Channel map positioning method based on beam domain channel covariance matrix
Through the channel map positioning method based on the beam domain channel covariance matrix, the problems of low positioning accuracy and high maintenance cost in NLoS scenarios are solved, and the positioning effect of 6G high precision and low latency is achieved.
Patent Information
- Application Number
- CN202510491442.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-18
- Publication Date
- 2025-08-08
AI Technical Summary
The existing positioning method has low positioning accuracy and high maintenance cost in NLoS scenarios, making it difficult to meet the positioning requirements of 6G with high precision and low latency.
The channel map positioning method based on the beam domain channel covariance matrix is adopted. By acquiring the data set, extracting the beam domain channel covariance matrix, cosine similarity and global difference measurement matrix are calculated, and sample location is predicted using channel map generation and positioning algorithms, and positioning accuracy is optimized in combination with semi-supervised learning.
Maintain high positioning accuracy under NLoS conditions, reduce the number of anchor points and maintenance costs, enhance the robustness and applicability of positioning, and achieve 6G high-precision and low-latency positioning.
Smart Images
Figure CN120455930A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of wireless communication technology, and in particular to a channel map positioning method based on a beam domain channel covariance matrix. Background Art
[0002] With advancements in sixth-generation wireless communication technology (6G), future wireless communication technologies are expected to support emerging applications such as extended reality, drones, and intelligent driving, all of which require highly accurate positioning. Geometry-based positioning methods (GPM) can achieve centimeter-level accuracy when only considering the direct light of sight (LoS) path between the transmitter and receiver. However, maintaining LoS conditions is often challenging in complex environments, especially indoors. In contrast, fingerprint-based positioning methods (FPM) can achieve reliable positioning accuracy even in non-light of sight (NLoS) scenarios. However, FPM uses channel measurements or simulations to build a channel fingerprint database and estimates position by comparing channel data from user equipments (UEs) with entries in the database. Therefore, FPM requires an expensive ground-truth reference system for accurate labeling, and environmental changes may require frequent updates, increasing deployment and maintenance costs.
[0003] Channel map-based positioning methods, as an alternative positioning technique, address the limitations of FPM. By leveraging unsupervised learning, they map high-dimensional channel state information (CSI) into low-dimensional virtual coordinates, effectively capturing the spatial geometry around the UE while reducing reliance on expensive location labels. Channel maps rely on unsupervised dimensionality reduction. Traditional methods include principal component analysis (PCA), Sammon mapping, multidimensional scaling, and isometric mapping. Deep neural network (DNN)-based methods are more flexible and effective in capturing complex nonlinear relationships in high-dimensional data. Autoencoders (AEs) achieve data dimensionality reduction through an unsupervised reconstruction mechanism without relying on explicit difference metrics. Deep neural networks based on Siamese architectures utilize difference metrics between pairs of samples for data dimensionality reduction. Triplet-based DNNs, based on difference metrics, enforce geometric consistency between neighboring samples in the original space in the latent space through triplet sample constraints. They focus solely on the relative values of the difference metrics between samples, rather than their absolute values.
[0004] Current positioning methods have problems such as high database maintenance costs and low positioning accuracy in NLoS scenarios. Summary of the Invention
[0005] The technical problem to be solved by the present invention is to overcome the shortcomings of the existing technology and provide a channel map positioning method based on the beam domain channel covariance matrix. The present invention can achieve 6G high-precision and low-latency positioning.
[0006] The present invention adopts the following technical solutions to solve the above technical problems:
[0007] A channel map positioning method based on a beam-domain channel covariance matrix proposed in the present invention includes:
[0008] Step 1: Obtain a data set for positioning and divide it into a training set and a validation set. The data set includes the CSI matrix and the real coordinates of the anchor points.
[0009] Extract the beam domain channel covariance matrix based on the CSI matrix in the data set;
[0010] Step 2: Calculate the cosine similarity based on the beam domain channel covariance matrix to obtain the channel covariance cosine similarity matrix, and obtain the global difference measurement matrix based on the channel covariance cosine similarity matrix;
[0011] Step 3: The beam domain channel covariance matrix and the global difference metric matrix are used to generate a channel map using a channel map generation algorithm. The channel map uses a channel map positioning algorithm to predict the position information of the sample to be positioned.
[0012] Step 4: Calculate the predicted distance matrix using the predicted position information of the sample to be located, and combine it with the global difference measurement matrix to obtain the updated global difference measurement matrix;
[0013] Step 5: Repeat steps 3-4 until the performance of the validation set no longer improves, and obtain the final location information of the sample to be located.
[0014] As a further optimization scheme of the channel map positioning method based on the beam domain channel covariance matrix described in the present invention, the CSI matrix is collected using wireless channel simulation software or wireless channel measurement equipment based on ray tracing. The CSI matrix includes channel state information, Gaussian white noise and time advance for each receiving antenna, each transmitting antenna and each subcarrier.
[0015] As a further optimization scheme of the channel map positioning method based on the beam-domain channel covariance matrix described in the present invention, in step 1, the beam-domain channel covariance matrix is extracted according to the CSI matrix in the data set; including:
[0016] Step 1.1: Perform covariance calculation on the CSI matrix to obtain the channel covariance matrix;
[0017] Step 1.2: Perform a two-dimensional Fourier transform on the channel covariance matrix to obtain the beam-domain channel covariance matrix.
[0018] As a further optimization scheme of the channel map positioning method based on the beam domain channel covariance matrix described in the present invention, the calculation method of the global difference metric matrix in step 2 is:
[0019] Step 2.1, from the i-th sample x in the dataset i First, calculate the channel covariance cosine similarity matrix D of its nearest k samples;
[0020] Step 2.2: Construct k-nearest neighbor graph G based on D k-NN ;
[0021] Step 2.3: Apply the shortest path algorithm to G k-NN Get the matrix D geo ';
[0022] Step 2.4: Repeat steps 2.1 to 2.3 for all samples to be located, and finally obtain D geo ' is the global difference measurement matrix D geo .
[0023] As a further optimization scheme of the channel map positioning method based on the beam domain channel covariance matrix described in the present invention,
[0024] Channel map generation algorithm: The samples in the training set are divided into anchor samples, positive samples, and negative samples. The samples in the training set are input into the Triplet deep neural network for training to generate a channel map. The difference metric between the anchor samples and the positive samples is smaller than the difference metric between the anchor samples and the negative samples. The goal of the channel map generation algorithm is to optimize the neural network parameters so that the channel map distance between the anchor samples and the positive samples is smaller than the channel map distance between the anchor samples and the negative samples.
[0025] As a further optimization scheme of the channel map positioning method based on the beam domain channel covariance matrix described in the present invention, the channel map positioning algorithm: using the channel map generated by the channel map generation algorithm, using the semi-supervised learning method to transform the Triplet loss function L Triplet The actual positioning coordinate error L with the anchor point Position Combined, gradually improve L during the training process Position The proportion of the channel map positioning algorithm is calculated until the channel map positioning algorithm converges, and finally the predicted position information of each sample is obtained.
[0026] As a further optimization scheme of the channel map positioning method based on the beam domain channel covariance matrix described in the present invention, step 4 includes:
[0027] Step 4.1: Use the position information of the samples to be located to calculate the predicted distance matrix D between the samples to be located P ;
[0028] Step 4.2: Use the predicted distance matrix D P Perform global geodesic algorithm to obtain the global prediction distance matrix D geo,P ;
[0029] Step 4.3: The global prediction distance matrix D geo,P and the global difference measure matrix D geo The updated global difference measurement matrix D is obtained by fusion renew .
[0030] As a further optimization scheme of the channel map positioning method based on the beam domain channel covariance matrix described in the present invention,
[0031] The Triplet loss function L Triplet The actual positioning coordinate error L with the anchor point Position Combined; specifically as follows:
[0032] Fusion Triplet loss function LTriplet The actual positioning coordinate error L with the anchor point Position , we get the neural network loss function Loss = αL Position +(1-α)L Triplet , where α is the error ratio of the real positioning coordinates of the anchor point.
[0033] As a further optimization scheme of the channel map positioning method based on the beam domain channel covariance matrix described in the present invention,
[0034]
[0035] Among them, f θ (·) is the channel map generation algorithm, M is the margin, T is the triplet sample set formed according to the generated channel map, x i ,x j ,x k are the i-th, j-th, and k-th samples taken from the Triplet sample set T;
[0036]
[0037] Among them, g θ (·) is the channel map positioning algorithm, P is the anchor point set, x p is the p-th anchor point sample, y p is the real coordinate of the p-th anchor point.
[0038] A computer device includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor executes the computer program, the steps of the channel map positioning method based on the beam domain channel covariance matrix are implemented as described above.
[0039] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:
[0040] (1) The present invention proposes a channel map positioning method based on the beam domain channel covariance matrix. As a positioning algorithm based on a channel map, the present invention has better positioning accuracy under NLoS and low signal-to-noise ratio conditions compared to GPM; compared to FPM, it can significantly reduce the number of anchor points and reduce the manpower and material resources consumed in anchor point measurement.
[0041] (2) The beam-domain channel covariance matrix, global difference metric, and difference metric update algorithm proposed in the present invention reduce the impact of noise and time advance on positioning in actual situations, ensure the robustness of the channel map, and enhance the applicability to real-world positioning.
[0042] (3) The present invention uses a channel map generation algorithm and a channel map positioning algorithm based on semi-supervised learning, which are superior to other positioning algorithms based on channel maps.
[0043] (4) In summary, the present invention is an effective positioning method, which provides conditions for further realizing 6G high-precision and low-latency positioning. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] Figure 1 This is a model framework diagram of a channel map positioning method based on a beam-domain channel covariance matrix provided in Example 1 of the present invention;
[0045] Figure 2 Schematic diagram of the global difference measurement size from any sample to all samples in Example 1 of the present invention;
[0046] Figure 3 Schematic diagram of the algorithm structure of the channel map generation algorithm in Example 1 of the present invention;
[0047] Figure 4 Schematic diagram of a Cumulative Distribution Function (CDF) curve of the mean absolute error of channel map positioning based on different features in Example 1 of the present invention;
[0048] Figure 5 Schematic diagram of CDF curves of mean absolute errors of different channel map positioning methods in Example 1 of the present invention;
[0049] Figure 6 Schematic diagrams of positioning results based on different channel map positioning methods in Example 1 of the present invention, where (a) is the PCA positioning result diagram, (b) is the AE positioning result diagram, (c) is the Siamese network positioning result diagram, (d) is the traditional Triplet positioning result diagram, (e) is the result diagram of the proposed channel map positioning algorithm, and (f) is the true value coordinate diagram of the sample to be positioned;
[0050] Figure 7 This is a flow chart of a channel map positioning method based on the beam domain channel covariance matrix of the present invention. DETAILED DESCRIPTION
[0051] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
[0052] like Figure 7 As shown, a channel map positioning method based on the beam domain channel covariance matrix includes:
[0053] Step 1: Obtain a data set for positioning and divide it into a training set and a validation set. The data set includes the CSI matrix and the real coordinates of the anchor points.
[0054] Extract the beam domain channel covariance matrix based on the CSI matrix in the data set;
[0055] Step 2: Calculate the cosine similarity based on the beam domain channel covariance matrix to obtain the channel covariance cosine similarity matrix, and obtain the global difference measurement matrix based on the channel covariance cosine similarity matrix;
[0056] Step 3: The beam domain channel covariance matrix and the global difference metric matrix are used to generate a channel map using a channel map generation algorithm. The channel map uses a channel map positioning algorithm to predict the position information of the sample to be positioned.
[0057] Step 4: Calculate the predicted distance matrix using the predicted position information of the sample to be located, and combine it with the global difference measurement matrix to obtain the updated global difference measurement matrix;
[0058] Step 5: Repeat steps 3-4 until the performance of the validation set no longer improves, and obtain the final location information of the sample to be located.
[0059] Example 1:
[0060] See also Figure 1 , this embodiment provides a channel map positioning method based on the beam domain channel covariance matrix, comprising the following steps:
[0061] Step 1: Obtain the data set used for positioning, including the transmitter antenna array parameters, the receiver antenna array parameters, the CSI matrix, and the real coordinates of the anchor points, and divide them into a training set and a validation set.
[0062] Specifically, in this embodiment, a Multiple Input Multiple Output (MIMO) system is used, in which the base station is equipped with a 64-element millimeter wave communication array. The system adopts a dual-polarization design, including 8 columns of single-polarization elements, each with 4 channels. The horizontal spacing between the channels is d h =0.5λ, vertical spacing is d v = 2.0λ, where λ represents the wavelength. An orthogonal frequency division multiplexing system operating in the 3.5 GHz band with a subcarrier spacing of 30 kHz is considered. The CSI matrix is generated using ray tracing-based wireless channel simulation software. This embodiment includes a total of 20,000 samples, including 19,500 samples to be located and 500 anchor points. 400 anchor points are divided into a training set, and 100 anchor points are divided into a validation set.
[0063] Step 2: Extract the beam-domain channel covariance matrix according to the CSI matrix.
[0064] The specific steps are as follows:
[0065] Step 201: Perform covariance calculation based on the CSI matrix to obtain the channel covariance matrix R:
[0066]
[0067] Where t = 1, 2, ..., N T ,r,r′=1,2,...,N R , N T ,N R are the number of transmitting antennas and receiving antennas respectively, H(t,r,f k ) is the channel CSI matrix, (·) H is the conjugate transpose.
[0068] Step 202: Calculate the beam domain channel covariance matrix F according to the channel covariance matrix R:
[0069]
[0070] in, represents the two-dimensional Fourier transform, R(·,m,n) represents the element in the mth row and nth column of the channel covariance matrix, and F(·,z,a) represents the element in the zth row and ath column of the beam-domain channel covariance matrix.
[0071] Step 3: Calculate the channel covariance cosine similarity matrix based on the beam domain channel covariance matrix.
[0072] Calculate the channel covariance cosine similarity matrix through cosine similarity:
[0073]
[0074] in, represents the channel covariance cosine similarity between the i-th sample and the j-th sample, F (i) 、F (j) represent the beam-domain channel covariance matrices of the i-th sample and the j-th sample, respectively.
[0075] Step 4: Calculate the global difference measure matrix based on the channel covariance cosine similarity matrix.
[0076] The specific steps are as follows:
[0077] Step 401: From the i-th sample x in the data set i First, calculate the channel covariance cosine similarity matrix D of its nearest k samples;
[0078] Step 402: Construct a k-nearest neighbor graph G based on D k-NN ;
[0079] Step 403: Apply the shortest path algorithm to G k-NN Get the matrix D geo ';
[0080] Step 404: Repeat steps 401 to 403 for all samples to be located, and finally obtain D geo ' is the global difference measurement matrix D geo .
[0081] Specifically, the global difference measurement diagram of the present invention is shown in Figure 2 .
[0082] Step 5: Using the beam-domain channel covariance matrix and the global difference metric matrix to generate a 32-dimensional channel map using a channel map generation algorithm based on a Triplet neural network.
[0083] The specific steps are as follows:
[0084] Step 501: Randomly select anchor point samples x from the training set a ;
[0085] Step 502: Based on the global difference measurement matrix D geo Define the positive sample set of anchor points: in, represents the qth quantile of all similarity measures;
[0086] Step 503: Based on the global difference measurement matrix D geo Define the negative sample set of anchor points: Where, q′=1-q;
[0087] Step 504: Randomly select Select positive sample x p ,from Select negative samples x n ;
[0088] Step 505: Repeat steps 501 to 504 until N sample samples are selected to form the Triplet sample set;
[0089] Step 506: Input the Triplet sample set into the channel map generation algorithm to obtain a 32-dimensional channel map.
[0090] The schematic diagram of the channel map generation algorithm based on Triplet neural network is as follows: Figure 3 shown.
[0091] Step 6: Apply a channel map positioning algorithm to the 32-dimensional channel map to obtain the position information of the sample to be located.
[0092] The specific steps are as follows:
[0093] Step 601: Calculate the traditional Triplet loss function L according to the channel map result. Triplet :
[0094]
[0095] Among them, f θ (·) is the channel map generation algorithm, which can map the input sample to the channel map coordinate, M is the margin, x i ,x j ,x k are the i-th, j-th, and k-th samples taken from the Triplet sample set T.
[0096] Step 602: Calculate the true coordinate error L based on the predicted anchor point position information and the true position information. Position :
[0097]
[0098] Among them, g θ (·) represents the channel map positioning algorithm, which maps the channel map coordinates to the predicted position coordinates, P is the anchor point set, x p is the p-th anchor point sample, y p is the real coordinate of the p-th anchor point.
[0099] Step 603: Integrate the traditional Triplet loss function L Triplet and the true coordinate error L Position , we get the neural network loss function Loss = αL Position +(1-α)L Triplet ;
[0100] In this example, the initial value of α is 0.1 and gradually increases to 0.9 as the number of training rounds increases.
[0101] Step 603: Gradually improve the true coordinate error L during the training process Position The proportion of is calculated until the network converges, and finally the predicted coordinates of each sample are obtained.
[0102] Step 7: Use the predicted position information of each sample to calculate the distance information matrix of each sample, and combine it with the channel covariance cosine similarity matrix to calculate the updated difference measurement matrix. The specific steps are as follows:
[0103] Step 701: Calculate the predicted distance matrix D between the samples to be located using the position information P(i) of each sample obtained from the initial training. P :
[0104] D P (i,j)=‖P(i)-P(j)‖
[0105] Among them, D P (i,j) is the predicted distance matrix D P The element in the i-th row and j-th column, P(j) is the position information of the j-th sample;
[0106] Step 702: Using the predicted distance matrix D P The global geodesic algorithm is used to obtain the global prediction distance matrix D geo,P ;
[0107] Step 703: The global prediction distance matrix D geo,P and the global difference measure matrix D geo The updated global difference measurement matrix D is obtained by fusion renew :
[0108]
[0109] Among them D geo (i,j) is the global difference measurement matrix D geo The element at row i and column j.
[0110] like Figure 4 As shown in the figure, the proposed channel beam domain covariance matrix and global difference metric matrix have excellent performance. In the absence of environmental noise, the positioning performance of the channel beam domain covariance matrix and global difference metric matrix proposed in this invention is slightly better than that of traditional features. In the presence of environmental noise, the former performs much better than traditional features and is similar to the positioning results of traditional features in the absence of environmental noise.
[0111] like Figure 5 As shown in the figure, the performance of the proposed channel map positioning method is significantly improved compared to traditional channel map positioning methods. Compared with the Siamese network, AE, and PCA, the accuracy is improved by 72%, 75%, and 75%, respectively. 80% of the samples to be positioned have a positioning error below 20m.
[0112] like Figure 6 Schematic diagram of positioning results based on different channel map positioning methods in Example 1 of the present invention, wherein: Figure 6 (a) in the figure is the PCA positioning result diagram. Figure 6 (b) in the figure is the AE positioning result. Figure 6 (c) in the figure is the Siamese network positioning result map. Figure 6 (d) is the traditional Triplet positioning result diagram. Figure 6 (e) in the figure is the result of the proposed channel map positioning algorithm. Figure 6 (f) in is the true value coordinate diagram of the sample to be located. Figure 6 As shown in the figure, the proposed channel map positioning method based on the beam domain channel covariance matrix can not only correctly distinguish the samples to be positioned in different areas, but also better restore the shape structure of the samples to be positioned, with excellent performance.
[0113] In summary, the present invention establishes a channel map positioning method based on the beam domain channel covariance matrix. Through the calculation of the beam domain channel covariance matrix and the global difference metric, it can ultimately effectively realize the positioning of actual noisy communication scenarios, providing an important prerequisite for the realization of 6G high-precision and low-latency positioning.
[0114] A computer device includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor executes the computer program, the steps of the channel map positioning method based on the beam domain channel covariance matrix are implemented as described above.
[0115] Anything not described in detail in the present invention is well known to those skilled in the art.
[0116] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a technician familiar with this technical field within the technical scope disclosed by the present invention should be covered by the scope of protection of the present invention.
Claims
1. A channel map positioning method based on beam domain channel covariance matrix, characterized in that: include: Step 1: Obtain a data set for positioning and divide it into a training set and a validation set. The data set includes the CSI matrix and the real coordinates of the anchor points. Extract the beam domain channel covariance matrix based on the CSI matrix in the data set; Step 2: Calculate the cosine similarity based on the beam domain channel covariance matrix to obtain the channel covariance cosine similarity matrix, and obtain the global difference measurement matrix based on the channel covariance cosine similarity matrix; Step 3: The beam domain channel covariance matrix and the global difference metric matrix are used to generate a channel map using a channel map generation algorithm. The channel map uses a channel map positioning algorithm to predict the position information of the sample to be positioned. Step 4: Calculate the predicted distance matrix using the predicted position information of the samples to be located, and combine it with the global difference measurement matrix to obtain the updated global difference measurement matrix; Step 5: Repeat steps 3-4 until the performance of the validation set no longer improves, and obtain the final location information of the sample to be located.
2. A channel map positioning method based on beam domain channel covariance matrix according to claim 1, characterized in that The CSI matrix is collected using wireless channel simulation software or wireless channel measurement equipment based on ray tracing. The CSI matrix includes channel state information of each receiving antenna, each transmitting antenna, and each subcarrier, Gaussian white noise, and time advance.
3. A channel map positioning method based on beam domain channel covariance matrix according to claim 1, characterized in that In step 1, the beam domain channel covariance matrix is extracted based on the CSI matrix in the data set; this includes: Step 1.1: Perform covariance calculation on the CSI matrix to obtain the channel covariance matrix; Step 1.2: Perform a two-dimensional Fourier transform on the channel covariance matrix to obtain the beam-domain channel covariance matrix.
4. A channel map positioning method based on beam domain channel covariance matrix according to claim 1, characterized in that The calculation method of the global difference measurement matrix in step 2 is: Step 2.1, from the i-th sample x in the dataset i First, calculate the channel covariance cosine similarity matrix D of its nearest k samples; Step 2.2: Construct k-nearest neighbor graph G based on D k-NN ; Step 2.3: Apply the shortest path algorithm to G k-NN Get the matrix D geo '; Step 2.4: Repeat steps 2.1 to 2.3 for all samples to be located, and finally obtain D geo ' is the global difference measurement matrix D geo .
5. A channel map positioning method based on beam domain channel covariance matrix according to claim 1, characterized in that Channel map generation algorithm: The samples in the training set are divided into anchor samples, positive samples, and negative samples. The samples in the training set are input into the Triplet deep neural network for training to generate a channel map. The difference metric between the anchor samples and the positive samples is smaller than the difference metric between the anchor samples and the negative samples. The goal of the channel map generation algorithm is to optimize the neural network parameters so that the channel map distance between the anchor samples and the positive samples is smaller than the channel map distance between the anchor samples and the negative samples.
6. A channel map positioning method based on beam domain channel covariance matrix according to claim 1, characterized in that Channel map positioning algorithm: Using the channel map generated by the channel map generation algorithm, a semi-supervised learning method is used to transform the Triplet loss function L Triplet The actual positioning coordinate error L with the anchor point Position Combined, gradually improve L during the training process Position The proportion of the channel map positioning algorithm is calculated until the channel map positioning algorithm converges, and finally the predicted position information of each sample is obtained.
7. A channel map positioning method based on beam domain channel covariance matrix according to claim 1, characterized in that Step 4 includes: Step 4.1: Use the position information of the samples to be located to calculate the predicted distance matrix D between the samples to be located P ; Step 4.2: Use the predicted distance matrix D P Perform global geodesic algorithm to obtain the global prediction distance matrix D geo,P ; Step 4.3: The global prediction distance matrix D geo,P and the global difference measure matrix D geo The updated global difference measurement matrix D is obtained by fusion renew .
8. A channel map positioning method based on beam domain channel covariance matrix according to claim 6, characterized in that: The Triplet loss function L Triplet The actual positioning coordinate error L with the anchor point Position Combined; specifically as follows: Fusion Triplet loss function L Triplet The actual positioning coordinate error L with the anchor point Position , we get the neural network loss function Loss = αL Position +(1-α)L Triplet , where α is the error ratio of the real positioning coordinates of the anchor point.
9. A channel map positioning method based on beam domain channel covariance matrix according to claim 8, characterized in that: Among them, f θ (·) is the channel map generation algorithm, M is the margin, T is the triplet sample set formed according to the generated channel map, x i ,x j ,x k are the i-th, j-th, and k-th samples taken from the Triplet sample set T; Among them, g θ (·) is the channel map positioning algorithm, P is the anchor point set, x p is the p-th anchor point sample, y p is the real coordinate of the p-th anchor point.
10. A computer device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: When the processor executes the computer program, the steps of the channel map positioning method based on the beam domain channel covariance matrix are implemented as described in any one of claims 1 to 9.