A location-based clustering method in single station terminal fingerprint positioning

Abstract

The application discloses a location-based clustering method in single-station terminal fingerprint positioning and belongs to the technical field of fingerprint positioning. The method of the application is aimed at the problem that the traditional fingerprint database clustering method, such as K-means clustering, in fingerprint positioning will obviously lose positioning accuracy, and a location-based clustering method is provided, wherein the method clusters the fingerprint database according to physical position characteristics under the premise of not losing positioning accuracy or losing the least positioning accuracy, can ensure that the positioning accuracy is not affected or is affected the least, and greatly improves the positioning speed.

Description

Technical Field

[0001] This invention relates to a location-based clustering method for fingerprint positioning of single-station terminals, belonging to the field of signal and information processing technology. Background Technology

[0002] Today, location-based services (LBS) are widely used in many fields such as express delivery and logistics, smart homes, and emergency rescue, and LBS has become a basic need for people's daily production and life. Currently, one of the most mature LBS technologies is the Global Positioning System (GPS), which originated for military applications. Traditional positioning methods are generally based on triangulation. The biggest problem with these methods is that in non-line-of-sight (NLOS) scenarios where electromagnetic wave scattering and diffraction phenomena occur, they are affected by multipath effects, resulting in a significant decrease in positioning accuracy.

[0003] In recent years, fingerprint positioning technology has received widespread attention and research due to its ability to overcome the problems of traditional wireless network positioning technology and to effectively locate targets by fully utilizing multipath information from the channel. In fingerprint positioning methods, when the area to be located is large or the grid side length is small, the number of reference points within the area will be very large, resulting in a large fingerprint database. When using the nearest neighbor algorithm for positioning, the online positioning speed will be slow. Therefore, to ensure real-time positioning, it is necessary to perform clustering processing on the fingerprint database offline. The most commonly used clustering algorithm is K-means clustering, but K-means clustering has two problems: first, the selection of initial cluster centers and the K value has a significant impact on the clustering results; second, the positioning accuracy will decrease significantly when clustering the fingerprint database. Therefore, it is of great significance to invent a clustering method that can significantly improve positioning speed while ensuring that positioning accuracy is not affected or is minimally affected. Summary of the Invention

[0004] To address the issue that traditional fingerprint database clustering methods in fingerprint positioning significantly reduce positioning accuracy, this invention proposes a location-based clustering method for single-station terminal fingerprint positioning. Based on physical location characteristics, fingerprint database clustering is performed with minimal or no loss of positioning accuracy, thereby significantly improving positioning speed while ensuring that positioning accuracy is not affected or is minimally affected.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0006] The present invention provides a location-based clustering method for single-station terminal fingerprint positioning, comprising the following steps:

[0007] Step 1: Extract the fingerprint matrix FRCPM of the reference points, compress the fingerprint FRCPM, and store the position coordinates of all reference points in the area to be located and their compressed fingerprints FCPV one by one to obtain the fingerprint database.

[0008] Step 2: Perform location-based clustering (PBC) on the data in the fingerprint database based on physical location features;

[0009] Step 3: Extract the fingerprint matrix FRCPM of the point to be located, and compress the fingerprint matrix FRCPM of the point to be located to obtain the compressed fingerprint FCPV of the point to be located; match the compressed fingerprint FCPV of the point to be located with the compressed fingerprint FCPV in the fingerprint database to perform online positioning.

[0010] Furthermore, the reference point mentioned in step 1 refers to dividing the area to be located into a uniform grid, with the grid nodes serving as reference points.

[0011] Furthermore, in step 1, the fingerprint matrix FRCPM of the reference points is extracted and the fingerprint FRCPM is compressed. The specific steps are as follows:

[0012] Step 1.1: Record the coordinates of each reference point and collect the uplink signal of the terminal device at each reference point. Perform Hadamard product on the CFR matrix and its conjugate matrix of each reference point to obtain the fingerprint FRCPM of each reference point.

[0013] Step 1.2: Sum the elements of each column of the fingerprint matrix FRCPM to obtain a matrix with N elements. l The row vector is the compressed fingerprint FCPV;

[0014] The compressed fingerprint FCPV of the k-th reference point is denoted as FC. k FC k The j-th element FC kj Represented as:

[0015]

[0016] Furthermore, in step 2, location-based clustering (PBC) is performed on the data in the fingerprint database based on physical location features. The specific steps are as follows:

[0017] Step 2.1: Using the test dataset collected in the offline phase, calculate the average localization error in the absence of clustering in the fingerprint database;

[0018] Step 2.2: Initialize the cluster center spacing and cluster side length. Calculate the total positioning time and average positioning error under this clustering condition using the test set. The combination of cluster center spacing and cluster side length needs to satisfy that the set of all classes can cover the entire area to be located, and there can be overlapping parts between different classes, that is, a reference point can belong to several different classes at the same time.

[0019] Step 2.3: Change the cluster center spacing or cluster side length, recalculate the total positioning time and average positioning error. The new combination of cluster center spacing and cluster side length still needs to meet the conditions in Step 2.2, and should not repeat the previous clustering cases.

[0020] Step 2.4: Repeat step 2.3 until all combinations of class center spacing and class side length that meet the conditions have been traversed;

[0021] Step 2.5: Compare the total positioning time of all cases where the average positioning error is not greater than that of the case without clustering, and select the case with the smallest time as the final clustering result. If the average positioning error of all cases is greater than that of the case without clustering, then select the case with the smallest average positioning error as the final clustering result.

[0022] Furthermore, in step 3, the compressed fingerprint FCPV of the point to be located is matched with the compressed fingerprint FCPV in the fingerprint database for online localization. The specific details are as follows:

[0023] Step 3.1: Using the fingerprint and coordinate data of all reference points in the established fingerprint database, determine the optimal K value for the weighted K-nearest neighbor algorithm;

[0024] Step 3.2: Calculate the Euclidean distance from the compressed fingerprint FCPV of the point to be located to all the compressed fingerprint FCPVs of the class centers, and assign the point to be located to the class whose fingerprint is closest to the fingerprint of the class center;

[0025] Step 3.3: In this class, the weighted K-nearest neighbor matching algorithm is used to estimate the coordinates of the point to be located, as shown in the formula: In the formula, Represents the coordinates of the point to be located. w represents the coordinates of the k-th reference point. k d represents the weight corresponding to the k-th reference point. k It is the fingerprint distance between the k-th reference point and the point to be located, where Δd is a very small positive number.

[0026] Furthermore, in step 3, cross-validation is used to determine the optimal K value for the weighted K-nearest neighbor algorithm.

[0027] To address the issue that traditional fingerprint database clustering methods for fingerprint localization significantly reduce localization accuracy, this invention employs the above technical solution to achieve the following technical effects: without sacrificing localization accuracy or with minimal loss of localization accuracy, it can greatly improve online localization speed, achieving an improvement in localization speed comparable to that of K-means clustering. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the steps of the PBC algorithm in the present invention;

[0029] Figure 2 This is a schematic diagram illustrating the principle of the PBC algorithm of the present invention;

[0030] Figure 3 This is a comparison chart of the localization performance of the method of this invention and the existing K-means clustering algorithm; Detailed Implementation

[0031] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0032] This invention proposes a location-based clustering method for fingerprint localization of single-station terminals. The method is divided into an offline fingerprint database establishment stage and an online matching and localization stage, and mainly includes the following specific steps.

[0033] The present invention discloses a location-based clustering method for single-site terminal fingerprint positioning, based on a MIMO-OFDM single-site system, wherein the base station is equipped with N in a uniform linear array (ULA) configuration. t With one antenna, the total number of OFDM subcarriers is N. l Numerous scatterers are randomly distributed within the area to be located; specifically, the method includes the following steps:

[0034] Step 1: Extract the fingerprint matrix FRCPM of reference points in the offline stage, compress the fingerprint FRCPM, and establish a fingerprint database;

[0035] The reference point refers to dividing the area to be located into a uniform grid, and using the grid nodes as reference points.

[0036] Step 1.1: Extract the fingerprint FRCPM of the reference point;

[0037] Record the coordinates of each reference point and acquire the uplink signal from the terminal device at each reference point. Perform a Hadamard product of the CFR matrix and its conjugate matrix for each reference point to obtain the fingerprint FRCPM for that reference point. The specific formula is as follows: Where H k The CFR matrix representing the k-th reference point, FRCPM is an N t×N l A real matrix, F k The frequency response channel power matrix (FRCPM) represents the frequency response channel power matrix at the k-th reference point.

[0038] The entire channel frequency response (CFR) matrix at the k-th reference point can be expressed as... H k It is an N t ×N l complex matrix; h k,n This represents the channel frequency response (CFR) vector of the nth link at the kth reference point.

[0039] Step 1.2: The fingerprint matrix FRCPM of all reference points is compressed.

[0040] Specifically, the elements of each column of the fingerprint matrix FRCPM are summed to obtain an element with N elements. l The row vector is the compressed fingerprint frequency channel power vector (FCPV).

[0041] The compressed fingerprint FCPV of the k-th reference point is denoted as FC. k FC k The j-th element FC kj Represented as:

[0042]

[0043] Step 1.3: Store the position coordinates of all reference points in the area to be located and their fingerprints FCPVs one by one to obtain the fingerprint database;

[0044] Step 2: Based on the physical location characteristics of the fingerprint database, perform Position Based Clustering (PBC) to minimize or eliminate the loss of positioning accuracy. The principle of the PBC algorithm is as follows: Figure 2 As shown, the steps of the PBC algorithm are as follows: Figure 1 As shown, specifically:

[0045] Step 2.1: Using the test dataset collected in the offline phase, calculate the average localization error in the absence of clustering in the fingerprint database;

[0046] Step 2.2: Initialize the cluster center spacing and cluster side length. Calculate the total positioning time and average positioning error under this clustering condition using the test set. The combination of cluster center spacing and cluster side length needs to satisfy that the set of all classes can cover the entire area to be located, and there can be overlap between different classes. That is, a reference point can belong to several different classes at the same time. Therefore, the PBC algorithm is a fuzzy clustering algorithm.

[0047] Step 2.3: Change the cluster center spacing or cluster side length, recalculate the total positioning time and average positioning error. The new combination of cluster center spacing and cluster side length still needs to meet the conditions in Step 2.2, and should not repeat the previous clustering cases.

[0048] Step 2.4: Repeat step 2.3 until all combinations of class center spacing and class side length that meet the conditions have been traversed;

[0049] Step 2.5: Compare the total positioning time of all cases where the average positioning error is not greater than that of the case without clustering, and select the case with the smallest time as the final clustering result. If the average positioning error of all cases is greater than that of the case without clustering, then select the case with the smallest average positioning error as the final clustering result.

[0050] Step 3: Use the weighted K-nearest neighbor matching algorithm to match the compressed fingerprint FCPV of the point to be located with the compressed fingerprint FCPV in the fingerprint database for online localization.

[0051] Step 3.1: Using the fingerprint and coordinate data of all reference points in the established fingerprint database, determine the optimal K value of the weighted K-nearest neighbor algorithm using cross-validation.

[0052] Step 3.2: Calculate the Euclidean distance from the compressed fingerprint FCPV of the point to be located to all the compressed fingerprint FCPVs of the class centers, and assign the point to be located to the class whose fingerprint is closest to the fingerprint of the class center;

[0053] Step 3.3: In this class, the weighted K-nearest neighbor matching algorithm is used to estimate the coordinates of the point to be located, as shown in the formula: In the formula, Represents the coordinates of the point to be located. w represents the coordinates of the k-th reference point. k d represents the weight corresponding to the k-th reference point. k It is the fingerprint distance between the k-th reference point and the point to be located. Δd is a very small positive number to prevent the denominator from being zero; it can be taken as 10. -6 .

[0054] like Figure 3As shown, under the same experimental conditions, the PBC algorithm of this invention is compared with the most commonly used K-means clustering algorithm. The experimental results show that the average positioning error of the method of this invention is exactly the same as that of the case without clustering, while the average positioning error of K-means clustering is significantly larger than that of the case without clustering. In addition, the PBC method of this invention can reduce the total positioning time from 360.65 seconds in the case without clustering to 9.98 seconds, which is on the same order of magnitude as 7.11 seconds of K-means clustering.

Claims

1. A location-based clustering method for single-station terminal fingerprint localization, characterized in that, Includes the following steps: Step 1: Extract the fingerprint matrix FRCPM of the reference points, compress the fingerprint FRCPM, and store the position coordinates of all reference points in the area to be located and their compressed fingerprints FCPV one by one to obtain the fingerprint database. Step 2: Perform location-based clustering (PBC) on the data in the fingerprint database based on physical location features. The specific steps are as follows: Step 2.1: Using the test dataset collected in the offline phase, calculate the average localization error in the absence of clustering in the fingerprint database; Step 2.2: Initialize the cluster center spacing and cluster side length. Calculate the total positioning time and average positioning error under this clustering condition using the test set. The combination of cluster center spacing and cluster side length needs to satisfy that the set of all classes can cover the entire area to be located, and there can be overlapping parts between different classes, that is, a reference point can belong to several different classes at the same time. Step 2.3: Change the cluster center spacing or cluster side length, recalculate the total positioning time and average positioning error. The new combination of cluster center spacing and cluster side length still needs to meet the conditions in step 2.2, and should not repeat the previous clustering cases. Step 2.4, repeat step 2.3, until all combinations of class center spacing and class side length that meet the conditions have been traversed; Step 2.5: Compare the total positioning time of all cases where the average positioning error is not greater than that of the case without clustering, and select the case with the shortest time as the final clustering result. If the average positioning error of all cases is greater than that of the case without clustering, then select the case with the smallest average positioning error as the final clustering result. Step 3: Extract the fingerprint matrix FRCPM of the point to be located, and compress the fingerprint matrix FRCPM of the point to be located to obtain the compressed fingerprint FCPV of the point to be located; match the compressed fingerprint FCPV of the point to be located with the compressed fingerprint FCPV in the fingerprint database for online localization.

2. A location-based clustering method for single-station terminal fingerprint positioning according to claim 1, characterized in that, The reference point refers to dividing the area to be located into a uniform grid, with the grid nodes serving as reference points.

3. A location-based clustering method for single-station terminal fingerprint positioning according to claim 1, characterized in that, Step 1 involves extracting the reference point fingerprint matrix FRCPM and compressing the fingerprint FRCPM. The specific steps are as follows: Step 1.1: Record the coordinates of each reference point and collect the uplink signal of the terminal device at each reference point. Perform Hadamard product on the CFR matrix and its conjugate matrix of each reference point to obtain the fingerprint FRCPM of each reference point. Step 1.2: Sum the elements of each column of the fingerprint matrix FRCPM to obtain a matrix with the following number of elements. The row vector is the compressed fingerprint FCPV; The first k The compressed fingerprint FCPV of each reference point is represented as follows: ,but The j element Represented as: 。 4. A location-based clustering method for single-station terminal fingerprint positioning according to claim 1, characterized in that, Step 3 involves matching the compressed fingerprint FCPV of the point to be located with the compressed fingerprint FCPV in the fingerprint database for online localization. The specific details are as follows: Step 3.1: Using the fingerprint and coordinate data of all reference points in the established fingerprint database, determine the optimal K value for the weighted K-nearest neighbor algorithm; Step 3.2: Calculate the Euclidean distance from the compressed fingerprint FCPV of the point to be located to all class center compressed fingerprint FCPVs, and assign the point to be located to the class whose fingerprint is closest to the class center fingerprint. Step 3.3: In this class, the weighted K-nearest neighbor matching algorithm is used to estimate the coordinates of the point to be located, as shown in the formula: In the formula, Represents the coordinates of the point to be located. Representing the k The coordinates of the reference points Representing the k The weights corresponding to each reference point It is the first k The fingerprint distance between a reference point and the point to be located It is a very small positive number.

5. A location-based clustering method for single-station terminal fingerprint positioning according to claim 4, characterized in that, In step 3, cross-validation is used to determine the optimal K value for the weighted K-nearest neighbor algorithm.

Images