009] Training phase:

[0010] (4) Establish an 8-layer convolutional neural network for the fingerprint positioning scenario, use the positioning label set as the label set of the convolutional neural network, and the positioning multipath fingerprint set as the data set of the convolutional neural network, and train to determine the network parameters and bias parameters;

[0011] Testing phase:

[0012] (5) The error label set is used as the label set of the convolutional neural network, and the error multipath fingerprint set is used as the data set of the convolutional neural network to test and train the multi-layer convolutional neural network to simulate the actual positioning accuracy;

[0013] Online stage:

[0014] (6) Collect the corresponding online multipath fingerprints and input them into the trained multi-layer convolutional neural network to obtain the corresponding predicted two-dimensional coordinates to complete the positioning.

[0015] Furthermore, step (1) is specifically as follows: positioning signal receivers are arranged at equal intervals in the area to be positioned, and error signal transceivers are arranged near the coordinates of the positioning signal transceivers according to a certain offset error distance; wherein the error signal transceivers correspond to the positioning signal transceivers one by one; and the two-dimensional coordinates of the actual geographical locations of all positioning signal transceivers are recorded to form a positioning reference point coordinate matrix, which is recorded as ,but:

[0016] ;

[0017] The interval distance is recorded as , set the bit signal transceiver to have The signal transceivers are distributed in m rows and n columns; It represents the horizontal coordinate value of the initial reference point coordinate. Represents the ordinate value of the initial reference point coordinate; 5% of the spacing distance of the positioning signal transceiver is taken as the bias error distance, and the bias error distance is randomly added to the actual coordinate of the wireless transceiver to generate the bias error coordinate. The bias error coordinates of all error transceivers constitute the error reference point coordinate matrix, recorded as ,but:

[0018] ;

[0019] in, Represents the offset error distance, and a and b are the offset assignment parameters in the formula.

[0020] Furthermore, in step (2), the positioning rolling transceiver sequence set is specifically formed as follows: three positioning signal transceivers are selected from the positioning signal transceivers to form a group, which is recorded as the positioning tag, that is, The transceivers are arranged in positioning tags; each positioning tag identifies the specific order of rolling transceiver signals, wherein the specific order of rolling transceiver signals is: the positioning signal transceiver indicated in the positioning tag transmits one cycle clockwise to the transmitting end and the receiving end, and then transmits the signal in the reverse direction counterclockwise for one cycle; the formula is as follows:

[0021] Suppose a set of positioning tags is , and the corresponding order of rolling receiving and sending signals is ,but:

[0022] ;

[0023] in, represents the position coordinates of the i-th positioning signal transceiver, x represents the horizontal coordinate value, and y represents the vertical coordinate value; The order of sending and receiving is represented as follows: the i-th positioning signal transceiver is the sending end, and the j-th positioning signal transceiver is the receiving end.

[0024] Furthermore, in step (2), the positioning label set and the corresponding multipath information constitute the positioning multipath fingerprint set as follows: In the corresponding positioning rolling receiving and sending sequence set Under the guidance of the receiving and sending order, the signal is sent and received in sequence to extract the received multipath signal; according to the delay information of the first six multipaths from high to low received signal strength, a positioning delay vector is formed ,in, It represents the delay vector composed of the delays of the first six multipaths sorted from high to low in strength, which are sent by positioning signal transceiver a and received by positioning signal transceiver b; positioning tag The corresponding positioning scrolling sending and receiving order is: , then the corresponding delay vector group is , , , , , The delay vector group is the positioning delay matrix Finally, we get the positioning delay fingerprint set corresponding to the positioning label set ,but:

[0025] ;

[0026] Send and receive signals in sequence, extract the phase information of the first six multipaths of the received multipath signal from high to low according to the received signal strength to form a positioning phase vector , represents the phase vector composed of the phases of the first six multipaths whose strengths are ranked from high to low when the positioning signal transceiver a sends a signal and the positioning signal transceiver b receives the signal; the positioning tag represents the phase vector composed of the phases of the first six multipaths whose strengths are ranked from high to low when the positioning signal transceiver sends a signal and the positioning signal transceiver receives the signal; the positioning tag The corresponding positioning scrolling sending and receiving order is: , then the corresponding phase vector group is , , , , , The phase vector group is the positioning phase matrix Finally, the positioning phase fingerprint set corresponding to the positioning label set is obtained ,but:

[0027] ;

[0028] The positioning delay fingerprint set and the positioning phase fingerprint set together constitute the positioning multipath fingerprint set, denoted as .

[0029] Furthermore, step (3) is specifically as follows: The construction method of step (2) is consistent, and the corresponding error rolling receiving and sending sequence set is established according to the error reference point coordinate matrix And the corresponding error label set ; According to the established error label set, the corresponding error multipath fingerprint set is constructed, where the error multipath fingerprint set Error delay fingerprint set and error phase fingerprint set ; Among them, the established error label set corresponds one-to-one to the positioning label set, and the error multipath fingerprint set corresponds one-to-one to the positioning multipath fingerprint set.

[0030] Furthermore, step (4) is specifically as follows: the 8-layer convolutional neural network specifically includes: a convolution path of the delay matrix, a convolution path of the phase matrix, a feature fusion module, a fully connected layer, and an output layer; wherein the first layer of the convolution path of the delay matrix is ​​set to convolution layer 1, the input is a 6*6 positioning delay matrix in the positioning delay fingerprint set, the convolution kernel is denoted by P, the first layer is set to 3*3, the step size is denoted by S, the first layer is set to 1, the padding is denoted by F, the first layer is set to 0, and the output size formula is:

[0031] ;

[0032] The output size of the first layer is a 4*4 matrix; the second layer is set as a pooling layer, the pooling kernel is set to 2*2, the step size is set to 2, and the padding is set to 0. According to the output formula, the output size is a 2*2 matrix; the third layer is set as convolution layer 2, the input is a 2*2 matrix, the convolution kernel is set to 2*2, the step size is set to 1, and the padding is set to 1. According to the output formula, the output size is a 2*2 matrix; the fourth layer is set as a ReLU activation layer, the input is a 2*2 matrix, and the output is a 2*2 matrix; the fifth layer is convolution layer 3, the input is a 2*2 matrix, the convolution kernel is set to 2*2, the step size is set to 1, and the padding is set to 0. According to the output formula, the output size is a 1*2 feature map.

[0033] Furthermore, the convolution path of the phase matrix is ​​as follows: the convolution path of the phase matrix has the same network structure as the convolution path of the delay matrix. The input is a 6*6 positioning phase matrix in the positioning phase fingerprint set, and the output is a feature map of size 1*2; the feature fusion module concatenates two 1x2 feature maps to obtain a 2x2 feature map; the concatenated feature vector is , the sixth layer of the convolutional neural network is set as the feature fusion layer, and the concatenated feature vector is:

[0034] ;

[0035] The seventh layer is set as a fully connected layer, with an input of 1*4 and an output of the intermediate layer feature vector, which is set to 256 dimensions; the eighth layer is the output layer, with an input of a 256-dimensional intermediate layer feature vector and an output of a 2D coordinate point; the hyperparameters and network bias parameters are adjusted by minimizing the mean square error loss function, and the model parameters and structure are saved after training.

[0036] Furthermore, step (5) is as follows: the positioning error threshold is set to twice the offset error distance, that is: The distance difference between the predicted coordinates and the positioning reference point coordinates is compared with the positioning error threshold. If it is less than the positioning error threshold, the positioning is accurate and the positioning accuracy meets the standard; the error multipath fingerprint set is input into the multi-layer convolutional neural network as a test set to obtain the corresponding prediction label of the error multipath fingerprint; then, the predicted label is compared with the error label in the error label set corresponding to the error multipath fingerprint, and the distance between the predicted coordinates and the positioning reference point coordinates is calculated. Finally, the average value of all distance differences is calculated as the predicted positioning error, and the distance between each set of predicted coordinates and the positioning reference point coordinates is compared with the positioning error threshold to calculate the positioning success rate.

[0037] Furthermore, step (6) is as follows: Assume that the total number of people to be located is When the number of people to be located is 1, they are randomly divided into groups. There is one person to be located left. The remaining person needs to form a three-device group with two supplementary signal transceivers and are divided into groups. The collected online multipath fingerprints and positioning fingerprint sets are in the same form and are divided into online delay fingerprints and online phase fingerprints. They are then input into the trained multi-layer convolutional neural network to obtain the corresponding predicted two-dimensional coordinates.

[0038] The invention discloses a fingerprint positioning system without access points based on multipath information, comprising: a central processing unit module, a central data storage module, a positioning signal transceiver module, an error signal transceiver module, an online data acquisition module, a supplementary signal transceiver module, a calculation processing module, a neural network module, a neural network training module and a positioning data output module; wherein the central processing unit module comprises a calculation processing module and a central data storage module, the calculation processing module is equipped with a neural network module and a neural network training module for training the network, the central data storage module is used to store intermediate data and important parameters, including: hyperparameters and bias parameters of the neural network; corresponding data set data collected by the positioning signal transceiver module and the error signal transceiver module, and online multipath fingerprint data of the device to be positioned collected by the online data acquisition module and the supplementary signal transceiver module; the collected online multipath fingerprint data is transmitted to the central data storage module, the calculation processing module calls the network parameters and the online data set in the central data storage module, and uses the trained neural network module to obtain the final two-dimensional positioning data, and outputs it to the positioning data output module.

[0039] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: Through the access point-free fingerprint positioning algorithm based on multipath information, the present invention achieves high-precision positioning in an environment where the positioning navigation system is limited and there is no fixed access point. The multi-layer convolutional neural network is used to process the delay and phase information, which improves the positioning accuracy and robustness. In addition, the problem of non-divisible number of devices is solved by supplementing the signal transceiver, the positioning speed and accuracy are optimized, and the applicability and reliability of the system in complex environments are greatly improved. This innovation not only enhances the stability of positioning, but also improves the positioning efficiency in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is a schematic diagram of the method of the present invention;

[0041] Figure 2 is a schematic diagram of the positioning reference point coordinates and the error reference point coordinates of the present invention;

[0042] Figure 3 It is a schematic diagram of the flow of the 8-layer convolutional neural network of the present invention;

[0043] Figure 4 It is a schematic diagram of the system terminal structure of the present invention. DETAILED DESCRIPTION

[0044] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0045] like Figure 1 As shown, an embodiment of the present invention provides a method for fingerprint positioning without access points based on multipath information, comprising the following steps:

[0046] Offline stage:

[0047] S1. Set up positioning signal transceivers and error signal transceivers at equal intervals in the positioning area, and record the coordinate matrix respectively. Specifically: Set up positioning signal receivers at equal intervals in the area to be positioned, and set up error signal transceivers near the coordinates of the positioning signal transceivers according to a certain offset error distance. The error signal transceivers correspond to the positioning signal transceivers one by one. Record the two-dimensional coordinates of the actual geographical locations of all positioning signal transceivers to form a positioning reference point coordinate matrix, recorded as ,but:

[0048] ;

[0049] The interval distance is recorded as , set the bit signal transceiver to have The signal transceivers are distributed in m rows and n columns; It represents the horizontal coordinate value of the initial reference point coordinate. Represents the ordinate value of the initial reference point coordinate; 5% of the spacing distance of the positioning signal transceiver is taken as the bias error distance, and the bias error distance is randomly added to the actual coordinate of the wireless transceiver to generate the bias error coordinate. The bias error coordinates of all error transceivers constitute the error reference point coordinate matrix, recorded as ,but:

[0050] ;

[0051] in, represents the bias error distance, a and b are the bias assignment parameters in the formula, and all the above matrices are transmitted to the central processing unit.

[0052] S2. The positioning signal transceivers transmit and receive signals in a rolling manner in groups of three. The transmission and reception order of any three positioning transceivers is recorded to form a positioning rolling transmission and reception order set. The positioning tag set corresponding to the positioning rolling transmission and reception order set and the positioning multipath fingerprint set composed of the corresponding multipath information are determined. Three positioning signal transceivers are selected from the positioning signal transceivers in a full arrangement to form a group, which is recorded as a positioning tag. The transceivers are arranged in positioning tags; each positioning tag identifies a specific rolling order of sending and receiving signals. The specific order of sending and receiving rolling signals is: the positioning signal transceiver indicated in the positioning tag transmits the signal to the transmitter and the receiver in a clockwise manner for one cycle, and then transmits the signal in a counterclockwise manner for one cycle; for example, suppose a group of positioning tags is , and the corresponding order of rolling receiving and sending signals is ,but:

[0053] ;

[0054] in, represents the position coordinates of the i-th positioning signal transceiver, x represents the horizontal coordinate value, and y represents the vertical coordinate value; The order of sending and receiving is represented as follows: the i-th positioning signal transceiver is the sending end, and the j-th positioning signal transceiver is the receiving end.

[0055] It must be pointed out that the rolling signal receiving method of the present invention is position-oriented. For a positioning tag composed of the same positioning signal transceivers, when the order of the positioning signal transceivers in the tag changes, the corresponding positioning rolling order will also change accordingly; the order of the positioning tag composed of the same three positioning signal transceivers is changed to: , then the corresponding order of positioning scrolling sending and receiving signals is changed to:

[0056] ;

[0057] in, Indicates the order of signal transmission and reception. The signal is sent by positioning signal transceiver No. 2 and received by positioning signal transceiver No. 3. It can be seen that although the positioning tag contains the same positioning signal transceiver, and They are not the same. The order of sending and receiving signals of positioning tags and positioning scrolling is one-to-one corresponding. All positioning tags are recorded to form a positioning tag set. ,but:

[0058] ;

[0059] in, Indicates the positioning signal transceiver in the rolling signal transmission and reception sequence, the positioning tags composed of a, b and c; the positioning rolling transmission and reception sequence set corresponding to the positioning tag set ,but:

[0060] ;

[0061] in, express The corresponding positioning scrolling receiving and sending sequence; and transmitting all the above data sets to the central processing unit.

[0062] The positioning multipath fingerprint set composed of the positioning tag set and the corresponding multipath information is as follows: In the corresponding positioning rolling receiving and sending sequence set Under the guidance of the receiving and sending order, the signal is sent and received in sequence to extract the received multipath signal; according to the delay information of the first six multipaths from high to low received signal strength, a positioning delay vector is formed ,in, It represents the delay vector composed of the delays of the first six multipaths sorted from high to low in strength, which are sent by positioning signal transceiver a and received by positioning signal transceiver b; positioning tag The corresponding positioning scrolling sending and receiving order is: , then the corresponding delay vector group is , , , , , The delay vector group is the positioning delay matrix Finally, we get the positioning delay fingerprint set corresponding to the positioning label set ,but:

[0063] ;

[0064] Send and receive signals in sequence, extract the phase information of the first six multipaths of the received multipath signal from high to low according to the received signal strength to form a positioning phase vector , represents the phase vector composed of the phases of the first six multipaths whose strengths are ranked from high to low when the positioning signal transceiver a sends a signal and the positioning signal transceiver b receives the signal; the positioning tag represents the phase vector composed of the phases of the first six multipaths whose strengths are ranked from high to low when the positioning signal transceiver sends a signal and the positioning signal transceiver receives the signal; the positioning tag The corresponding positioning scrolling sending and receiving order is: , then the corresponding phase vector group is , , , , , The phase vector group is the positioning phase matrix Finally, the positioning phase fingerprint set corresponding to the positioning label set is obtained ,but:

[0065] ;

[0066] The positioning delay fingerprint set and the positioning phase fingerprint set together constitute the positioning multipath fingerprint set, denoted as . And transmit all the above data sets to the central processing unit.

[0067] S3, the error signal transceivers transmit and receive signals in a rolling manner in groups of three, and record the transmission and reception order of any three positioning transceivers to form an error rolling transmission and reception order set, and determine the error label set corresponding to the error rolling transmission and reception order set and the error multipath fingerprint set composed of the corresponding multipath information. The error rolling transmission and reception order set is consistent with the order method of the rolling transmission and reception signal in step S2, and the corresponding error rolling transmission and reception order set is established according to the error reference point coordinate matrix And the corresponding error label set ; According to the established error label set, the corresponding error multipath fingerprint set is constructed, where the error multipath fingerprint set Error delay fingerprint set and error phase fingerprint set ; Among them, the error label set established is in one-to-one correspondence with the positioning label set, and the error multipath fingerprint set is in one-to-one correspondence with the positioning multipath fingerprint set. The corresponding data set of the error signal transceiver under the small position offset of the positioning signal transceiver is established to simulate the positioning error caused by environmental interference in the actual positioning environment, to verify the positioning effect in the actual scene, and to improve the robustness and reliability of the positioning algorithm.

[0068] Training phase:

[0069] S4. Establish an 8-layer convolutional neural network suitable for fingerprint positioning scenarios, use the positioning label set as the label set of the convolutional neural network, and use the positioning multipath fingerprint set as the data set of the convolutional neural network. Train and determine the network parameters and bias parameters for subsequent high-precision positioning. The establishment of an 8-layer convolutional neural network suitable for fingerprint positioning scenarios specifically includes: the first part is the convolution path for processing the delay matrix: the first layer is set to convolution layer 1, the input is the 6*6 positioning delay matrix in the positioning delay fingerprint set, the convolution kernel is denoted by P, the first layer is set to 3*3, the step size is denoted by S, the first layer is set to 1, the padding is denoted by F, the first layer is set to 0, and the output size formula is:

[0070] ;

[0071] The output size of the first layer is a 4*4 matrix; the second layer is set as a pooling layer, the pooling kernel is set to 2*2, the step size is set to 2, and the padding is set to 0. According to the output formula, the output size is a 2*2 matrix; the third layer is set as convolution layer 2, the input is a 2*2 matrix, the convolution kernel is set to 2*2, the step size is set to 1, and the padding is set to 1. According to the output formula, the output size is a 2*2 matrix; the fourth layer is set as a ReLU activation layer, the input is a 2*2 matrix, and the output is a 2*2 matrix; the fifth layer is convolution layer 3, the input is a 2*2 matrix, the convolution kernel is set to 2*2, and the step size is set to 1, padding is set to 0, and the output size is 1*2 feature map according to the output formula; the second part is the convolution path for processing the phase matrix: because the phase matrix is ​​exactly the same as the experimental matrix, the five-layer convolutional neural network in the second part has the same network structure as the first part, with the input being the 6*6 positioning phase matrix in the positioning phase fingerprint set, and the output size is 1*2 feature map; the third part is feature fusion, and the convolution paths for processing the experimental matrix and the phase matrix both output 1*2 feature maps, and the two 1x2 feature maps are spliced ​​to obtain a 2x2 feature map; the feature vector after splicing is , the sixth layer of the convolutional neural network is set as the feature fusion layer, and the concatenated feature vector is :

[0072] ;

[0073] The seventh layer is set as a fully connected layer, with an input of 1*4 and an output of the intermediate layer feature vector, which is set to 256 dimensions. The formula is as follows:

[0074] ;

[0075] The 8th layer is the output layer. The input is a 256-dimensional intermediate layer feature vector and the output is a 2-dimensional coordinate point. The formula is as follows:

[0076] ;

[0077] Through the above steps, an effective hierarchical neural network is established, and the obtained positioning multipath fingerprint set data is imported. By minimizing the mean square error loss function, it is recorded as The hyperparameters and network bias parameters can be tuned, the model parameters and structure can be saved after the training is completed, and the multi-layer convolutional neural network, the corresponding hyperparameters and bias values ​​after the training are transmitted to the central processing unit; the above mean square error loss function is as follows:

[0078] ;

[0079] in, Indicates the two-dimensional coordinates corresponding to the positioning label, Represents the two-dimensional coordinates predicted by the neural network; the 8-layer convolutional neural network designed by the present invention can automatically extract useful features from the original data. The use of the activation function enhances the nonlinearity of the network, and the feature maps of the delay matrix and the phase matrix are fused and spliced ​​together, utilizing information from two different feature sources, which helps to improve the expressiveness and accuracy of the model.

[0080] Testing phase:

[0081] S5. The error label set is used as the label set of the convolutional neural network, and the error multipath fingerprint set is used as the data set of the convolutional neural network to test the generalization ability of the trained multi-layer convolutional neural network and simulate the actual positioning accuracy. Specifically, the positioning error threshold is set to twice the bias error distance, that is: The distance difference between the predicted coordinates and the coordinates of the positioning reference point is compared with the positioning error threshold. If it is less than the positioning error threshold, it is recorded as accurate positioning and the positioning accuracy meets the standard; import the error label set and the error multipath fingerprint set, import the trained multi-layer convolutional neural network, and input the error multipath fingerprint set as a test set into the multi-layer convolutional neural network to obtain the predicted label corresponding to the error multipath fingerprint; then, compare the predicted label with the error label in the error label set corresponding to the error multipath fingerprint, calculate the distance between the predicted coordinates and the positioning reference point coordinates, and finally calculate the average value of all distance differences as the predicted positioning error Compare the distance between each set of predicted coordinates and the positioning reference point coordinates with the positioning error threshold to calculate the positioning success rate The predicted positioning error and positioning success rate obtained above are transmitted to the central processing unit.

[0082] Online stage:

[0083] S6. Let the personnel and equipment to be located in the positioning area carry signal transceivers, assign groups, roll the signal transmission and reception within the group, collect the corresponding online multipath fingerprints, input the trained multi-layer convolutional neural network to obtain the corresponding predicted two-dimensional coordinates, and complete the positioning. The details are as follows: set up two supplementary signal transceivers with known coordinates that are always in working state in the positioning area; randomly divide the personnel and equipment to be located into groups of three, and all groups complete the three-by-three positioning at the same time; when the total number of personnel and equipment to be located is not an integer multiple of three, select from the set supplementary signal transceivers to complete the group supplement; for example, when the total number of personnel to be located is When the number of people is 1, they can be randomly divided into n groups, and there is one person left to be located. This remaining person needs to form a three-device group with two supplementary signal transceivers, so the total number of people is group; the collected online multipath fingerprints and positioning fingerprint sets are in the same form, divided into online delay fingerprints and online phase fingerprints, and the online multipath fingerprint sets collected by all groups are transmitted to the central processing unit; the trained multi-layer convolutional neural network is input to obtain the corresponding predicted two-dimensional coordinates; it should be pointed out that, unlike the traditional fingerprint positioning method, the multi-layer convolutional neural network using deep learning greatly improves the positioning speed and accuracy, and is divided into a group of three, which not only solves the problem that effective fingerprint positioning cannot be performed without an access point, but also greatly improves the positioning speed, and creatively introduces a supplementary signal transceiver, so that the positioning scene is not affected by the number of devices and personnel to be positioned, and has wide applicability and strong robustness.

[0084] Specifically, taking a certain area in Kowloon City, Hong Kong as an example, positioning signal transceivers and error signal transceivers are set, the positioning area size is set to 200m*150m, and the spacing distance is 20*15. The signal transceivers are distributed in 20 rows and 15 columns. 5% of the spacing distance of the positioning signal transceivers is taken as the offset error distance, and the corresponding offset error distance is ; Get the corresponding positioning reference point coordinate matrix and the corresponding error reference point matrix like Figure 2 As shown, then:

[0085] ;

[0086] ;

[0087] Determine the positioning label set and positioning multipath fingerprint set corresponding to the positioning rolling receiving and sending sequence; Figure 2 Taking the group of positioning signal transceivers shown in the middle circle as an example, the corresponding order of rolling transmission and reception of signals by the three positioning signal transceivers is: , the positioning tag is , then its corresponding positioning multipath fingerprint contains the positioning delay fingerprint and localization phase fingerprint , the corresponding data are as follows:

[0088] ;

[0089] ;

[0090] According to the above method, the corresponding positioning delay fingerprint set under all positioning tags is obtained and the corresponding positioning phase fingerprint set The positioning phase fingerprint set and the positioning experiment fingerprint set constitute the positioning multipath fingerprint set The error multipath fingerprint set corresponding to the positioning multipath fingerprint set is obtained to simulate the positioning effect in the presence of interference in actual scenes and improve the robustness of positioning.

[0091] like Figure 3 As shown, the corresponding process of the 8-layer convolutional neural network provided by the present invention is demonstrated. The 8-layer convolutional neural network designed by the present invention can automatically extract useful features from the original data. The use of activation function enhances the nonlinearity of the network, and the feature maps of the delay matrix and the phase matrix are fused and spliced ​​together, utilizing information from two different feature sources, which helps to improve the expressiveness and accuracy of the model.

[0092] The error multipath fingerprint set and the corresponding error label set are used to verify the generalization ability of the trained neural network and the positioning effect in the simulated actual environment. The positioning error threshold is set to The predicted coordinates and the actual coordinates of the reference points are shown in Table 1:

[0093] Table 1 Neural network predicted coordinates and true coordinates

[0094] ;

[0095] Calculate the average of all distance differences as the predicted positioning error , compare the distance between each set of predicted coordinates and the positioning reference point coordinates with the positioning error threshold, and calculate the positioning success rate The predicted positioning error is less than the positioning error threshold, meeting the positioning accuracy requirements and the red positioning success rate in complex environments is as high as 93%, meeting the expected requirements.

[0096] The group receives and sends signals in a rolling manner, collects the corresponding online multipath fingerprints, inputs the trained multi-layer convolutional neural network to obtain the corresponding predicted two-dimensional coordinates, and completes the positioning; the method of allocating groups and the collected online multipath fingerprint data specifically include: setting two supplementary signal transceivers with known coordinates that are always in working state in the positioning area; randomly dividing the personnel and equipment to be located into groups of three, and all groups complete three-three positioning at the same time; when the total number of personnel and equipment to be located is not an integer multiple of three, select from the set supplementary signal transceivers to complete the group supplement; in this embodiment, the total number of positioning personnel is 13, which can be randomly divided into 4 groups, and there is 1 person to be located. This remaining person needs to form a three-device group with 2 supplementary signal transceivers, so a total of 5 The collected online multipath fingerprint and positioning fingerprint set are in the same form, which are divided into online delay fingerprint and online phase fingerprint, and the online multipath fingerprint sets collected by all groups are transmitted to the central processor; the trained multi-layer convolutional neural network is input to obtain the corresponding predicted two-dimensional coordinates. In the actual scenario, the predicted coordinates and the actual coordinates of the actual person to be located are shown in Table 2:

[0097] Table 2 Neural network predicted coordinates and personnel actual coordinates

[0098] ;

[0099] It should be pointed out that, unlike traditional fingerprint positioning methods, the use of multi-layer convolutional neural networks based on deep learning has greatly improved the speed and accuracy of positioning. Dividing them into groups of three not only solves the problem of inability to perform effective fingerprint positioning without access points, but also greatly improves the positioning speed. It creatively introduces supplementary signal transceivers, so that the positioning scenario is not affected by the number of devices and personnel to be positioned, and has wide applicability and strong robustness.

[0100] The embodiment of the present invention also provides a fingerprint positioning system without access point based on multipath information, such as Figure 4As shown, it includes the following modules: a central processing unit module, a central data storage module, a positioning signal transceiver module, an error signal transceiver module, an online data acquisition module, a supplementary signal transceiver module, a calculation processing module, a neural network module, a neural network training module and a positioning data output module; the central processing unit module includes a calculation processing module and a central data storage module, the calculation processing module is equipped with the neural network module provided by the present invention and the neural network training module used to train the network, the central data storage module is used to store intermediate data and important parameters, including: hyperparameters of the neural network, bias parameters, corresponding data set data collected by the positioning signal transceiver module and the error signal transceiver module, and online multipath fingerprint data of the device to be positioned collected by the online data acquisition module and the supplementary signal transceiver module; the collected online multipath fingerprint data is transmitted to the central data storage module, the calculation processing module calls the network parameters and online data sets in the central data storage module, and uses the trained neural network module to obtain the final two-dimensional positioning data, and outputs it to the positioning data output module.

Claims

1. A fingerprint positioning method without access point based on multipath information, characterized in that: The following steps are involved: Offline stage: (1) Positioning signal transceivers and error signal transceivers are set at equal intervals in the positioning area, and the coordinate matrices are recorded respectively; (2) The positioning signal transceivers are grouped into three and receive and transmit signals in a rolling manner. The receiving and transmitting order of any three positioning transceivers is recorded to form a positioning rolling receiving and transmitting order set. The positioning label set corresponding to the positioning rolling receiving and transmitting order set and the positioning multipath fingerprint set composed of the corresponding multipath information are determined. The positioning rolling receiving and transmitting order set is specifically composed as follows: three positioning signal transceivers are selected from the positioning signal transceivers in a full arrangement to form a group, which is recorded as the positioning label, i.e. The transceivers are arranged in positioning tags; each positioning tag identifies the specific order of rolling transceiver signals, wherein the specific order of rolling transceiver signals is: the positioning signal transceiver indicated in the positioning tag transmits one cycle clockwise to the transmitting end and the receiving end, and then transmits the signal in the reverse direction counterclockwise for one cycle; the formula is as follows: Suppose a set of positioning tags is , and the corresponding order of rolling receiving and sending signals is ,but: ； in, represents the position coordinates of the i-th positioning signal transceiver, x represents the horizontal coordinate value, and y represents the vertical coordinate value; The order of transmission and reception represented by is: the i-th positioning signal transceiver is the transmitting end, and the j-th positioning signal transceiver is the receiving end; The positioning multipath fingerprint set composed of the positioning tag set and the corresponding multipath information is as follows: In the corresponding positioning rolling receiving and sending sequence set Under the guidance of the receiving and sending order, the signal is sent and received in sequence to extract the received multipath signal; according to the delay information of the first six multipaths from high to low received signal strength, a positioning delay vector is formed ,in, It represents the delay vector composed of the delays of the first six multipaths sorted from high to low in strength, which are sent by positioning signal transceiver a and received by positioning signal transceiver b; positioning tag The corresponding positioning scrolling sending and receiving order is: , then the corresponding delay vector group is , , , , , The delay vector group is the positioning delay matrix Finally, we get the positioning delay fingerprint set corresponding to the positioning label set ,but: ； Send and receive signals in sequence, extract the phase information of the first six multipaths of the received multipath signal from high to low according to the received signal strength to form a positioning phase vector , represents the phase vector composed of the phases of the first six multipaths whose strengths are ranked from high to low when the positioning signal transceiver a sends a signal and the positioning signal transceiver b receives the signal; the positioning tag represents the phase vector composed of the phases of the first six multipaths whose strengths are ranked from high to low when the positioning signal transceiver sends a signal and the positioning signal transceiver receives the signal; the positioning tag The corresponding positioning scrolling sending and receiving order is: , then the corresponding phase vector group is , , , , , The phase vector group is the positioning phase matrix Finally, the positioning phase fingerprint set corresponding to the positioning label set is obtained ,but: ； The positioning delay fingerprint set and the positioning phase fingerprint set together constitute the positioning multipath fingerprint set, denoted as ; (3) The error signal transceivers transmit and receive signals in a rolling manner in groups of three. The transmission and reception order of any three positioning transceivers is recorded to form an error rolling transmission and reception order set. The error label set corresponding to the error rolling transmission and reception order set and the error multipath fingerprint set composed of the corresponding multipath information are determined. The specific details are as follows: The construction method of step (2) is consistent. The corresponding error rolling transmission and reception order set is established according to the error reference point coordinate matrix. And the corresponding error label set ; According to the established error label set, the corresponding error multipath fingerprint set is constructed, where the error multipath fingerprint set Error delay fingerprint set and error phase fingerprint set ; Among them, the error label set established is in one-to-one correspondence with the positioning label set, and the error multipath fingerprint set is in one-to-one correspondence with the positioning multipath fingerprint set; Training phase: (4) Establish an 8-layer convolutional neural network for the fingerprint positioning scenario, use the positioning label set as the label set of the convolutional neural network, and the positioning multipath fingerprint set as the data set of the convolutional neural network, and train to determine the network parameters and bias parameters; Testing phase: (5) The error label set is used as the label set of the convolutional neural network, and the error multipath fingerprint set is used as the data set of the convolutional neural network to test and train the multi-layer convolutional neural network to simulate the actual positioning accuracy; Online stage: (6) Collect the corresponding online multipath fingerprints and input them into the trained multi-layer convolutional neural network to obtain the corresponding predicted two-dimensional coordinates to complete the positioning.

2. According to claim 1, a fingerprint positioning method without access point based on multipath information is characterized in that: Step (1) is as follows: positioning signal receivers are arranged at equal intervals in the area to be positioned, and error signal transceivers are arranged near the coordinates of the positioning signal transceivers according to a certain offset error distance; wherein the error signal transceivers correspond to the positioning signal transceivers one by one; and the two-dimensional coordinates of the actual geographical locations of all positioning signal transceivers are recorded to form a positioning reference point coordinate matrix, which is recorded as ,but: ； The interval distance is recorded as , set the bit signal transceiver to have The signal transceivers are distributed in m rows and n columns; It represents the horizontal coordinate value of the initial reference point coordinate. Represents the ordinate value of the initial reference point coordinate; 5% of the spacing distance of the positioning signal transceiver is taken as the bias error distance, and the bias error distance is randomly added to the actual coordinate of the wireless transceiver to generate the bias error coordinate. The bias error coordinates of all error transceivers constitute the error reference point coordinate matrix, recorded as ,but: ； in, Represents the offset error distance, and a and b are the offset assignment parameters in the formula.

3. The method for fingerprint positioning without access point based on multipath information according to claim 1, characterized in that: Step (4) is as follows: The 8-layer convolutional neural network specifically includes: a convolution path of the delay matrix, a convolution path of the phase matrix, a feature fusion module, a fully connected layer, and an output layer; wherein the first layer of the convolution path of the delay matrix is ​​set to convolution layer 1, the input is a 6*6 positioning delay matrix in the positioning delay fingerprint set, the convolution kernel is denoted by P, the first layer is set to 3*3, the step size is denoted by S, the first layer is set to 1, the padding is denoted by F, the first layer is set to 0, and the output size formula is: ； The output size of the first layer is a 4*4 matrix; the second layer is set as a pooling layer, the pooling kernel is set to 2*2, the step size is set to 2, and the padding is set to 0. According to the output formula, the output size is a 2*2 matrix; the third layer is set as convolution layer 2, the input is a 2*2 matrix, the convolution kernel is set to 2*2, the step size is set to 1, and the padding is set to 1. According to the output formula, the output size is a 2*2 matrix; the fourth layer is set as a ReLU activation layer, the input is a 2*2 matrix, and the output is a 2*2 matrix; the fifth layer is convolution layer 3, the input is a 2*2 matrix, the convolution kernel is set to 2*2, the step size is set to 1, and the padding is set to 0. According to the output formula, the output size is a 1*2 feature map.

4. The method for fingerprint positioning without access point based on multipath information according to claim 3 is characterized in that: The convolution path of the phase matrix is ​​as follows: The convolution path of the phase matrix has the same network structure as the convolution path of the delay matrix. The input is a 6*6 positioning phase matrix in the positioning phase fingerprint set, and the output size is a feature map of 1*2. feature The fusion module concatenates two 1x2 feature maps to obtain a 2x2 feature map. The concatenated feature vector is , the sixth layer of the convolutional neural network is set as the feature fusion layer, and the concatenated feature vector is: ； The seventh layer is set as a fully connected layer, with an input of 1*4 and an output of the intermediate layer feature vector, which is set to 256 dimensions; the eighth layer is the output layer, with an input of a 256-dimensional intermediate layer feature vector and an output of a 2D coordinate point; the hyperparameters and network bias parameters are adjusted by minimizing the mean square error loss function, and the model parameters and structure are saved after training.

5. The method for fingerprint positioning without access point based on multipath information according to claim 1, characterized in that: Step (5) is as follows: Set the positioning error threshold to twice the offset error distance, that is: The distance difference between the predicted coordinates and the positioning reference point coordinates is compared with the positioning error threshold. If it is less than the positioning error threshold, the positioning is accurate and the positioning accuracy meets the standard; the error multipath fingerprint set is input into the multi-layer convolutional neural network as a test set to obtain the corresponding prediction label of the error multipath fingerprint; then, the predicted label is compared with the error label in the error label set corresponding to the error multipath fingerprint, and the distance between the predicted coordinates and the positioning reference point coordinates is calculated. Finally, the average value of all distance differences is calculated as the predicted positioning error, and the distance between each set of predicted coordinates and the positioning reference point coordinates is compared with the positioning error threshold to calculate the positioning success rate.

6. The method for fingerprint positioning without access point based on multipath information according to claim 1, characterized in that: Step (6) is as follows: Assume that the total number of people to be located is When the number of people to be located is 1, they are randomly divided into groups. There is one person to be located left. The remaining person needs to form a three-device group with two supplementary signal transceivers and are divided into groups. The collected online multipath fingerprints and positioning fingerprint sets are in the same form and are divided into online delay fingerprints and online phase fingerprints. They are then input into the trained multi-layer convolutional neural network to obtain the corresponding predicted two-dimensional coordinates.

7. A system according to any one of claims 1 to 6, characterized in that: include: A central processing unit module, a central data storage module, a positioning signal transceiver module, an error signal transceiver module, an online data acquisition module, a supplementary signal transceiver module, a calculation processing module, a neural network module, a neural network training module and a positioning data output module; wherein the central processing unit module includes a calculation processing module and a central data storage module, the calculation processing module is equipped with a neural network module and a neural network training module for training the network, the central data storage module is used to store intermediate data and important parameters, including: hyperparameters and bias parameters of the neural network; corresponding data set data collected by the positioning signal transceiver module and the error signal transceiver module, and online multipath fingerprint data of the device to be located collected by the online data acquisition module and the supplementary signal transceiver module; the collected online multipath fingerprint data is transmitted to the central data storage module, the calculation processing module calls the network parameters and online data set in the central data storage module, and uses the trained neural network module to obtain the final two-dimensional positioning data, and outputs it to the positioning data output module.

Images