A method for high-precision indoor positioning based on CSI of a sense-integrated system

Abstract

The application belongs to the technical field of indoor positioning, and particularly relates to a CSI-based indoor high-precision positioning method of a sensing-integrated system, comprising the following steps: S1: an algorithm initialization stage, signal preprocessing is performed; S2: an SVM is used to identify and classify a LoS path, according to a received signal, accurate identification and classification of the LoS path are realized; S3: according to the number of LoS paths, a corresponding positioning method is adopted to obtain a user position; and S4: output of the user position. The application considers the influence of non-direct paths in a complex indoor environment, and based on the CSI characteristics of ISAC signals, through fingerprint and model dual-mode positioning, the signal classification process is mainly to identify whether the signal received by the receiver is a LoS signal through a support vector machine, and the fingerprint positioning is used to accurately determine the position of the user, the algorithm has the advantages of high positioning precision, low complexity and suitability for complex indoor environments, and has a bright future in future indoor positioning.

Description

Technical Field

[0001] This invention belongs to the field of indoor positioning technology, specifically relating to a high-precision indoor positioning method based on CSI for a sensor-integrated system. Background Technology

[0002] Driven by emerging applications, future 5G and 6G wireless systems will need to provide high-precision sensing services. Meanwhile, due to the widespread application of millimeter wave and MIMO technologies, communication signals exhibit high resolution in both the time and angular domains, making it possible to achieve high-precision sensing using communication signals and greatly promoting the research on the integration of communication and sensing.

[0003] High-precision positioning is the foundation for navigation, surveillance, and crowd sensing. Although the GPS system has achieved satisfactory results in outdoor positioning and navigation, its performance in indoor environments is not ideal. In addition, using GPS technology requires additional hardware costs and consumes more energy. Therefore, in indoor positioning, using ISAC signals to achieve location sensing is a very effective solution to get rid of dependence on the GPS system.

[0004] Traditional positioning often uses signal strength of arrival (RSS), time of arrival or time difference of arrival (TOA and TDOA), or angle of arrival (AOA), but all of these have certain limitations.

[0005] RSS requires measuring the strength of the received signal and using a propagation model to estimate the relationship between distance and signal strength. However, the fundamental reasons why RSS is unsuitable for indoor positioning stem from two aspects. First, RSS is measured from the radio frequency signal on each data packet, making it difficult to obtain an accurate value. Second, RSS is easily altered by multipath effects, as multiple signals may reach the receiver via different paths in addition to the Loss of Position (LoS) signal. This multipath effect is more pronounced in indoor environments with ceilings, floors, and walls.

[0006] TOA and TDOA are positioning technologies based on wireless signal transmission time. Both technologies require the location of at least three anchor points to be known during the positioning phase, and both have high requirements for time synchronization, which is not easy to achieve.

[0007] AOA requires measuring the incident angle of the wireless signal, and the receiver must have a directional antenna array. Due to the complex indoor environment (such as people moving around and signal interference), measuring the incident angle becomes extremely difficult.

[0008] On the other hand, fingerprint-based methods do not require knowledge of the anchor point's location, nor do they require measurement of time and angle. They are easy to implement, easy to measure, and low in cost, and have gradually become the mainstream trend in indoor positioning technology.

[0009] Traditional fingerprint-based positioning methods typically use the collected RSS (Receiving Signal Segmentation) as a channel feature. However, due to multipath effects, the RSS is extremely unstable. Although preprocessing methods such as Kalman filtering can reduce this instability, large errors will occur when the received signal fluctuates significantly, greatly affecting positioning accuracy. Therefore, there is a need to develop a positioning method with high accuracy and low complexity for use in complex indoor environments. Summary of the Invention

[0010] The purpose of this invention is to provide a high-precision indoor positioning method based on CSI for a sensor-integrated system. This method has many advantages, such as high positioning accuracy, low complexity, and applicability to complex indoor environments, and has a bright future in indoor positioning.

[0011] The specific technical solution adopted by this invention is as follows:

[0012] A high-precision indoor positioning method based on CSI for a sensor-integrated system includes the following steps:

[0013] S1: Algorithm initialization phase, signal preprocessing is performed;

[0014] S2: SVM identifies and classifies LosS paths, achieving accurate identification and classification of LosS paths based on the received signal;

[0015] S3: Use the appropriate positioning method based on the number of LoS paths to obtain the user's location;

[0016] S4: Output of user location.

[0017] Furthermore, the signal preprocessing in S1 involves dimensionality reduction, outlier processing, and normalization of the collected CSI.

[0018] The dimensionality reduction performed on the CSI is PCA dimensionality reduction, which includes the following steps:

[0019] At each position y, assume there exist T p-dimensional vectors, i.e. remember

[0020] The key to PCA is finding the mean matrix U to calculate the variance. First, the mean vector of the CSI vector must be found. Furthermore, the covariance matrix can be expressed as V = U'U / T. Find the q largest eigenvalues ​​of V and their corresponding eigenvectors. This allows the p-dimensional CSI vector to be... Convert to a q-dimensional data vector d pca (q<p), the conversion formula is d pca =(d-γ)Φ pca ;

[0021] The abnormal data processing performed on the CSI is abnormal data filtering, which includes the following steps:

[0022] Using the Pauta criterion, the mean of a sufficient amount of collected CSI data is first calculated. Then, each raw data point is compared with the mean. Data points with a difference greater than 3σ from the mean are considered outliers and need to be filtered out. Here, 3σ represents the standard deviation of the raw data.

[0023] The normalization operation performed on the CSI is called data normalization, which includes the following steps:

[0024] The original CSI data range is limited to 0 to 1.

[0025] Further, S2 includes the following steps:

[0026] In the process of using SVM, LosS data is labeled as 1 and NLoS data is labeled as -1: For this dataset, SVM is used to classify the amplitude and phase information of CSI, and then the solution is obtained.

[0027]

[0028]

[0029] Among them, c i ∈{-1,1}, where n represents the number of samples, and w∈R q ξ represents the direction of the classification hyperplane, b∈R represents the position of the hyperplane, C represents the tradeoff coefficient between the classification error and the maximum separable region, and ξ=(ξ1,ξ2,...,ξ n ) represents a slack variable;

[0030] By solving the above equation, we obtain the following classification function:

[0031]

[0032] Where, α i Let K(CSI) be the Lagrange factor and the kernel function be K(CSI). normalized,i CSI) is

[0033] K(x i ,x j )=exp(-γ||x i -x j || 2 )

[0034] Where γ > 0 is the kernel parameter, f(CSI) > 0 indicates LOS data, and vice versa indicates NLOS data.

[0035] Further, S3 includes the following steps:

[0036] If there are three or more Loss paths, the user's location is calculated directly using a mathematical model-based method.

[0037] If there are two Loss paths, the intersection point is calculated using the model, and the user's location is calculated by locating the intersection point through the correlation of CSI.

[0038] If a Loss path exists, the user's location can be obtained using a hybrid model-fingerprint method.

[0039] If no Loss path exists, but only an NLoS path exists, fingerprinting can be used to achieve online matching and determine the user's location.

[0040] Furthermore, in step S4, if a Loss path exists, the user location can be directly output; if no Loss path exists, different fingerprint matching methods are mixed, and the different locations obtained are weighted to output the final user location.

[0041] The technical effects achieved by this invention are as follows:

[0042] The present invention discloses a high-precision indoor positioning method based on CSI for a sensor-integrated system. This method considers the influence of non-direct light paths in complex indoor environments and, based on the CSI characteristics of ISAC signals, uses fingerprint and model dual-mode positioning. The signal classification process mainly uses a support vector machine to identify whether the signal received by the receiver is a Loss signal. Fingerprint positioning is used to accurately determine the user's location. The algorithm has many advantages such as high positioning accuracy, low complexity, and applicability to complex indoor environments, and has a bright future in indoor positioning. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of the indoor positioning system of the integrated communication and sensing system of the present invention;

[0044] Figure 2 This is a schematic diagram of the indoor positioning algorithm of the present invention;

[0045] Figure 3 This is a schematic diagram of the positioning algorithm of the present invention in the case of three LoS paths;

[0046] Figure 4 This is a schematic diagram of the positioning algorithm of the present invention in the case of two Loss paths;

[0047] Figure 5 This is a schematic diagram of the positioning of the present invention in an NLoS environment;

[0048] Figure 6This is a performance graph of the positioning error of the positioning algorithm of this invention under different path loss factors;

[0049] Figure 7 This is a performance graph of the positioning error of the positioning algorithm of this invention under different fingerprint database grids. Detailed Implementation

[0050] To make the objectives and advantages of this invention clearer, the invention will be specifically described below with reference to embodiments. It should be understood that the following text is merely used to describe one or more specific embodiments of the invention and does not strictly limit the scope of protection specifically claimed by the invention.

[0051] Example 1:

[0052] like Figure 1 As shown, a high-precision indoor positioning method based on CSI (Communication, Sensing, and Signal Processing) in a sensor-integrated system is proposed. Based on integrated communication and sensing signals, it considers the impact of NLOS (No-LoS) path interference in the indoor environment on positioning. It utilizes the CSI characteristics of the signal and employs a fingerprint model for hybrid positioning. This high-precision indoor positioning method comprises two parts: signal classification and fingerprint positioning. The signal classification process mainly uses a Support Vector Machine (SVM) to identify whether the signal received by the receiver is a Loss of Sight (LoS) signal. Fingerprint positioning is used to determine the user's location. Different positioning methods are adopted based on the type of received signal, fully utilizing LoS path information and removing NLoS path interference. It is suitable for scenarios providing accurate positioning services to mobile users in complex indoor environments with NLoS paths. It is applicable to indoor environments where each access point (AP) is equipped with an M... T In an OFDM system with a single transmitting antenna, the access point (AP) node sends an ISAC signal and receives the user's echo. The frequency domain information of the echo is extracted to perform positioning calculations. This invention is applicable to communication environments where the wireless link between the user and the wireless AP node exhibits NLoS (Network Loss of Speed). Figure 1 A typical indoor positioning scenario for an integrated communication and sensing system is presented.

[0053] Specifically, an indoor positioning process is disclosed here:

[0054] When the target is active indoors, it will receive ISAC signals from one or more indoor devices to achieve positioning, such as... Figure 1 As shown, it is assumed that each AP in the indoor environment has M T The target has M-type transmitting antennas. R Considering an OFDM system with a single transmit antenna, since OFDM can acquire CSI from multiple subcarriers, assuming there are K subcarriers, then in a pair of transmit and receive antenna links, the frequency domain representation of the signal received by the target for each subcarrier is as follows:

[0055] y m,k =Hm,k x m,k +w m,k

[0056] in, This indicates that the user receives the signal. This indicates the transmitted ISAC signal. This represents additive white Gaussian noise with zero mean. This represents the channel matrix. The channel matrix is ​​generally related to the user's location; therefore, the channel matrix for user u is represented as H. m,k [u], under user u, APm, and subcarrier k, the channel matrix of the link between transmit antenna l and receive antenna r can be expressed as:

[0057]

[0058] in, Indicates the amplitude response of the channel. This represents the phase response of subcarrier k.

[0059] In the presence of Loss of Space (LoS), assuming that signal propagation only experiences free space propagation loss, the distance d and the channel matrix have the following relationship:

[0060]

[0061] Where c represents the signal transmission speed, n represents the path loss factor, which is environment-dependent, and f c CSI represents the center frequency of the entire bandwidth. eff σ represents the effective value of the channel matrix with LOS between user u and APm, and σ represents the environmental factor, which is related to the environment.

[0062] The goal of the positioning algorithm is to receive signal y m,k Locate the user's location.

[0063] The indoor high-precision positioning method of this invention is based on fingerprint recognition and Loss and Non-LoS identification and classification. The main body of the method is divided into four stages: algorithm initialization, SVM identification and classification of LosS paths, adopting corresponding positioning methods according to the number of LosS paths, and user location output. Specifically, it includes the following steps:

[0064] S1: Algorithm initialization phase;

[0065] In this stage, signal preprocessing is performed, including dimensionality reduction, outlier processing, and normalization of the collected CSI data.

[0066] The dimensionality reduction performed on CSI is PCA dimensionality reduction, which includes the following steps:

[0067] At each position y, assume there exist T p-dimensional vectors, i.e. remember

[0068] The key to PCA is finding the mean matrix U to calculate the variance. First, the mean vector of the CSI vector must be found. Furthermore, the covariance matrix can be expressed as V = U'U / T. Find the q largest eigenvalues ​​of V and their corresponding eigenvectors. This allows the p-dimensional CSI vector to be... Convert to a q-dimensional data vector d pca (q<p), the conversion formula is d pca =(d-γ)Φ pca .

[0069] The abnormal data processing performed on CSI is abnormal data filtering, which includes the following steps:

[0070] Using the Pauta criterion, the mean of a sufficient amount of collected CSI data is first calculated. Then, each raw data point is compared with the mean. Data points with a difference greater than 3σ from the mean are considered outliers and need to be filtered out, where 3σ represents the standard deviation of the raw data.

[0071] The normalization operation performed on CSI is data normalization, which includes the following steps:

[0072] The original CSI data range is limited to 0 to 1.

[0073] S2: SVM identifies and classifies LosS paths;

[0074] In this stage, based on the received signal, accurate identification and classification of the LosS path are achieved. In this technical solution, SVM is used to mark the amplitude and phase data of CSI to distinguish between LosS and NLoS.

[0075] Specifically, S2 includes the following steps:

[0076] In the process of using SVM, LosS data is labeled as 1 and NLoS data is labeled as -1: For this dataset, SVM is used to classify the amplitude and phase information of CSI, and then the solution is obtained.

[0077]

[0078]

[0079] Among them, c i ∈{-1,1}, where n represents the number of samples, and w∈R qξ represents the direction of the classification hyperplane, b∈R represents the position of the hyperplane, C represents the tradeoff coefficient between the classification error and the maximum separable region, and ξ=(ξ1,ξ2,...,ξ n ) represents a slack variable, which allows some samples to make mistakes on the classification hyperplane;

[0080] By solving the above equation, we obtain the following classification function:

[0081]

[0082] Where, α i Let K(CSI) be the Lagrange factor and the kernel function be K(CSI). normalized,i CSI) is

[0083] K(x i ,x j )=exp(-γ||xi-xj|| 2 )

[0084] Where γ > 0 is the kernel parameter. f(CSI) > 0 indicates LOS data, and vice versa.

[0085] S3: Based on the number of Loss paths, adopt the corresponding positioning method to obtain the user's location.

[0086] After step S2, the LosS and NLoS labels for each CSI of the target can be obtained. Therefore, the algorithm is applied to the processed measurements for localization. The overall framework of the algorithm is as follows: Figure 2 As shown.

[0087] Specifically, S3 will be described in further detail here, and S3 includes the following steps:

[0088] If there are three or more Loss paths, the user's location can be calculated directly using a mathematical model-based method, as follows:

[0089] Among the CSIs labeled LOS, three LOS path CSIs with the highest confidence levels and located at different APs are selected. The distance 'd' is calculated, and the intersection of the three circles is obtained with the reference AP as the center and the distance as the radius. However, due to noise, the three circles cannot intersect at a single point; only a region can be determined. Within this region, the target's location needs to be further determined, such as... Figure 3 As shown. Let the intersection points be A, B, and C, with coordinates (x, y, y) respectively. A ,y A ), (x B ,y B ), (x C ,y CIf the coordinates of its centroid are taken as the starting point coordinates, the gradient descent method is used to minimize the error between the estimated position and the true position, thus obtaining the optimal position estimate. The error function is:

[0090]

[0091] Where p = [x, y] T p is the estimated location of the target. i =[x i ,y i ] T Let d be the position of the i-th reference AP. i Let L represent the distance between the target and the i-th AP, and let L represent the number of APs that can be received, where L = 3.

[0092] If two Loss paths exist, the intersection point is calculated using the model, and the user's location is calculated by locating the intersection point through CSI correlation, as detailed below:

[0093] First, calculate the distance d between the two LOS values. i At this point, the distance d can be used to determine the distance. i The two intersecting circles are obtained, such as Figure 4 As shown, the two intersecting circles correspond to two intersection points, therefore the estimated target position is one of these two intersection points. Let the intersection point be (x... A ,y A ), (x B ,y B At this point, the fingerprint near the intersection coordinates is correlated with the measured CSI, and the intersection point where the area with high correlation is located is taken as the target location.

[0094] If a Loss-of-Stake (LoS) path exists, the user's location can be obtained using a hybrid model-fingerprint approach, as follows:

[0095] First, the distance *d* corresponding to the LOS path is calculated, indicating that the target's location lies on the circle. The collected target fingerprints are grouped by time interval (assuming n groups). The average CSI of the fingerprints within each time interval is calculated. Then, the average CSI fingerprint is matched against a fingerprint database. Using the KNN algorithm, n estimated coordinates are obtained. The distances from these n coordinates to the circle are calculated, and the coordinate closest to the circle is taken as the estimated target location coordinate.

[0096]

[0097] Here, A represents the set of matched coordinates.

[0098] If no Loss path exists, and only an NLoS path exists, then fingerprinting is used to achieve online matching and thus determine the user's location, as follows:

[0099] When no LOS path exists, a multi-fingerprint algorithm is adopted, using KNN+SVM regression and two fingerprint algorithms for weighted localization:

[0100] During training, SVM regression is used to establish the dependency relationship between the target position and the amplitude and phase of the CSI (Constant Sensor Indicator). That is, a representation (CSI) is established. normalized The functional relationship f of (x, x) x With characterization (CSI) normalized The functional relationship f of ,y) y Then, using the trained SVM architecture, an estimate of the user's location (x1, y1) can be obtained.

[0101] Next, for the second fingerprint matching algorithm, KNN, a distance model is used for assistance. First, the distance is calculated to "straighten" the NLOS, such as... Figure 5 As shown, all NLOS data are treated as LOS data first. The distances in the graph have the following relationships.

[0102] d 1,NLOS +d 2,NLOS =d 1,NLOS +d 3,NLOS

[0103] Then calculate the distances of all CSIs received by the target, and denote the minimum distance as d. min Therefore, in subsequent cases where the value is less than d min Perform a full match within the specified positions, matching both LOS and NLOS data; for positions greater than d... min The region only performs partial matching, that is, only matching data where CSI is NLOS. For the matched NLOS data, a "straightening" operation is also performed, and d is calculated using the model. i , will fall into d min Collect a range of CSIs within a specific domain, denoted as the selected matching CSIs, and then match the selected matching CSIs with values ​​less than d. min The CSI with LOS label is used to obtain another estimate of the target location (x2, y2) using KNN.

[0104] S4: Output of user location. If a Loss path exists, the user location can be output directly; if no Loss path exists, different fingerprint matching methods are used to weight the obtained locations and output the final user location.

[0105] When in use, the positioning algorithm first executes the algorithm initialization step S1; then steps S2 and S3 are executed sequentially; finally, the algorithm outputs the user result according to step S4. The steps of the algorithm are summarized in Table 1.

[0106]

[0107] Table 1. Indoor positioning methods based on CSI

[0108] Example 2:

[0109] This embodiment is disclosed based on embodiment 1. Specifically, in a wireless AP with 64 antennas in URA-shaped antenna array, 100 subcarriers are transmitted each time, with a center frequency of 2.61GHz and a bandwidth of 20MHz. Four users walk along a predetermined trajectory, collect CSI values ​​at different locations, and create a fingerprint dataset.

[0110] Figure 6 This paper demonstrates the error performance of the distance model using different path loss factors and environmental factors under the LOS path model. As shown in the figure, if all rooms use the same environmental factor, the impact of the path loss factor on the error is random in different rooms, making it difficult to compare the algorithm accuracy across different rooms. However, if the minimum error criterion established under different environments is used in each room, the errors in each room become less different and can be approximated as the same, achieving "environmental factor independence." In this case, the error increases with the increase of the path loss factor. When the path loss factor is 2, the error is around 0.35m, achieving decimeter-level accuracy.

[0111] Figure 7 The proposed algorithm, along with KNN and SVM, is demonstrated using ISAC signals at different grid spacings. It can be seen that the proposed method has the lowest error, maintaining decimeter-level accuracy even when the grid spacing varies from 0.5m to 2.5m. This is because, compared to KNN and SVM algorithms, the proposed method effectively utilizes the Loss of Space (LoS) data from the measurement and uses virtual users for CSI matching, thereby further improving positioning accuracy. Therefore, the proposed method outperforms traditional fingerprint matching algorithms.

[0112] In summary, this technical solution has many advantages, such as high positioning accuracy, low complexity, and applicability to complex indoor environments, and has a bright future in indoor positioning.

[0113] The above description is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained in this invention are implemented according to conventional methods in the art unless otherwise specified or limited.

Claims

1. A high-precision indoor positioning method based on CSI for a sensor-integrated system, characterized in that: Includes the following steps: S1: Algorithm initialization phase, signal preprocessing is performed; S2: SVM identifies and classifies LosS paths, achieving accurate identification and classification of LosS paths based on the received signal; S3: Use the appropriate positioning method based on the number of LoS paths to obtain the user's location; S4: Output of user location; The signal preprocessing in S1 involves dimensionality reduction, outlier processing, and normalization of the collected CSI. The dimensionality reduction performed on the CSI is PCA dimensionality reduction, which includes the following steps: In each location At this point, assuming it exists indivual A dimensional vector, that is ,remember ; The key to PCA is finding the mean matrix. To calculate the variance, the mean vector of the CSI vector must first be found. And the covariance matrix can be expressed as ,turn up of The largest eigenvalue is related to its corresponding eigenvector. Then you can 3D CSI vector Convert to 3D data vector , The conversion formula is: ; S2 includes the following steps: In the process of using SVM, LosS data is labeled as 1 and NLoS data is labeled as -1: For this dataset, SVM is used to classify the amplitude and phase information of CSI, and then the solution is obtained. in, , Indicates the number of samples. This indicates the direction of the classification hyperplane. Indicates the position of the hyperplane. This represents the trade-off between classification error and the maximum separable region. Represents slack variables; By solving the above equation, we obtain the following classification function: in, Let Lagrange multiplier and kernel function be the Lagrange multiplier. for in, For kernel parameters, It is represented as LOS data, and vice versa as NLOS data.

2. The indoor high-precision positioning method based on CSI for a sensor-integrated system according to claim 1, characterized in that: The abnormal data processing performed on the CSI is abnormal data filtering, which includes the following steps: Using the Pauta criterion, first, the mean of a sufficient amount of collected CSI data is calculated. Then, each raw data point is compared to the mean. Data points with a difference greater than 100% are considered valid. Data that is considered abnormal and needs to be filtered out includes... This represents the standard deviation of the original data.

3. The indoor high-precision positioning method based on CSI for a sensor-integrated system according to claim 1, characterized in that: The normalization operation performed on the CSI is called data normalization, which includes the following steps: The original CSI data range is limited to 0 to 1.

4. The indoor high-precision positioning method based on CSI for a sensor-integrated system according to claim 1, characterized in that: S3 includes the following steps: If there are three or more Loss paths, the user's location is calculated directly using a mathematical model-based method. If there are two Loss paths, the intersection point is calculated using the model, and the user's location is calculated by locating the intersection point through the correlation of CSI. If a Loss path exists, the user's location can be obtained using a hybrid model-fingerprint method. If no Loss path exists, but only an NLoS path exists, fingerprinting can be used to achieve online matching and determine the user's location.

5. The indoor high-precision positioning method based on CSI for a sensor-integrated system according to claim 4, characterized in that: In step S4, if a Loss path exists, the user's location can be directly output; if no Loss path exists, different fingerprint matching methods are combined, and the different locations obtained are weighted to output the final user location.

Images