Construction method, training method and application of Wi-Fi passive tracking target model training data

By constructing the training data of the Wi-Fi passive tracking target model in a non-line-of-sight environment, using the environmental coordinate system and signal flooding to generate a reflected signal path length heat map, combining user trajectory simulation and PLCR sequence, Wi-Fi passive tracking is solved The system's positioning accuracy reduction in non-line-of-sight environments has achieved higher positioning accuracy.

CN118913283APending Publication Date: 2024-11-08THE ACAD OF TIANJIN UNIV HEFEI
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410974715.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-19
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

In non-line-of-sight environments, the Wi-Fi passive tracking system reduces positioning accuracy due to the complexity of signal propagation and obstacle interference. Traditional models cannot effectively handle signal reflection, attenuation and multipath effects, resulting in limited noise and accuracy. question.

Method used

By establishing environmental coordinate system and mesh division, simulated signal flooding, and generating reflected signal path length heat maps, combining user trajectory simulation and PLCR sequences, target model training data that is more in line with actual conditions, and use long and short-term memory networks to model train.

Benefits of technology

It improves the passive tracking accuracy in non-line-of-sight environments, corrects the inaccuracy problem of the original model, and enhances the practicality and accuracy of the signal propagation model.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118913283A_ABST
    Figure CN118913283A_ABST
Patent Text Reader

Abstract

The invention provides a construction method and application of Wi-Fi passive tracking target model training data in a non-line-of-sight scene, and the core method is that a global reflection signal propagation heat map model is pre-established, and accurate non-line-of-sight positioning is realized by calculating the signal reflection path length of each indoor point. Specifically, the method comprises the following steps: generating an analogue simulation data set, training a neural network by using the data set, firstly determining a tracking area and transceiver position coordinates according to a real environment, generating training data by using a preset transceiver link and a human body simulation track in a simulation training stage, and then training the human body simulation track according to the training data. Training a network model by adopting a long-short term memory (LSTM) network; in a positioning test stage, actual channel state information (CSI) data is collected, and the characteristics of a signal reflection path length change rate (PLCR) are analyzed and input into the trained model for prediction, so that accurate non-line-of-sight positioning is realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of indoor tracking technology, and in particular to a method for constructing training data of a Wi-Fi passive tracking target model in a non-line-of-sight scenario, a training method, and an application. Background Art

[0002] In today's society, location information has become an indispensable and important source of information. Although outdoor positioning technologies such as the global positioning system are quite mature and widely used in modern scientific and technological society, these positioning technologies are difficult to achieve accurate positioning and tracking in indoor environments due to obstacles and interference. However, Wi-Fi-based indoor positioning technology has the advantages of low hardware deployment cost and wide application scenarios, making it a feasible solution for indoor positioning. In particular, with the development of wireless LAN technology and the popularization of Wi-Fi wireless routers, Wi-Fi indoor positioning technology has become more popular and feasible. At the same time, compared with simple measurement indicators such as traditional received signal strength (RSS), the channel state information (CSI) of Wi-Fi signals provides more fine-grained information, including detailed information on signal paths, reflections, multipath effects, and attenuation. By observing and analyzing the channel state information of Wi-Fi signals, we can achieve passive positioning of users in indoor environments, complete the extraction of user motion features, and ultimately achieve accurate indoor positioning. Therefore, Wi-Fi indoor positioning technology has significant advantages such as low cost, wide application scenarios, and high accuracy, making it an important part of modern positioning technology.

[0003] The Wi-Fi signal received by the receiver can be divided into two parts: static part and dynamic part. The static part refers to the signal directly transmitted between the transmitter and the receiver and the signal reflected after passing through static obstacles such as walls and furniture. The dynamic part refers to the signal reflected by the moving human body. These two signals will be superimposed together during the propagation process and recorded by the receiver. The general idea of ​​Wi-Fi passive positioning is to achieve passive positioning and tracking of users by separating the static and dynamic parts and extracting the location-related features in the dynamic signal. This method can observe signal propagation and changes in more detail, thereby improving the performance and reliability of the wireless communication system, and can also achieve accurate positioning and tracking in indoor environments.

[0004] Although the most advanced tracking systems can demonstrate relatively good tracking accuracy in complex experimental scenarios, they still have the following problems: (1) When the line-of-sight signal is blocked by obstacles, the signal transmission path will change significantly, and the CSI will fluctuate greatly, causing the simulated positioning effect to be seriously inaccurate compared to the actual path; (2) The existing tracking system does not consider the complexity of transmission in the signal propagation theory model, such as the existence of signal reflection, attenuation, and multipath effects. Therefore, the positioning data based on the traditional model has obvious noise in the non-line-of-sight environment, which not only greatly increases the processing difficulty, but also limits the room for improving accuracy. Summary of the invention

[0005] The technical problem to be solved by the present invention is how to construct model training data that is more in line with actual conditions.

[0006] The present invention solves the above technical problems through the following technical means:

[0007] The method for constructing training data of a Wi-Fi passive tracking target model in a non-line-of-sight scenario includes the following steps:

[0008] Steps of establishing the environmental coordinate system and dividing the mesh: determine the coordinates of the tracking area and the transceiver location according to the real environment, and establish the environmental coordinate system and divide the environmental mesh, which will serve as the basis for generating environmental parameters for simulation data;

[0009] The steps of meshing the simulation environment are as follows: With the transmitter location as the origin, simulated signal flooding is performed in all directions according to the set angle step; when the signal encounters obstacles and walls, it will bounce back in a full reflection manner and continue to propagate until the signal life cycle ends or the number of reflections exceeds the threshold;

[0010] Steps for updating signal propagation time: For the signal beam processed at a certain moment, consider its propagation time to the current map grid. If the time is less than the shortest propagation time recorded in the current grid, update it, otherwise do not update it; repeat this process until all signal beams have been propagated.

[0011] Steps of generating a heat map of the reflected signal path length: referring to the steps of meshing the simulation environment and updating the signal propagation time, the signal flooding is also performed with the receiver as the origin, so that the signal propagation time matrix corresponding to each transceiver reaching each point in the indoor environment can be obtained, and the elements of the matrix are correspondingly superimposed and multiplied by the signal propagation speed to obtain the heat map of the reflected signal path length corresponding to the pair of transceivers;

[0012] Steps of user trajectory simulation: In a simulation environment, user trajectory simulation is performed based on the acceleration-velocity-position sequence as a simulation principle to obtain a simulation trajectory that can cover the entire tracking area;

[0013] The steps of PLCR sequence and data set generation are as follows: the reflection signal path length heat map is used to obtain the reflection signal length sequence corresponding to each simulation trajectory, and then the PLCR sequence is obtained;

[0014] The combination of the simulation trajectory and the PLCR sequence is used as target model training data.

[0015] Furthermore, in the steps of establishing the environment coordinate system and dividing the grid, the tracking area and the transceiver coordinate position are determined according to the real environment; assuming that the size of the tracking area is L x ×L y , a two-dimensional rectangular coordinate system is established in the tracking area, with the lower left corner as the origin, the right as the positive direction of the x-axis, and the upward as the positive direction of the y-axis. The transmitter is deployed at P T (x T ,y T ) and two receivers are deployed at P R2 (x R1 ,y R1 ) and P R2 (x R2 ,y R2 ) and set the grid size to δ x ×δ y , the tracking area can be divided into OK, A grid matrix of columns; where Represents the smallest integer not less than x.

[0016] Furthermore, in the step of flooding the transmitter signal, the angle step of the signal flooding is set to Δθ, and the angle range to be traversed is 0° to 360°; assuming that the signal is reflected in the direction θ after the i-th total reflection after encountering an obstacle i Propagation, the propagation distance is d i , then the cumulative propagation path of the signal after i total reflections can be expressed as:

[0017]

[0018] Where d0 represents the distance from the transmitter to the first obstacle. The signal propagation will continue until the signal life cycle ends or the number of reflections exceeds the set threshold n. max .

[0019] Furthermore, in the step of updating the signal propagation time, the shortest propagation time t of the signal from the transmitter or receiver to the current grid is recorded. min ; For the signal beam processed at a certain moment, let the propagation speed of the signal beam be v, then the propagation time t i It can be expressed as:

[0020]

[0021] If i <t min , then update t min =t i , and this process is repeated until all signal beams have been propagated.

[0022] Furthermore, in the step of generating the reflected signal path length heat map, the signal flooding is also performed on the two receivers with reference to the steps of transmitter signal flooding and signal propagation time updating. So far, for each grid, a signal propagation time matrix T corresponding to a transmitter and two receivers has been obtained. T , T R1 and T R2 , add the signal propagation time matrix of the transmitter to the signal propagation time matrix of the two receivers respectively, and multiply it by the signal propagation speed v to obtain the heat map H of the reflected signal path length corresponding to the two pairs of transceivers:

[0023] H1=v·(T T +T R1 )

[0024] H2=v·(T T +T R2 ).

[0025] Furthermore, in the user trajectory simulation step, for each user in the simulation environment, there is an initial position (x0, y0), an initial velocity (v x0 , v y0 ) and initial acceleration (a x0 , a y0 ), for each simulation time step Δt, the position and velocity are updated using the following formula:

[0026] v xt =v x(t-1) +a xt ·Δt

[0027] v yt =v y(t-1) +a yt ·Δt

[0028] x t =x t-1 +v xt ·Δt

[0029] y t =y t-1 +v yt ·Δt

[0030] Repeat the entire simulation process to obtain a simulation trajectory that can cover the entire tracking area above the set bar.

[0031] Furthermore, in the step of generating the PLCR sequence and data set, after obtaining the user trajectory, the reflection signal path length heat map is used to calculate the reflection signal length sequence corresponding to each user trajectory; along each position of the user trajectory, the shortest arrival length corresponding to the current grid is found, that is, the required reflection signal length, and the reflection signal length sequence can be obtained after traversing each position of the user trajectory; the PLCR sequence generation is completed using the difference method, the reflection signal length sequence is differenced, and the PLCR is obtained by dividing it by the time interval δt; for all the transceiver links preset in the simulation environment, the position sequence corresponding to each simulation process is obtained. PLCR sequence Where N is the total number of simulation trajectories, and L is the sequence length of a single trajectory.

[0032] The present invention also provides a Wi-Fi passive tracking target model training method for non-line-of-sight scenarios.

[0033] The position sequence PLCR sequence As the training data of the long short-term memory network, the feature set and label set are defined as follows:

[0034] Set feature =[rk(i,j),P init (i)],Set label =[P(i, j)]

[0035] The feature set contains the PLCR sequences r of different links k (i, j), k∈[0, 1], the feature set also contains the two-dimensional coordinates P corresponding to each trajectory init (i); the label set contains the predicted target, i.e., the position sequence P (i, j); the feature set and label set are respectively used as inputs of the LSTM neural network to perform network model training, and finally obtain the target model.

[0036] The present invention also provides an application of a Wi-Fi passive tracking target model in a non-line-of-sight scenario.

[0037] Furthermore, the input of the target model is a PLCR sequence, and the PLCR sequence is obtained as follows:

[0038] Configure a transmitter and two receivers to receive CSI data, and use the inter-antenna conjugate multiplication method to remove noise and highlight dynamic signals:

[0039]

[0040] Among them, H i (f, t) represents the channel state information observed by the ith antenna at frequency f and time t, represents the conjugate CSI observed by the jth antenna; by performing short-time Fourier transform on C(f, t), the PLCR sequences corresponding to the two links can be obtained.

[0041] The two sets of PLCR sequences obtained above are combined with the starting point coordinates corresponding to the real path as the data of the target model, and the positioning trajectory map is output.

[0042] The advantages of the present invention are:

[0043] In the training phase, this method uses the indoor floor plan as the basis and integrates the reflection propagation model of Wi-Fi signals to generate the corresponding simulation trajectory data set. Compared with traditional methods, this method is more in line with the actual indoor layout and improves the accuracy of passive tracking in non-line-of-sight environments on the basis of limiting the user's feasible range.

[0044] The distribution of indoor signal reflection path length is revealed theoretically, making the signal propagation model more in line with the actual situation. On the basis of obtaining the reflection signal path length heat map, the inaccuracy problem of the original model in non-line-of-sight scenarios is effectively corrected, thereby improving the accuracy of non-line-of-sight positioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a logical block diagram of the training data construction process, training process and application in embodiments 1, 2 and 3 of the present invention;

[0046] Figure 2 Schematic diagram of reflected signal flooding in embodiment 1 of the present invention;

[0047] Figure 3 Heat map of the reflected signal path length under real conditions in Example 3 of the present invention.

[0048] Figure 4 A diagram showing the passive tracking effect of the method in Example 3 of the present invention. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] Terminology explanation:

[0051] Wi-Fi Passive Tracking

[0052] Wi-Fi passive tracking is a technology that tracks user locations based on Wi-Fi reflected signals. Unlike active tracking with the help of mobile terminals, passive tracking does not require users to carry any electronic devices. Instead, it captures and analyzes Wi-Fi signals reflected by the human body, detects changes in the phase and amplitude of the signal, and then uses relevant algorithms to calculate and complete positioning and tracking.

[0053] Channel State Information

[0054] Channel state information (CSI) is the channel property of a communication link. In wireless communication systems, the sender and receiver feedback the channel state information, namely CSI, based on the received signal. This includes information such as the channel gain, phase, and amplitude, which is used to reflect the impact of attenuation factors such as distance on the signal. When using wireless signals for user tracking, CSI can be used to infer the path and reflection of the signal during transmission.

[0055] LSTM Neural Network

[0056] Long Short Term Memory (LSTM) is a recursive neural network that circulates in time. In traditional recursive neural networks (RNNs), each output is determined by the current input and previous information, and the information it stores is usually short-term information, which determines that RNNs only have short-term memory but not long-term memory. LSTM is different from RNNs. Its internal repetitive modules have a multi-layer structure and a special interaction method, so that it can learn long-distance dependencies, effectively transmit and express information from a long time ago, and avoid the situation where effective information is ignored in long-term storage. Because it performs well on many problems, it is now widely used.

[0057] Reflection path length change rate

[0058] The path length changing rate (PLCR) refers to the rate of change of the path length of a signal after it is reflected by the human body. With the help of the Fresnel zone theory, we can determine the number of Fresnel zones that a link crosses by PLCR, and then determine its normal velocity. By combining the normal velocities of multiple links, we can finally get the user's true velocity and direction, thereby achieving tracking and positioning.

[0059] Non-Line-of-Sight

[0060] Non Line of Sight (NLOS) transmission means that in the presence of obstacles, wireless signals cannot be directly transmitted between the transmitter and the receiver as in the Line of Sight (LOS) model, but can only reach the receiver through reflection, scattering, etc. Compared with LOS, occlusion not only causes signal attenuation, but also reflection, diffraction and penetration loss. The occlusion of different objects has different effects, resulting in inaccuracy of the original passive positioning model and higher processing difficulty.

[0061] Example 1

[0062] This embodiment describes a method for constructing Wi-Fi passive tracking target model training data in non-line-of-sight scenarios, such as Figure 1 As shown, the following steps are included:

[0063] Step 1: Determine the tracking area and transceiver location coordinates according to the real environment, and establish the environmental coordinate system and divide the environmental grid, which will serve as the basis for generating environmental parameters for simulation data;

[0064] Step 2: With the transmitter location as the origin, simulate signal flooding in all directions according to the set angle step; when the signal encounters obstacles and walls, it will bounce back in a full reflection manner and continue to propagate until the signal life cycle ends or the number of reflections exceeds the threshold;

[0065] Step 3: For the signal beam processed at a certain moment, consider its propagation time to the current map grid. If the time is less than the shortest propagation time recorded in the current grid, update it; otherwise, do not update it. Repeat this process until all signal beams have been propagated.

[0066] Step 4: Referring to the steps of meshing the simulation environment and updating the signal propagation time, the signal flooding is also performed with the receiver as the origin. Thus, the signal propagation time matrix corresponding to each transceiver reaching each point in the indoor environment can be obtained. The elements of the matrix are correspondingly superimposed and multiplied by the signal propagation speed to obtain the heat map of the reflected signal path length corresponding to the pair of transceivers;

[0067] Step 5: In the simulation environment, the user trajectory is simulated based on the acceleration-velocity-position sequence as the simulation principle to obtain a simulation trajectory that can cover the entire tracking area;

[0068] Step 6: Use the reflection signal path length heat map to obtain the reflection signal length sequence corresponding to each simulation trajectory, and then obtain the PLCR sequence;

[0069] The combination of the simulation trajectory and the PLCR sequence is used as target model training data.

[0070] The specific process of establishing the environment coordinate system and dividing the grid in step 1 is as follows: determine the tracking area and transceiver coordinate position according to the real environment; assume that the size of the tracking area is L x ×L y , a two-dimensional rectangular coordinate system is established in the tracking area, with the lower left corner as the origin, the right as the positive direction of the x-axis, and the upward as the positive direction of the y-axis. The transmitter is deployed at P T (x T ,y T ) and two receivers are deployed at P R2 (X R1 ,y R1 ) and P R2 (x R2 ,y R2 ) and set the grid size to δ x ×δ y , the tracking area can be divided into OK, A grid matrix of columns; where Represents the smallest integer not less than x.

[0071] The specific process of transmitter signal flooding in step 2 is as follows: set the angle step of signal flooding traversal to Δθ, and the angle range to be traversed is 0° to 360°; assume that the signal is reflected in the direction θ after the i-th total reflection after encountering an obstacle i Propagation, the propagation distance is d i , then the cumulative propagation path of the signal after i total reflections can be expressed as:

[0072]

[0073] Where d0 represents the distance from the transmitter to the first obstacle. The signal propagation will continue until the signal life cycle ends or the number of reflections exceeds the set threshold n. max .

[0074] The specific process of updating the signal propagation time in step 3 is: record the shortest propagation time t of the signal from the transmitter or receiver to the current grid min ; For the signal beam processed at a certain moment, let the propagation speed of the signal beam be v, then the propagation time t i It can be expressed as:

[0075]

[0076] If i <t min , then update t min =t i , and this process is repeated until all signal beams have been propagated.

[0077] The specific process of generating the reflected signal path length heat map in step 4 is as follows: Referring to steps 2 and 3, the signal flooding is also performed on the two receivers; so far, for each grid, the signal propagation time matrix T corresponding to a transmitter and two receivers has been obtained. T , T R1 and T R2 , add the signal propagation time matrix of the transmitter to the signal propagation time matrix of the two receivers respectively, and multiply it by the signal propagation speed v to obtain the heat map H of the reflected signal path length corresponding to the two pairs of transceivers:

[0078] H1=v·(T T +T R1 )

[0079] H2=v·(T T +T R2 ).

[0080] The specific process of user trajectory simulation in step 5 is as follows: for each user in the simulation environment, there is an initial position (x0, y0), initial velocity (v x0 , v y0 ) and initial acceleration (a x0 , a y0 ), for each simulation time step Δt, the position and velocity are updated using the following formula:

[0081] v xt =v x(t-1) +a xt ·Δt

[0082] v yt =v y(t-1) +a yt ·Δt

[0083] x t =x t-1 +v xt ·Δt

[0084] y t =y t-1 +v yt ·Δt

[0085] The entire simulation process is repeated to obtain more than 100,000 simulation trajectories that can cover the entire tracking area.

[0086] The specific process of generating the PLCR sequence and data set in step 6 is as follows: after obtaining the user trajectory, use the reflection signal path length heat map to calculate the reflection signal length sequence corresponding to each user trajectory; along each position of the user trajectory, find the shortest arrival length corresponding to the current grid, that is, the required reflection signal length, and traverse each position of the user trajectory to obtain the reflection signal length sequence; use the difference method to complete the PLCR sequence generation, make a difference on the reflection signal length sequence, and divide it by the time interval δt to obtain the PLCR; for all the transceiver links preset in the simulation environment, obtain the position sequence corresponding to each simulation process PLCR sequence Where N is the total number of simulation trajectories, and L is the sequence length of a single trajectory. PLCR sequence As training data for long short-term memory networks.

[0087] Example 2

[0088] This embodiment trains the Wi-Fi passive tracking target model in a non-line-of-sight scenario based on the training data constructed in Embodiment 1.

[0089] First, the network constructed contains an input layer, an LSTM layer, a fully connected layer, and an output layer. The LSTM layer contains 64 neurons and outputs the output sequence of all time steps. The Adam optimization algorithm is used for network training, and the initial learning rate is set to 0.03, the batch size is 512, and the maximum number of iterations is 8. Experiments show that the model can achieve convergence in the second iteration process.

[0090] Based on the training data obtained in Example 1, the feature set and label set are defined as follows:

[0091] Set feature =[r k (i, j), P init (i)],Set label =[P(i, j)]

[0092] The feature set contains the PLCR sequences r of different links k (i, j), here we use the experimental configuration of two links, so k∈[0,1], and the feature set also contains the two-dimensional coordinates P corresponding to each trajectory init (i), so Set feature The dimension of is 10w×100×4; the label set contains our prediction target, i.e., the position sequence P(i, j), and the data dimension is 10w×100×2. The above two sets are used as the input and output of the LSTM neural network respectively to train the network model.

[0093] Example 3

[0094] Based on the target model of Example 2, this example records the input data processing process in the target model application process, which is as follows: Application of Wi-Fi passive tracking target model in non-line-of-sight scenarios.

[0095] like Figure 2 , Figure 3 , Figure 4 As shown in the figure, in this stage, one transmitter and two receivers are deployed on site according to the established configuration, and the three transceivers are remotely controlled through the SSH protocol. Among them, the transmitter is turned on the transmission (iniect) mode, the packet rate is 1000 times / s, and its signal transmission is controlled by the command line; the two receivers are configured in the monitoring (monitor) mode at the same time to ensure that they can detect the CSI data packets of the transmitted signal at the same time. For the received CSI data, it is necessary to remove noise and highlight dynamic signals first. For this purpose, we use the method of conjugate multiplication between antennas:

[0096]

[0097] Among them, H i (f, t) represents the channel state information observed by the ith antenna at frequency f and time t, represents the conjugate CSI observed by the second antenna (reference antenna). The conjugate multiplication operation can remove the phase offset that changes with time, thereby weakening the static signal and highlighting the dynamic signal. By performing a short-time Fourier transform on C(f, t), the PLCR sequences corresponding to the two links can be obtained.

[0098] The two sets of PLCR sequences obtained above are combined with the starting point coordinates (known) corresponding to the real path to form a vector with a dimension of 100×4, which is used as the test set Set in the positioning test phase. test , input it into the pre-trained positioning network model, and we can get Figure 4 The predicted positioning trajectory is shown.

[0099] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing Wi-Fi passive tracking target model training data in non-line-of-sight scenarios, characterized in that: The following steps are involved: Steps of establishing the environmental coordinate system and dividing the mesh: determine the coordinates of the tracking area and the transceiver location according to the real environment, and establish the environmental coordinate system and divide the environmental mesh, which will serve as the basis for generating environmental parameters for simulation data; The steps of meshing the simulation environment are as follows: With the transmitter location as the origin, simulated signal flooding is performed in all directions according to the set angle step; when the signal encounters obstacles and walls, it will bounce back in a full reflection manner and continue to propagate until the signal life cycle ends or the number of reflections exceeds the threshold; Steps for updating signal propagation time: For the signal beam processed at a certain moment, consider its propagation time to the current map grid. If the time is less than the shortest propagation time recorded in the current grid, update it, otherwise do not update it; repeat this process until all signal beams have been propagated. Steps of generating a heat map of the reflected signal path length: referring to the steps of meshing the simulation environment and updating the signal propagation time, the signal flooding is also performed with the receiver as the origin, so that the signal propagation time matrix corresponding to each transceiver reaching each point in the indoor environment can be obtained, and the elements of the matrix are correspondingly superimposed and multiplied by the signal propagation speed to obtain the heat map of the reflected signal path length corresponding to the pair of transceivers; Steps of user trajectory simulation: In a simulation environment, user trajectory simulation is performed based on the acceleration-velocity-position sequence as a simulation principle to obtain a simulation trajectory that can cover the entire tracking area; The steps of PLCR sequence and data set generation are as follows: the reflection signal path length heat map is used to obtain the reflection signal length sequence corresponding to each simulation trajectory, and then the PLCR sequence is obtained; The combination of the simulation trajectory and the PLCR sequence is used as target model training data.

2. The method for constructing Wi-Fi passive tracking target model training data for non-line-of-sight scenarios according to claim 1, characterized in that: In the steps of establishing the environment coordinate system and dividing the grid, the tracking area and the transceiver coordinate position are determined according to the real environment; assuming that the size of the tracking area is L x ×L y , a two-dimensional rectangular coordinate system is established in the tracking area, with the lower left corner as the origin, the right as the positive direction of the x-axis, and the upward as the positive direction of the y-axis. The transmitter is deployed at P T (x T ,y T ) and two receivers are deployed at P R2 (x R1 ,y R1 ) and P R2 (x R2 ,y R2 ) and set the grid size to δ x ×δ y , the tracking area can be divided into OK, A grid matrix of columns; where Represents the smallest integer not less than x.

3. The method for constructing Wi-Fi passive tracking target model training data for non-line-of-sight scenarios according to claim 1, characterized in that: In the step of flooding the transmitter signal, the angle step of the signal flooding is set to Δθ, and the angle range to be traversed is 0° to 360°; assuming that the signal is reflected in the direction θ after the i-th total reflection after encountering an obstacle i Propagation, the propagation distance is d i , then the cumulative propagation path of the signal after i total reflections can be expressed as: Where d0 represents the distance from the transmitter to the first obstacle. The signal propagation will continue until the signal life cycle ends or the number of reflections exceeds the set threshold n. max .

4. The method for constructing Wi-Fi passive tracking target model training data for non-line-of-sight scenarios according to claim 1, characterized in that: In the step of updating the signal propagation time, the shortest propagation time t of the signal from the transmitter or receiver to the current grid is recorded. min ; For the signal beam processed at a certain moment, let the propagation speed of the signal beam be v, then the propagation time t i It can be expressed as: If i <t min , then update t min =t i , and this process is repeated until all signal beams have been propagated.

5. The method for constructing Wi-Fi passive tracking target model training data for non-line-of-sight scenarios according to claim 1, characterized in that: In the step of generating the reflected signal path length heat map, the signal flooding step and the signal propagation time updating step of the transmitter are referred to, and the signal flooding is also performed on the two receivers; so far, for each grid, a signal propagation time matrix T corresponding to a transmitter and two receivers has been obtained. T 、T R1 and T R2 , add the signal propagation time matrix of the transmitter to the signal propagation time matrix of the two receivers respectively, and multiply it by the signal propagation speed v to obtain the heat map H of the reflected signal path length corresponding to the two pairs of transceivers: H1=v·(T T +T R1 ) H2=v·(T T +T R2 )。 6. The method for constructing Wi-Fi passive tracking target model training data for non-line-of-sight scenarios according to any one of claims 1 to 5, characterized in that: In the user trajectory simulation step, each user in the simulation environment has an initial position (x0, y0), an initial velocity (v x0 ,v y0 ) and initial acceleration (a x0 ,a y0 ), for each simulation time step Δt, the position and velocity are updated using the following formula: v xt =v x(t-1) +a xt ·Δt v yt =v y(t-1) +a yt ·Δt x t =x t-1 +v xt ·Δt y t =y t-1 +v yt ·Δt Repeat the entire simulation process to obtain a simulation trajectory that can cover the entire tracking area above the set bar.

7. The method for constructing Wi-Fi passive tracking target model training data in non-line-of-sight scenarios according to claim 6, characterized in that: In the step of generating the PLCR sequence and data set, after obtaining the user trajectory, the reflection signal path length heat map is used to calculate the reflection signal length sequence corresponding to each user trajectory; along each position of the user trajectory, the shortest arrival length corresponding to the current grid is found, that is, the required reflection signal length, and the reflection signal length sequence can be obtained after traversing each position of the user trajectory; the PLCR sequence generation is completed using the difference method, the reflection signal length sequence is differenced, and the PLCR is obtained by dividing it by the time interval δt; for all the transceiver links preset in the simulation environment, the position sequence corresponding to each simulation process is obtained PLCR sequence Where N is the total number of simulation trajectories, and L is the sequence length of a single trajectory.

8. A Wi-Fi passive tracking target model training method for non-line-of-sight scenarios, characterized in that: The position sequence PLCR sequence As the training data of the long short-term memory network, the feature set and label set are defined as follows: Set feature =[r k (i,j),P init (i)],Set label =[P(i,j)] The feature set contains the PLCR sequences r of different links k (i, j), k∈[0,1], the feature set also contains the two-dimensional coordinates P corresponding to each trajectory init (i); the label set contains the predicted target, i.e., the position sequence P(i, j); the feature set and label set are respectively used as inputs of the LSTM neural network to perform network model training, and finally obtain the target model.

9. Application of Wi-Fi passive tracking target model in non-line-of-sight scenarios.

10. The application of the Wi-Fi passive tracking target model in non-line-of-sight scenarios according to claim 9, characterized in that: The input of the target model is a PLCR sequence, and the PLCR sequence is obtained as follows: Configure a transmitter and two receivers to receive CSI data, and use the inter-antenna conjugate multiplication method to remove noise and highlight dynamic signals: Among them, H i (f, t) represents the channel state information observed by the ith antenna at frequency f and time t, represents the conjugate CSI observed by the jth antenna; by performing a short-time Fourier transform operation on C(f, t), the PLCR sequences corresponding to the two links can be obtained. The two sets of PLCR sequences obtained above are combined with the starting point coordinates corresponding to the real path as the data of the target model, and the positioning trajectory map is output.

Citation Information

Cited By

  • Self-adaption Wi-Fi tracking method based on environmental obstacle constraint

    CN119835756A