Direct path distinguishing method based on adaptive parameter DBSCAN clustering

Abstract

The invention provides a direct path distinguishing method based on adaptive parameter DBSCAN clustering. Firstly, a distance distribution matrix is calculated, and an average nearest neighbor distance matrix is obtained through sorting and averaging operation; on the basis, a Savitzky-Golay filter is used for filtering a normalized distribution probability histogram obtained according to the distance matrix. Secondly, searching by adopting a peak searching algorithm to obtain an interval of an optimal neighborhood radius, and calculating a minimum point number corresponding to the interval of the optimal neighborhood radius by utilizing a mathematical expectation method; thirdly, DBSCAN clustering is carried out according to the neighborhood radius and the minimum point number, a contour coefficient is calculated, and a clustering result corresponding to the maximum value of the contour coefficient is selected; and finally, a maximum likelihood function is calculated according to the point number, the average value, the variance and the ToF features of the clusters, and then the direct path is distinguished. According to the method, under the condition that the number of clusters, the neighborhood radius and the minimum point number in the neighborhood radius do not need to be determined, clustering is carried out according to the characteristics of data, and the direct path distinguishing capacity of a system is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of active indoor WiFi positioning, and specifically is a direct path resolution method based on adaptive parameter DBSCAN clustering, which realizes the resolution of direct path by processing angle of arrival (AoA) and time of flight (ToF) data. Background Art

[0002] With the rapid development of modern science and technology, indoor positioning is of great significance in many fields such as precise navigation, security management, and equipment tracking. In the development of indoor positioning technology, WiFi positioning has become a research focus due to its wide coverage and ready-made infrastructure. In the early stage, traditional indoor positioning technology mainly used the received signal strength indicator (RSSI) to estimate the distance between the device and the signal source by measuring and analyzing the physical characteristics of the wireless signal to achieve positioning. However, this method is greatly affected by obstacles, multipath effects, and interference in the environment in practical applications, and the positioning accuracy is relatively low. Subsequently, the researchers extracted the channel state information (CSI) by modifying the firmware to achieve higher-precision positioning. Thanks to the Multiple-Input Multiple-Output (MIMO) technology and Orthogonal Frequency Division Multiplexing (OFDM) technology, CSI has richer channel information and stronger anti-multipath effect ability than RSSI, so it performs better in wireless positioning, especially indoor positioning. In addition, CSI has the ability to distinguish different paths, allowing the positioning system to more accurately identify and process multipath signals in complex environments, thereby achieving higher-precision positioning.

[0003] In the field of wireless technology, indoor positioning systems are roughly divided into two categories: fingerprint positioning solutions based on deep learning and positioning methods based on multidimensional parameter estimation. Due to the high cost of deployment, maintenance and resource consumption of fingerprint positioning solutions based on deep learning, as well as its sensitivity to environmental changes, its practical application is limited. However, the positioning method based on multidimensional parameter estimation uses a variety of parameters to reduce the dependence of the positioning system on environmental changes, and has low deployment costs, so it has gradually become the mainstream. However, due to the indoor multipath effect of the existing positioning method based on multidimensional parameter estimation, in addition to the AoA and ToF parameters of the mixed direct path and reflection path, there are also a small number of outliers in the estimation results. Therefore, it is inappropriate to simply regard the AoA parameter with a smaller ToF value as a direct path parameter. To solve the above problems, the present invention proposes a direct path resolution method based on adaptive parameter DBSCAN clustering. First, the distance distribution matrix is ​​calculated and the average nearest neighbor distance matrix is ​​obtained by sorting and averaging operations. On this basis, the normalized distribution probability histogram obtained according to the distance matrix is ​​filtered using a Savitzky-Golay filter. Secondly, the peak-finding algorithm is used to search for the interval of the optimal neighborhood radius, and the mathematical expectation method is used to calculate the minimum number of points corresponding to the interval of the optimal neighborhood radius. Again, DBSCAN clustering is performed according to the neighborhood radius and the minimum number of points, and the silhouette coefficient is calculated, and the clustering result corresponding to the maximum value is selected. Finally, the maximum likelihood function is calculated based on the number of points, mean value, variance and ToF characteristics of the cluster, and then the direct diameter is identified. The present invention performs clustering according to the characteristics of the data itself without determining the number of clusters, the neighborhood radius and the minimum number of points within the neighborhood radius, thereby improving the system's ability to resolve direct diameters. Summary of the invention

[0004] The purpose of the present invention is to provide a method for distinguishing direct paths based on adaptive parameter DBSCAN clustering in a WiFi positioning system. Clustering is performed according to the characteristics of the data itself without determining the number of clusters, the neighborhood radius and the minimum number of points within the neighborhood radius, thereby improving the system's ability to distinguish direct paths.

[0005] The present invention discloses a method for distinguishing direct paths based on adaptive parameter DBSCAN clustering, which specifically comprises the following steps:

[0006] Step 1: For the CSI data collected on the array antenna using commercial WiFi devices such as ASUS RT-AC86U or hosts equipped with Intel 5300 network cards, the AoA and ToF parameters of the transmitter are estimated using subspace algorithms such as Matrix Pencil (MP). The set of raw estimation results is expressed as:

[0007] Est={(θ i ,τi )i∈{1,2,...,I}}

[0008] Among them, θ i and τ i They represent the AoA and ToF parameters of data point i (1≤i≤I), and I is the total number of data points. Due to the different scales of AoA and ToF, they need to be normalized before clustering:

[0009]

[0010] Where X is the estimated value of AoA {θ i} or ToF estimate {τ i}, X i represents the i-th (1≤i≤I) element in the set X, and the normalized parameter set is recorded as:

[0011]

[0012] Step 2: The average nearest neighbor distance represents the distance distribution characteristics of the data points. First, calculate the distance distribution matrix:

[0013] D={Distance(u,v)u,v∈{1,2,...,I}}

[0014] Where Distance(u,v) represents the set Est nor The Euclidean distance from data point u to data point v is:

[0015]

[0016] Among them, (x u ,y u ) and (x v ,y v ) represent the normalized coordinates of data point u and data point v respectively.

[0017] Then, sort each row of the distance matrix D in ascending order to obtain D sort :

[0018]

[0019] Secondly, calculate the average of the vth (1≤v≤I) column in D and get the average nearest neighbor distance of all data points

[0020]

[0021] Among them, the vth (1≤v≤I) element Represents the average distance from any data point to the nearest vth data point.

[0022] Step 3: Determine the candidate neighborhood radius interval and the minimum number of points corresponding to the candidate neighborhood radius. Draw a distribution probability histogram:

[0023]

[0024] Among them, z represents the set of the left boundary of the histogram, b represents the interval width in the histogram, and f(z) represents the frequency, that is, The number of elements in that fall within the interval [z,z+b).

[0025] Secondly, the histogram H(z) is smoothed by Savitzky-Golay filter to obtain the fitting curve to reduce the number of false peaks in the histogram. Thirdly, the position of the first peak point of the fitting curve is obtained by the peak finding algorithm, and the interval where it is located is regarded as the optimal neighborhood radius interval.

[0026] Finally, the interval is divided by the minimum interval Δeps to obtain a set of E candidate neighborhood radii eps = {eps1, eps2, ..., eps E}, and then use the mathematical expectation method to find the minimum number of points corresponding to each neighborhood radius in eps:

[0027]

[0028] Among them, p(eps e ,i) represents the neighborhood radius eps e The number of neighborhood points of the i-th data point under the condition .

[0029] Step 4: Perform DBSCAN clustering based on eps and minpts and calculate the silhouette coefficient {s u}:

[0030]

[0031] Among them, u and v represent the index of the data point in the cluster, q and w represent the sequence number of the cluster, and J q and J w Respectively represent the set of data point indices in cluster numbers q and v, n q and n w Respectively represent the total number of data points in cluster numbers q and v. u Indicates the data point u to its cluster J q The average Euclidean distance of other data points v in (u≠v), y u Indicates that the data point u is transferred to other clusters J wThe minimum value of the average Euclidean distance of the inner data point v (w≠q). In order to select the optimal neighborhood radius and the minimum number of points in the neighborhood radius, calculate the average silhouette coefficient:

[0032]

[0033] The average silhouette coefficient of the DBSCAN clustering results obtained by different neighborhood radius sets eps and different minimum point number sets minpts in the neighborhood radius is Select the clustering result corresponding to the maximum value in set s

[0034] Step 5: Calculate the maximum likelihood function l for the kth (1≤k≤K) cluster in the clustering result obtained in step 4 according to the following formula: k :

[0035] l k =exp(ω N N k -ω A σ A,k -ω T σ T,k -ω m m T,k )

[0036] Among them, N k represents the total number of data points in the kth cluster, σ A,k represents the variance of the AoA dimension of the data points in the kth cluster, σ T,k represents the variance of the ToF dimension of the data points in the kth cluster, m T,k Represents the average value of the ToF dimension of the data points in the kth cluster. ω N ,ω A ,ω T and ω m are all constants and represent the corresponding weights.

[0037] Finally, by calculating Get the cluster number k * , that is, the kth cluster in the clustering result * The AoA and ToF parameters of each cluster represent the resolved direct path, and the centroid of the direct path data point is calculated using the following formula

[0038]

[0039] Where u represents the index of the data point in the cluster, Indicates cluster number k * The collection of data point indices in , Indicates cluster number k * The total number of data points in .

[0040] Beneficial Effects

[0041] The present invention proposes a direct path resolution method based on adaptive parameter DBSCAN clustering. First, the distance distribution matrix is ​​calculated and the average nearest neighbor distance matrix is ​​obtained by sorting and averaging operations. On this basis, the normalized distribution probability histogram obtained according to the distance matrix is ​​filtered using a Savitzky-Golay filter. Secondly, the peak search algorithm is used to search for the interval of the optimal neighborhood radius, and the mathematical expectation method is used to calculate the minimum number of points corresponding to the interval of the optimal neighborhood radius. Thirdly, DBSCAN clustering is performed according to the neighborhood radius and the minimum number of points, and the silhouette coefficient is calculated, and the clustering result corresponding to the maximum value is selected. Finally, the maximum likelihood function is calculated based on the number of points, mean value, variance and ToF characteristics of the cluster to identify the direct path.

[0042] The present invention performs clustering according to the characteristics of the data itself without determining the number of clusters, the neighborhood radius and the minimum number of points within the neighborhood radius, thereby improving the system's ability to resolve direct paths. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 The block diagram of the direct path resolution method based on adaptive parameter DBSCAN clustering of the present invention is Specific implementation plan

[0044] The purpose of the present invention is to provide a method for distinguishing direct paths based on adaptive parameter DBSCAN clustering in a WiFi positioning system. Clustering is performed according to the characteristics of the data itself without determining the number of clusters, the neighborhood radius and the minimum number of points within the neighborhood radius, thereby improving the system's ability to distinguish direct paths.

[0045] The present invention discloses a method for distinguishing direct paths based on adaptive parameter DBSCAN clustering, and the specific implementation scheme includes the following steps:

[0046] Step 1: For the CSI data collected on the array antenna using commercial WiFi devices such as ASUS RT-AC86U or hosts equipped with Intel 5300 network cards, the AoA and ToF parameters of the transmitter are estimated using subspace algorithms such as Matrix Pencil (MP). The set of raw estimation results is expressed as:

[0047] Est={(θ i ,τ i )i∈{1,2,...,I}}

[0048] Among them, θ i and τ iThey represent the AoA and ToF parameters of data point i (1≤i≤I), and I is the total number of data points. Due to the different scales of AoA and ToF, they need to be normalized before clustering:

[0049]

[0050] Where X is the estimated value of AoA {θ i} or ToF estimate {τ i}, X i represents the i-th (1≤i≤I) element in the set X, and the normalized parameter set is recorded as:

[0051]

[0052] Step 2: The average nearest neighbor distance represents the distance distribution characteristics of the data points. First, calculate the distance distribution matrix:

[0053] D={Distance(u,v)u,v∈{1,2,...,I}}

[0054] Where Distance(u,v) represents the set Est nor The Euclidean distance from data point u to data point v is:

[0055]

[0056] Among them, (x u ,y u ) and (x v ,y v ) represent the normalized coordinates of data point u and data point v respectively.

[0057] Then, sort each row of the distance matrix D in ascending order to obtain D sort :

[0058]

[0059] Secondly, calculate the average of the vth (1≤v≤I) column in D and get the average nearest neighbor distance of all data points

[0060]

[0061] Among them, the vth (1≤v≤I) element Represents the average distance from any data point to the nearest vth data point.

[0062] Step 3: Determine the candidate neighborhood radius interval and the minimum number of points corresponding to the candidate neighborhood radius. Draw a distribution probability histogram:

[0063]

[0064] Among them, z represents the set of the left boundary of the histogram, b represents the interval width in the histogram, and f(z) represents the frequency, that is, The number of elements in that fall within the interval [z,z+b).

[0065] Secondly, the histogram H(z) is smoothed by Savitzky-Golay filter to obtain the fitting curve to reduce the number of false peaks in the histogram. Thirdly, the position of the first peak point of the fitting curve is obtained by the peak finding algorithm, and the interval where it is located is regarded as the optimal neighborhood radius interval.

[0066] Finally, the interval is divided by the minimum interval Δeps to obtain a set of E candidate neighborhood radii eps = {eps1, eps2, ..., eps E}, and then use the mathematical expectation method to find the minimum number of points corresponding to each neighborhood radius in eps:

[0067]

[0068] Among them, p(eps e ,i) represents the neighborhood radius eps e The number of neighborhood points of the i-th data point under the condition .

[0069] Step 4: Perform DBSCAN clustering based on eps and minpts and calculate the silhouette coefficient {s u}:

[0070]

[0071] Among them, u and v represent the index of the data point in the cluster, q and w represent the sequence number of the cluster, and J q and J w Respectively represent the set of data point indices in cluster numbers q and v, n q and n w Respectively represent the total number of data points in cluster numbers q and v. u Indicates the data point u to its cluster J q The average Euclidean distance of other data points v in (u≠v), y u Indicates that the data point u is transferred to other clusters J w The minimum value of the average Euclidean distance of the inner data point v (w≠q). In order to select the optimal neighborhood radius and the minimum number of points in the neighborhood radius, calculate the average silhouette coefficient:

[0072]

[0073] The average silhouette coefficient of the DBSCAN clustering results obtained by different neighborhood radius sets eps and different minimum point number sets minpts in the neighborhood radius is Select the clustering result corresponding to the maximum value in set s

[0074] Step 5: Calculate the maximum likelihood function l for the kth (1≤k≤K) cluster in the clustering result obtained in step 4 according to the following formula: k :

[0075] l k =exp(ω N N k -ω A σ A,k -ω T σ T,k -ω m m T,k )

[0076] Among them, N k represents the total number of data points in the kth cluster, σ A,k represents the variance of the AoA dimension of the data points in the kth cluster, σ T,k represents the variance of the ToF dimension of the data points in the kth cluster, m T,k Represents the average value of the ToF dimension of the data points in the kth cluster. ω N ,ω A ,ω T and ω m are all constants and represent the corresponding weights.

[0077] Finally, by calculating Get the cluster number k * , that is, the kth cluster in the clustering result * The AoA and ToF parameters of each cluster represent the resolved direct path, and the centroid of the direct path data point is calculated using the following formula

[0078]

[0079] Where u represents the index of the data point in the cluster, Indicates cluster number k * The collection of data point indices in , Indicates cluster number k * The total number of data points in .

Claims

1. A straight path resolution method based on adaptive parameter DBSCAN clustering, comprising the following steps: Step 1: For the CSI data collected on the array antenna using commercial WiFi devices such as ASUS RT-AC86U or hosts equipped with Intel 5300 network cards, the AoA and ToF parameters of the transmitter are estimated using subspace algorithms such as Matrix Pencil (MP). The set of raw estimation results is expressed as: Is={(θ i ,τ i )|i∈{1,2,...,I}} Among them, θ i and τ i They represent the AoA and ToF parameters of data point i (1≤i≤I), and I is the total number of data points. Due to the different scales of AoA and ToF, they need to be normalized before clustering: Where X is the estimated value of AoA {θ i } or ToF estimate {τ i }, X i represents the i-th (1≤i≤I) element in the set X, and the normalized parameter set is recorded as: Step 2: The average nearest neighbor distance represents the distance distribution characteristics of the data points. First, calculate the distance distribution matrix: D={Distance(u,v)|u,v∈{1,2,...,I}} Then, sort each row of the distance matrix D in ascending order to obtain D sort . Secondly, calculate the average of the vth (1≤v≤I) column in D and get the average nearest neighbor distance of all data points Step 3: Determine the candidate neighborhood radius interval and the minimum number of points corresponding to the candidate neighborhood radius. Draw a distribution probability histogram. Secondly, the histogram is smoothed by Savitzky-Golay filter to obtain a fitting curve to reduce the number of false peaks in the histogram. Thirdly, the position of the first peak point of the fitting curve is obtained by the peak finding algorithm, and the interval where it is located is regarded as the optimal neighborhood radius interval. Finally, the interval is divided by the minimum interval Δeps to obtain a set of E candidate neighborhood radii eps={eps1,eps2,...,eps E }, and then use the mathematical expectation method to find the minimum number of points corresponding to each neighborhood radius in eps: Among them, p(eps e ,i) represents the neighborhood radius eps e The number of neighborhood points of the i-th data point under the condition . Step 4: Perform DBSCAN clustering based on eps and minpts and calculate the silhouette coefficient {s u }: Among them, u and v represent the index of the data point in the cluster, q and w represent the sequence number of the cluster, and J q and J w Respectively represent the set of data point indices in cluster numbers q and v, n q and n w Respectively represent the total number of data points in cluster numbers q and v. u Indicates the data point u to its cluster J q The average Euclidean distance of other data points v in (u≠v), y u Indicates that the data point u is transferred to other clusters J w The minimum value of the average Euclidean distance of the inner data point v (w≠q). In order to select the optimal neighborhood radius and the minimum number of points in the neighborhood radius, calculate the average silhouette coefficient: The average silhouette coefficient of the DBSCAN clustering results obtained by different neighborhood radius sets eps and different minimum point number sets minpts in the neighborhood radius is Select the clustering result corresponding to the maximum value in set s Step 5: Calculate the maximum likelihood function l for the kth (1≤k≤K) cluster in the clustering result obtained in step 4 according to the following formula: k : l k =exp(ω N N k -oh A s A,k -oh T s T,k -oh m m T,k ) Among them, N k represents the total number of data points in the kth cluster, σ A,k represents the variance of the AoA dimension of the data points in the kth cluster, σ T,k represents the variance of the ToF dimension of the data points in the kth cluster, m T,k Represents the average value of the ToF dimension of the data points in the kth cluster. ω N ,ω A ,ω T and ω m are all constants and represent the corresponding weights. Finally, by calculating Get the cluster number k * , that is, the kth cluster in the clustering result * The AoA and ToF parameters of each cluster represent the resolved direct path, and the centroid of the direct path data point is calculated using the following formula Where u represents the index of the data point in the cluster, Indicates cluster number k * The collection of data point indices in , Indicates cluster number k * The total number of data points in .

Images