A feature compensation and passive localization method and system

Abstract

This invention provides a feature compensation and passive positioning method and system. The method includes: performing signal feature processing operations; extracting WiFi signal features by analyzing channel state information to obtain a PLCR matrix; processing DFS data based on the reflection path change rate (PLCR); deriving an observation-based matrix P1 and a reliability-based matrix R from the PLCR matrix; assigning weights to each type of prediction for combined prediction operations; calculating each type of prediction; combining the weights to obtain the final PLCR prediction; processing to obtain applicable PLCR prediction values ​​and user trajectories; determining the user's speed; and predicting the wireless signal features at the next moment to obtain the complete trajectory. This invention solves the technical problem of low accuracy in WiFi passive positioning operations under non-continuous communication scenarios due to insufficient utilization of the correlation between WiFi links and the loss of wireless signal features in real-world application scenarios.

Description

Technical Field

[0001] This invention relates to the field of indoor WiFi passive positioning under wireless sensing technology, specifically to a feature compensation and passive positioning method and system. Background Technology

[0002] In recent years, indoor tracking and positioning have attracted widespread interest from researchers. Existing tracking methods can generally be divided into two categories: active tracking and passive tracking. Active tracking methods mainly rely on received signal strength indication; although widely used, they are difficult to achieve high accuracy, while methods utilizing channel state information can provide more accurate tracking. Passive tracking methods do not require users to carry equipment, which is an advantage, but they usually require continuous link communication, limiting their application.

[0003] Because common mobile devices (such as mobile phones, tablets, and laptops) can easily function as WiFi receivers, providing Received Signal Strength Indicators (RSSI), RSSI has become a widely used feature for tracking. For example, the existing invention patent application CN114757237B, entitled "A Speed-Independent Gait Recognition Method Based on WiFi Signals," describes a method that includes: extracting the time-varying CSI amplitude of WiFi signals during a person's walking process; preprocessing the CSI amplitude; determining whether someone is walking in the environment and extracting walking activity segments; converting the walking activity segments into time-frequency maps of the same size; building a DANN-based speed-independent gait recognition model, which includes a feature extractor, an identity recognizer, and a speed recognizer. The feature extractor extracts latent features from the input time-frequency map, the identity recognizer uses the features extracted by the feature extractor to predict the identity of the target, and the speed recognizer uses the features extracted by the feature extractor to predict the speed of the target; training the speed-independent gait recognition model and outputting the identity of the target. And the existing invention patent application document CN117197888A, entitled "IMU-WiFi-based Hot Deployment Cross-Mode Gait Recognition System and Method," includes: a gait feature extraction unit, which can calculate the attitude based on the raw IMU data obtained from the inertial measurement unit and combined with extended Kalman filtering to obtain the attitude that compensates for drift error, perform processing to reduce ankle IMU drift and suppress waist IMU drift to obtain footprint and trunk velocity curves respectively, and extract IMU-based gait feature vectors from the footprint and trunk velocity curves and save them to a feature database; a gait recognition unit, which can receive CSI data of WiFi signal and eliminate high-frequency noise of CSI data through adaptive PCA method, generate a spectrum map from CSI data and extract WiFi-based gait feature vectors from it, and compare the WiFi-based gait feature vectors with the IMU-based gait feature vectors through a pre-trained classification neural network model, and identify the pedestrian's identity based on the comparison result. However, multipath effects and asynchrony between devices make high-precision RSSI tracking difficult to achieve. Because CSI provides more information through multiple antennas and carriers, it can now be used to achieve relatively accurate active localization and tracking. Existing research has utilized multiple-input multiple-output (MIMO) techniques to construct antenna arrays to analyze signal angles of arrival and achieve mobile device tracking. Inspired by RSSI indoor positioning, some studies have shown that CSI fingerprints can eliminate the effects of multipath propagation. Developing schemes to automatically update CSI fingerprint databases can improve positioning efficiency without requiring on-site fingerprint collection. Although active localization technologies have made progress in indoor tracking, they all require users to carry electronic devices to function, which undoubtedly reduces user willingness to use them.

[0004] Compared to active tracking, passive tracking offers significant advantages because it extracts and analyzes signal features without requiring the user to carry any equipment. The Widar system, proposed in 2017, was the first to infer user speed using the path length variation rate of multiple WiFi links. Widar2.0 and md-Track, proposed in 2018 and 2019 respectively, further combined angle of arrival, time of flight, and Doppler shift to achieve single-link tracking. Additionally, work has implemented methods to eliminate tracking limitations when users cross WiFi links, using smart microphones for tracking and gait recognition. The newly proposed NNE-Tracking in 2024 invented a wireless sensing architecture that can be trained by generating large-scale datasets while utilizing mathematical models to supervise the training process, effectively combating environmental noise. Although line-of-sight wireless positioning has seen significant improvements, the aforementioned idealized models often perform poorly when dealing with significant environmental obstructions. To overcome these limitations, researchers have also begun exploring methods for tracking non-line-of-sight paths. The NLoc proposed in 2022 models obstructed reflections and virtual direct signals to achieve non-line-of-sight positioning. HyperTracking, proposed in 2024, developed a non-line-of-sight tracking method that combines spatial model features and neural networks to eliminate the interference of obstacles on localization.

[0005] With the development of IoT technology, numerous smart devices are commonly found in everyday environments. Multiple transceiver links allow for the acquisition of rich user motion information, but communication within each link is not continuous. The aforementioned passive tracking systems cannot track users with missing link information, and they all require continuous link information. Communication Duty Cycles (CDC) is used to measure the degree of missing wireless signal characteristics; it refers to the proportion of effective communication data packets available for sensing. Existing work is based on the assumption that CDC is 100%.

[0006] In summary, existing technologies suffer from several technical problems, including underutilization of the correlation between signals between WiFi links and loss of wireless signal characteristics in real-world application scenarios, resulting in low accuracy of passive WiFi positioning operations in non-continuous communication situations. Summary of the Invention

[0007] The technical problem to be solved by this invention is: how to solve the technical problem that the existing technology does not make full use of the correlation between signals between WiFi links and the loss of wireless signal characteristics in real application scenarios, resulting in low accuracy of WiFi passive positioning operation in non-continuous communication scenarios.

[0008] The present invention solves the above-mentioned technical problems by adopting the following technical solution: a feature compensation and passive positioning method comprising:

[0009] S1. Perform signal feature processing operations. By analyzing channel state information, extract WiFi signal features to obtain the PLCR matrix. Collect and extract the reflection path change rate (PLCR) from the CSI data, and process it to obtain DFS data. Derive the observation-based matrix P1 and the reliability-based matrix R from the PLCR matrix. Assign weights to the first type prediction, the second type prediction, and the third type prediction for combined prediction operations.

[0010] S2. Prepare for PLCR prediction in the tracking phase by obtaining the first type of prediction, the second type of prediction and the third type of prediction, combining them with weights to obtain the final PLCR prediction, and processing the final PLCR prediction to obtain the applicable PLCR prediction value and user trajectory.

[0011] S3. Determine the user's speed and predict the wireless signal characteristics at the next moment. Design and utilize a neural network to map the user's trajectory based on the reflection path change rate (PLCR). Obtain the complete trajectory by performing tracking, prediction, algorithm adjustment, and local trajectory prediction optimization operations.

[0012] This invention utilizes WiFi multi-link communication to achieve accurate passive tracking of the human body in indoor environments where wireless signal characteristics are severely lacking. The goal of this invention is to compensate for the missing wireless signal characteristics in real-world application scenarios by mining and utilizing the correlation between signals from multiple WiFi links, thereby achieving precise WiFi passive positioning in non-continuous communication situations.

[0013] In a more specific technical solution, S1 includes:

[0014] S11. Acquire CSI data. By performing a Short-Time Fourier Transform (STFT), extract the reflection path change rate (PLCR) from the raw CSI readings. The CSI function with respect to frequency f and time t can be expressed as follows:

[0015]

[0016] In the formula, H s (f,t) represents the static CSI component, and L(t) represents the dynamic CSI component H. d The path length corresponding to (f,t), where λ is the wavelength, and A(f,t) is the signal amplitude, which is e -j2πL(t) / λ For phase;

[0017] S12. Based on the Doppler effect, the following formula is derived:

[0018]

[0019] In the formula, f D represents DFS data, r represents the reflection path change rate PLCR, and L(t) is the dynamic path length at time t;

[0020] S13. Check each element in the PLCR matrix. When a missing value is detected, scan upward from the position of the missing value until the nearest observed true value in the same link is encountered, and fill the nearest observed true value into the preset observation matrix to obtain the matrix P1 based on the observed values;

[0021] S14. In the reliability-based matrix R, when the actual reflection path change rate PLCR can be observed at a specific position, the weight assigned to the first type of prediction is 1; when the signal characteristics are continuously lost, the weight of the first type of prediction is reduced; when a link experiences continuous feature loss, a quadratic function is used to reduce the weight of the first type of prediction, and weights are assigned to the second type of prediction and the third type of prediction.

[0022] In a more specific technical solution, in S14, the following logic is used to reduce the weight of the first type of prediction:

[0023]

[0024] In a more specific technical solution, S2 includes:

[0025] S21. Calculate the prediction based on the observed data as the first type of prediction. Denote the t-th element of the n-th link in the matrix P1 based on the observed values as P1(t, n), and obtain the prediction based on the observed data:

[0026]

[0027] In the formula, t ′ is the smallest time index such that P(t ′ , n)≠0 and 1≤t’<t.

[0028] S22. Calculate the prediction based on the proportional relationship as the second type of prediction. Obtain and use the non-missing values to calculate the prediction based on the proportional relationship. When all the reflection path change rates PLCR at t = t2 are missing, jump to use the prediction of the mathematical model;

[0029] S23. Calculate the prediction based on the mathematical model as the third type of prediction. Obtain the PLCR prediction matrix P3 based on the mathematical model through mathematical modeling, and inversely deduce the reflection path change rate PLCR for the first t time slots therefrom;

[0030] S24. Perform a weighted operation on the first type of prediction, the second type of prediction, and the third type of prediction to obtain the final PLCR prediction, and update the elements of the PLCR prediction matrix.

[0031] S25. Use the final PLCR prediction to fill in the missing features and obtain the user trajectory.

[0032] This invention expands the application scenarios of traditional passive positioning systems, relaxes the unrealistic requirements of traditional sensing technologies for long-term continuous communication, and further realizes the integration of communication and sensing in practical applications of WiFi-based wireless sensing systems.

[0033] In a more specific technical solution, in S22, the following logic is used to obtain a prediction based on the proportional relationship:

[0034]

[0035] In a more specific technical solution, S23 includes:

[0036] S231. For the pre-set link, let the transmitter position be: l t =(x t ,y t ); Let the receiver position be: l r =(x r ,y r The current person's position is: l h =(x h ,y h The person's speed is: v = (v x ,v y ) T ;

[0037] S232. Using the following logic, obtain the PLCR prediction matrix based on the mathematical model:

[0038] P3(t,n)=A×v=a x v x +a y v y .

[0039] S233. When all PLCR data of all links are lost during the preset time slot, PLCR prediction is performed based on the known location sequence.

[0040] In a more specific technical solution, the PLCR prediction matrix based on the mathematical model satisfies:

[0041]

[0042] In a more specific technical solution, in S24, each element of the PLCR prediction matrix P is updated using the following logic:

[0043]

[0044] In the formula, w represents the weight of the prediction based on the observation.

[0045] This invention proposes three mechanisms for effectively compensating for missing WiFi features in passive tracking scenarios. By combining these three signal feature prediction methods, this invention proposes an algorithm called simultaneous tracking and prediction. This algorithm achieves accurate passive tracking even in the presence of severe WiFi feature loss.

[0046] In a more specific technical solution, S3 includes:

[0047] S31. During the tracking process, the reflection path change rate (PLCR) is predicted in reverse by using local user trajectories.

[0048] S32. Compensate for missing PLCR features by continuously iterating through tracking and prediction operations.

[0049] S33. Continuously execute algorithm adjustment operations and local trajectory prediction optimization operations to obtain the complete trajectory.

[0050] In a more specific technical solution, a feature compensation and passive positioning system includes:

[0051] The signal feature processing module is used to perform signal feature processing operations. By analyzing channel state information, it extracts WiFi signal features to obtain the PLCR matrix; it collects and extracts the reflection path change rate (PLCR) from CSI data, processes it to obtain DFS data, and derives the observation-based matrix P1 and the reliability-based matrix R from the PLCR matrix. It assigns weights to the first type prediction, the second type prediction, and the third type prediction for combined prediction operations.

[0052] The prediction and combination module is used to prepare for PLCR prediction in the tracking stage. It obtains the first type of prediction, the second type of prediction and the third type of prediction, and uses weighted combination to obtain the final PLCR prediction. The final PLCR prediction is processed to obtain the applicable PLCR prediction value and user trajectory. The prediction and combination module is connected to the signal feature processing module.

[0053] The wireless signal feature prediction module is used to determine the user's speed and predict the wireless signal features at the next moment. It designs and utilizes a neural network to map the user's trajectory based on the reflection path change rate (PLCR). By performing tracking operations, prediction operations, algorithm adjustment operations, and local trajectory prediction optimization operations, the complete trajectory is obtained. The wireless signal feature prediction module is connected to the prediction combination module.

[0054] The present invention has the following advantages over the prior art:

[0055] This invention utilizes WiFi multi-link communication to achieve accurate passive tracking of the human body in indoor environments where wireless signal characteristics are severely lacking. The goal of this invention is to compensate for the missing wireless signal characteristics in real-world application scenarios by mining and utilizing the correlation between signals from multiple WiFi links, thereby achieving precise WiFi passive positioning in non-continuous communication situations.

[0056] This invention expands the application scenarios of traditional passive positioning systems, relaxes the unrealistic requirements of traditional sensing technologies for long-term continuous communication, and further realizes the integration of communication and sensing in practical applications of WiFi-based wireless sensing systems.

[0057] This invention proposes three mechanisms for effectively compensating for missing WiFi features in passive tracking scenarios. By combining these three signal feature prediction methods, this invention proposes an algorithm called simultaneous tracking and prediction. This algorithm achieves accurate passive tracking even in the presence of severe WiFi feature loss.

[0058] This invention solves the technical problems in the prior art, such as the underutilization of the correlation between signals between WiFi links and the lack of wireless signal characteristics in real application scenarios, which leads to low accuracy of WiFi passive positioning operations in non-continuous communication scenarios. Attached Figure Description

[0059] Figure 1 This is a schematic diagram of the basic steps of a feature compensation and passive positioning method according to Embodiment 1 of the present invention;

[0060] Figure 2 This is a schematic diagram of data flow processing for a feature compensation and passive positioning method according to Embodiment 1 of the present invention;

[0061] Figure 3 This is a schematic diagram of the Fresnel zone cutting for human movement in Embodiment 1 of the present invention;

[0062] Figure 4 This is a schematic diagram illustrating the calculation of the PLCR matrix, the observation-based matrix, and the reliability matrix in Embodiment 1 of the present invention;

[0063] Figure 5 This is a schematic diagram illustrating the use of an LSTM neural network to establish a mapping between wireless signal features and trajectories in Embodiment 1 of the present invention;

[0064] Figure 6 This is a diagram illustrating the passive positioning effect of Embodiment 1 of the present invention. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] Example 1

[0067] like Figure 1 and Figure 2 As shown, the feature compensation and passive localization method provided by the present invention includes the following basic steps:

[0068] S1. Perform signal feature processing operations;

[0069] like Figure 3 As shown, in this embodiment, WiFi signal features are extracted by analyzing channel state information to obtain the PLCR matrix; specifically, since the Fresnel zone ellipse is cut when the user moves, the receiver can calculate the PLCR from the received signal; the matrix based on the observation and the reliability matrix are calculated to assist the PLCR prediction in the next stage; the system alternates between tracking and prediction operations and repeats the cycle.

[0070] In this embodiment, the system first acquires the CSI; then, it extracts the PLCR from the raw CSI reading by performing a Short-Time Fourier Transform (STFT). In this embodiment, the wavelength is represented by λ, the signal amplitude by A(f,t), and the phase by e. -j2πL(t) / λ Then, the CSI function with respect to frequency f and time t can be expressed as follows:

[0071]

[0072] Among them, H s (f,t) represents the static CSI component, and L(t) represents the dynamic CSI component H. d The path length corresponding to (f,t). PLCR is mathematically defined as:

[0073]

[0074] Where L(t) is the dynamic path length at time t. Due to the Doppler effect, PLCR is a constant multiple of DFS; therefore, the following equation can be derived:

[0075]

[0076] Among them, f Dr and r represent DFS and PLCR, respectively.

[0077] In this embodiment, the PLCR matrix can be represented by, for example, a two-dimensional matrix P of size T×N, where T is the total number of time slots and N is the number of links. The row index t represents the time index, and the column index n represents the signal characteristics of different WiFi links, i.e., the signal characteristics received by different receivers. When there is no WiFi communication at the location corresponding to time t and link n, the system sets the corresponding value P(t,n) to 0. This matrix is ​​continuously updated in each iteration of the prediction phase.

[0078] In this embodiment, the observation-based matrix P1 is the same size as the PLCR matrix and can be directly derived from it. Specifically, this invention compensates for missing WiFi features based on the most recently observed true PLCR features within the same WiFi link. To calculate the observation matrix, each element in the original PLCR matrix needs to be examined. When a missing value is detected, the system scans upwards from that position until a non-missing PLCR value is encountered, i.e., the most recently observed true value within the same link. This value is filled into the observation matrix as an observation-based prediction. The elements in the observation matrix are referred to as observation-based predictions.

[0079] In this embodiment, the reliability matrix is ​​denoted by R and has the same size as the two matrices mentioned above. However, its elements are not PLCRs, but weights between 0 and 1. Although the system has obtained a basic prediction based on observations, it cannot achieve accurate tracking by relying solely on it. To improve the accuracy of the final prediction, this invention introduces two other types of predictions obtained through other methods, denoted as the second type prediction and the third type prediction, respectively.

[0080] In this embodiment, to effectively and accurately combine the three predictions, the system assigns weights to them and then calculates the final PLCR prediction value. Since the PLCR in a given link remains stable over short time intervals, the system tends to trust the observation-based prediction as long as the signal in a link is not missing for an extended period. To measure the degree of trust in the observation-based prediction, each element in the reliability matrix is ​​a weight assigned to prediction 1, representing the system's level of confidence in the observation-based prediction at the corresponding time slot. When an actual PLCR can be observed at a specific location, the weight assigned to the observation-based prediction is 1, because the observation is real and accurate. However, if signal features are continuously lost, the reliability of past PLCR observations decreases. Nevertheless, the most recent observable value for the same link remains meaningful. Based on the above analysis, the weight assigned to the observation-based prediction gradually decreases over time until after T... wIt decreases to 0 after a certain time index. When a link experiences continuous feature loss, a quadratic function is used in this embodiment to reduce this weight:

[0081]

[0082] where w represents the above-mentioned weight. This weight reduces to zero at time slot t = T w . According to the specific situation of signal feature loss, another part of the weight (1 - w) will be assigned to the prediction based on proportional relationship or the prediction based on the model. In the above process, the calculation processes of the PLCR matrix, the matrix based on observations, and the reliability matrix are as Figure 4 shown.

[0083] S2. Prepare for the PLCR prediction in the tracking stage, obtain and combine the first-type prediction, the second-type prediction, and the third-type prediction to obtain the final PLCR prediction;

[0084] In this embodiment, the first-type prediction, the second-type prediction, and the third-type prediction are combined. Specifically, the first-type prediction can be: the prediction based on observations; the second-type prediction can be: the prediction based on proportional relationship; the third-type prediction can be: the prediction based on the model;

[0085] In this embodiment, the prediction based on observations is calculated directly from a small number of actual PLCR values that can be directly observed; the prediction based on proportional relationship is determined by the correlation between different WiFi links; the system considers the degree of feature loss and combines the above predictions with the prediction based on the model with weights to obtain the final PLCR prediction.

[0086] In this embodiment, the prediction based on observed data is calculated. Specifically, the original PLCR matrix P stores the actual PLCR values that can be observed, and the positions of the missing data are marked with 0. For each position of the missing data, the system scans upward until it encounters a non-missing value. This value is the most recently observed PLCR value of the same link, and it is regarded as the prediction based on observed data for this position. Represent the t-th element of the n-th link in the observation matrix as P1(t, n), then the corresponding prediction based on observed data can be calculated as:

[0087]

[0088] where t ′ is the smallest time index such that P(t ′ , n) ≠ 0 and 1 ≤ t' < t.

[0089] In this embodiment, a prediction based on a proportional relationship is calculated. Specifically, in two given links, the ratio of the PLCR at two adjacent times t = t1 (the previous time) and t = t2 (the next time) is approximately constant. Assuming the system is currently processing data corresponding to t = t2 in the current iteration, the system can confirm that the PLCR of each link at its previous time t = t1 has already been obtained in the previous iteration and is therefore not missing. Therefore, in link n1, as long as there is at least one non-missing PLCR at time t = t2, the system can fill in the missing data in link n1 using the proportional relationship. Representing the missing PLCR at t = t2 as P2(t2, n1), the system can use three non-missing values ​​to calculate a prediction based on the proportional relationship:

[0090]

[0091] Otherwise, if all PLCR data at t=t2 is missing, the system must resort to predictions based on mathematical models.

[0092] In this embodiment, a prediction based on a mathematical model is calculated. Specifically, the prediction matrix based on the mathematical model, derived from mathematical modeling, is represented as P3. In previous work, mathematical models have been used to obtain trajectories from PLCR features. However, this invention utilizes mathematical models from another perspective, namely, inferring the PLCR features corresponding to the trajectory through a model-based approach. For a specific link, assume the positions of the transmitter and receiver are l... t =(x t ,y t ) and l r =(x r ,y r The current person's position is l h =(x h ,y h The person's speed is v = (v x ,v y ) T Therefore, the model-based PLCR prediction can be calculated as follows:

[0093] P3(t,n)=A×v=a x v x +a y v y .

[0094] in,

[0095]

[0096] Specifically, if trajectory predictions for the first t time slots can be obtained through iteration, the system can derive velocity predictions for the first t time slots by approximate differencing of the position sequence. With this velocity prediction, the system can then reverse-engineer the PLCR for the first t time slots. In fact, the continuous nature of the PLCR allows the PLCR at time (t+1) to be approximately fitted using the value at time t. Therefore, even if PLCR data for all links is lost at a specific time slot, this system can still perform PLCR predictions based on the previous position sequence.

[0097] In this embodiment, after calculating the three predictions, the three predictions are combined using the following weighting method. Assume that in the current iteration, the first t rows have been processed, and the (t+1)th row is being processed. Then, for each value in the (t+1)th row: If the value is not missing, the system can simply use the first type prediction, i.e., the observed value, without additional processing; if the value is missing, and there is at least one observable non-missing PLCR value in the (t+1)th row, the first type prediction and the second type prediction are combined; if the value is missing, and all values ​​in the (t+1)th row are missing, the first type prediction and the third type prediction are combined.

[0098] After obtaining the three predictions, each element of the final PLCR prediction matrix P can be updated as follows:

[0099]

[0100] Here, w represents the weight of the prediction based on the observations. The system then fills in the missing features using the weighted final PLCR prediction. After integrating the three predictions, the system obtains the complete PLCR matrix for the first (t+1) time slots. The system then sets t to (t+1) and continues processing the next row of the PLCR matrix. This process continues until the last row of the PLCR matrix is ​​processed, yielding the final trajectory.

[0101] S3. Determine the user's speed and predict the wireless signal characteristics at the next moment;

[0102] In this embodiment, the neural network model provides local tracking results for the initial period; the system obtains velocity estimates by calculating the differences between adjacent positions of the trajectory; through the Fresnel zone model, a model-based PLCR prediction can be derived for each time index; due to the continuity of human motion, this prediction can be reused in the next time slot, and the system continues to perform the prediction phase operation for the next iteration, processing the data of the next time slot until the iteration ends and the final trajectory is obtained.

[0103] like Figure 5As shown, in this embodiment, a neural network is designed to map the PLCR (input) to the user's trajectory (output). The basic idea of ​​the simultaneous tracking and prediction algorithm is to predict the PLCR in reverse through local trajectories during the tracking process. By continuously iterating through simultaneous tracking and prediction, the system can gradually compensate for missing PLCR features. To achieve this, the system must obtain a relatively accurate local trajectory prediction at the beginning. Otherwise, without an initial local trajectory prediction, it is impossible to compensate for missing features in reverse. Subsequently, the system gradually adjusts and optimizes the local trajectory prediction by continuing to execute the algorithm, and gradually obtains the complete trajectory.

[0104] like Figure 6 As shown, in this embodiment, to obtain initial local trajectory predictions, the system can only rely on the first type of prediction based on observations. This is because only this prediction can be obtained when the user location sequence is unknown. Therefore, the system extracts the first N values ​​from the observation matrix. f The data are processed and used as input to a neural network to obtain a preliminary local trajectory prediction. Empirically, N... f This is set to the number of time slots corresponding to 1 second. It's worth noting that at this point, the system only obtains the data corresponding to the first N... f The trajectory is incomplete because it consists of a sequence of positions in a time slot. The system then iteratively performs simultaneous tracking and prediction. Since this algorithm uses a method of progressively filling in missing data, the system can fill in one missing row of data in the original PLCR matrix after each iteration. By using the prediction integration method described above, the system can acquire a local trajectory through tracking and obtain the PLCR prediction for the next time step in each iteration. With this prediction, the system can compensate for missing signal features and then re-predict a more complete trajectory until the iteration ends. By continuously iterating and simultaneously tracking and predicting, the system can eventually obtain the final complete tracking result. For the passive localization effect of this method, see [link to documentation]. Figure 6 Experiments show that when the communication duty cycle is as low as 20%, the tracking error of this invention is only 0.47m. Compared with the state-of-the-art work, this invention reduces the tracking error by 79.19%.

[0105] Example 2

[0106] Example: The system was deployed on five commercial WiFi devices equipped with Intel 5300 wireless network cards. One device was designated as the transmitter, equipped with a single antenna. The remaining four devices served as receivers, each equipped with three antennas arranged in a linear array (antenna spacing of 2.5 cm). Linux 802.11n CSITool was installed on the devices to collect CSI readings. Data packets were transmitted at a frequency of 1000 Hz. The transmitter was configured in injection mode, and the receivers operated in monitoring mode on channel 64 and the 5.32 GHz band. The transmitter was located at (2.4, -2.4), and the four receivers were located at (2.4, 2.4), (-2.4, -2.4), (2.4, 0), and (0, -2.4), respectively.

[0107] This invention utilizes commercial WiFi devices to implement a system prototype. Experimental results show that when the communication duty cycle is 20.00%, the system's tracking error is 0.47m. Compared to state-of-the-art work, this invention reduces the tracking error by 79.19%.

[0108] In summary, this invention utilizes WiFi multi-link communication to achieve accurate passive tracking of the human body in indoor environments where wireless signal characteristics are severely lacking. The goal of this invention is to compensate for the missing wireless signal characteristics in real-world application scenarios by mining and utilizing the correlation between signals from multiple WiFi links, thereby achieving precise WiFi passive positioning in non-continuous communication scenarios.

[0109] This invention expands the application scenarios of traditional passive positioning systems, relaxes the unrealistic requirements of traditional sensing technologies for long-term continuous communication, and further realizes the integration of communication and sensing in practical applications of WiFi-based wireless sensing systems.

[0110] This invention proposes three mechanisms for effectively compensating for missing WiFi features in passive tracking scenarios. By combining these three signal feature prediction methods, this invention proposes an algorithm called simultaneous tracking and prediction. This algorithm achieves accurate passive tracking even in the presence of severe WiFi feature loss.

[0111] This invention solves the technical problems in the prior art, such as the underutilization of the correlation between signals between WiFi links and the lack of wireless signal characteristics in real application scenarios, which leads to low accuracy of WiFi passive positioning operations in non-continuous communication scenarios.

[0112] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A feature compensation and passive localization method, characterized in that, The method includes: S1. Perform signal feature processing operations: analyze channel state information to extract WiFi signal features and obtain a PLCR matrix; collect and extract the reflection path change rate (PLCR) from the CSI data, process it to obtain DFS data, and derive an observation-based matrix from the PLCR matrix. and reliability-based matrix Weights are assigned to the first type of prediction, the second type of prediction, and the third type of prediction for combined prediction operations; Specifically, each element in the PLCR matrix is ​​examined. When a missing value is detected, the scan proceeds upwards from the position of the missing value until the most recently observed true value in the same link is encountered. This most recently observed true value is then filled into a preset observation matrix to obtain the observation-based matrix. ; S2. Prepare for PLCR prediction in the tracking phase by obtaining the first type of prediction, the second type of prediction and the third type of prediction, combining them with the weights to obtain the final PLCR prediction, and processing the final PLCR prediction to obtain the applicable PLCR prediction value and user trajectory. Specifically, a prediction based on the observed data is calculated as the first type of prediction, and the matrix based on the observed values ​​is used as the prediction. The Middle The first link Each element is represented as The prediction based on the observation data is obtained as follows: In the formula, It makes and Minimum time index; Calculate the proportional relationship-based prediction as the second type of prediction, obtain and utilize the non-missing values, calculate the proportional relationship-based prediction, all of the reflection path change rate PLCRs at t=t2 are missing, jump to the prediction using the mathematical model; Calculate predictions based on mathematical models, as a third type of prediction, using the PLCR prediction matrix derived through mathematical modeling. Based on this, the previous result was derived in reverse. The reflection path change rate PLCR for each time slot; S3. Determine the user's speed and predict the wireless signal characteristics at the next moment. Design and utilize a neural network to map the user's trajectory based on the reflection path change rate PLCR. Obtain the complete trajectory by executing the tracking operation, the prediction operation, the algorithm adjustment operation, and the local trajectory prediction optimization operation. S3 includes: S31. During the tracking process, the reflection path change rate (PLCR) is predicted in reverse using the local user trajectory. S32. Compensate for missing PLCR features by continuously iterating through the tracking operation and the prediction operation; S33. Continuously execute the algorithm adjustment operation and the local trajectory prediction optimization operation to obtain the complete trajectory.

2. The feature compensation and passive localization method according to claim 1, characterized in that, S1 includes: CSI data is acquired, and the reflection path change rate (PLCR) is extracted from the raw CSI readings by performing a short-time Fourier transform (STFT). The CSI as a function of frequency f and time t can be expressed as follows: In the formula, Indicates static CSI components. It is a dynamic CSI component The corresponding path length, For wavelength, The signal amplitude is... For phase; Based on the Doppler effect, the following formula can be derived: In the formula, This refers to the DFS data. This represents the reflection path change rate PLCR. It is in time The dynamic path length; In the reliability-based matrix In the process, when the actual reflection path change rate (PLCR) can be observed at a specific location, the weight assigned to the first type of prediction is 1; when signal features are continuously lost, the weight of the first type of prediction is reduced; when a link experiences continuous loss of the features, a quadratic function is used to reduce the weight of the first type of prediction and assign weights to the second type of prediction and the third type of prediction.

3. The feature compensation and passive positioning method according to claim 2, characterized in that, In step S14, the weight of the first type of prediction is reduced using the following logic: 。 4. The feature compensation and passive positioning method according to claim 1, characterized in that, S2 includes: A weighted operation is performed on the first type of prediction, the second type of prediction, and the third type of prediction to obtain the final PLCR prediction, and the elements of the PLCR prediction matrix are updated. The missing features are filled in using the final PLCR prediction to obtain the user trajectory.

5. The feature compensation and passive positioning method according to claim 4, characterized in that, In step S22, the prediction based on the proportional relationship is obtained using the following logic: 。 6. The feature compensation and passive positioning method according to claim 4, characterized in that, S23 includes: S231. For the pre-set link, the transmitter position is set as follows: Assume the receiver's location is: The current location of the person is: The speed of a person is: ; S232. Using the following logic, obtain the PLCR prediction matrix based on the mathematical model: S233. When all PLCR data of all links are lost during the preset time slot, PLCR prediction is performed based on the known location sequence.

7. The feature compensation and passive positioning method according to claim 6, characterized in that, The PLCR prediction matrix based on the mathematical model satisfies: 。 8. The feature compensation and passive positioning method according to claim 1, characterized in that, In step S24, the PLCR prediction matrix is ​​updated using the following logic. Each element: In the formula, This represents the weight of the prediction based on the observations.

9. A feature compensation and passive positioning system, used to execute the feature compensation and passive positioning method according to any one of claims 1 to 8, characterized in that, The system includes: The signal feature processing module performs signal feature processing operations. It analyzes channel state information to extract WiFi signal features and processes them to obtain a PLCR matrix. It also collects and extracts the reflection path change rate (PLCR) from the CSI data, processes it to obtain DFS data, and derives an observation-based matrix from the PLCR matrix. and reliability-based matrix Weights are assigned to the first type of prediction, the second type of prediction, and the third type of prediction for combined prediction operations; The prediction and combination module is used to prepare for PLCR prediction in the tracking phase. It obtains the first type of prediction, the second type of prediction and the third type of prediction, combines them with the weights to obtain the final PLCR prediction, and processes the final PLCR prediction to obtain the applicable PLCR prediction value and user trajectory. The prediction and combination module is connected to the signal feature processing module. The wireless signal feature prediction module is used to determine the user's speed and predict the wireless signal features at the next moment. It designs and utilizes a neural network to map the user's trajectory based on the reflection path change rate (PLCR). By executing the tracking operation, the prediction operation, the algorithm adjustment operation, and the local trajectory prediction optimization operation, the complete trajectory is obtained. The wireless signal feature prediction module is connected to the prediction combination module.

Images