Global positioning method for constructing channel graph based on deep learning difference matrix measurement

Abstract

The invention belongs to the technical field of mobile communication, and discloses a global positioning method for constructing a channel graph based on deep learning difference matrix measurement, and the method comprises the steps: collecting and processing channel state information data, constructing a difference measurement matrix based on deep learning, and constructing the channel graph through employing a multi-dimensional scaling manifold learning technology, and global positioning is realized in combination with means such as rotation, conformal and affine transformation, so that the universality of the method for constructing the channel graph and performing global positioning based on the predicted value of the difference measurement model is ensured.

Description

Technical Field

[0001] The present invention belongs to the technical field of mobile communication, and specifically relates to a global positioning method for constructing a channel map based on deep learning difference matrix measurement. Background Art

[0002] In order to cope with the substantial growth in traffic volume, the number of terminals, and the number of users, as well as the demand for high-reliability and low-latency communication, large-scale multiple-input multiple-output (mMIMO) technology has been designed for future wireless communication systems. mMIMO technology significantly enhances wireless signal coverage by multiplying the system communication capacity and using technologies such as spatio-temporal beamforming, multi-user separation, and data precoding. The base station (BS) utilizes mMIMO antenna technology to collect rich channel state information (CSI) from user equipment (UE), and based on the similarity of this information in high-dimensional space, a self-supervised learning strategy can be used to draw a two-dimensional or even three-dimensional map of the radio environment (hereinafter referred to as: channel map). Based on the channel map, global positioning of user equipment (UE) can be achieved without relying on location tags or relying on very few location tags.

[0003] In recent years, domestic and foreign scholars have conducted in-depth research on channel map positioning technology. Typical methods include: the positioning method proposed by Jaakko Pihlajasalo et al. in 2020 based on multi-base station channel map generation, combination, and mapping transformation, and the positioning method proposed by Florian Euchne et al. in 2023 that fuses channel state information with classical source localization (triangulation and multilateration) information. In these efficient channel map construction and positioning methods, the differential measurement of channel state information is a key technology. Domestic and foreign scholars have also proposed a series of methods, including: the differential measurement method based on timestamps proposed by P. Ferrand et al. in 2021, the measurement method using cosine similarity proposed by L. LeMagoarou in 2021; the dissimilarity measurement method based on channel impulse response amplitude (CIRA) proposed by M. Stahlke et al. in 2023, and the differential matrix calculation method based on angle-delay curve (ADP) proposed by Phillip Stephan et al. in 2024.

[0004] However, the above differential measurement of channel state information relies on specific channel characteristics such as amplitude attenuation and phase change, which makes the user equipment positioning method based on the channel map usually have the following disadvantages: The positioning method based on specific characteristics such as amplitude attenuation and phase change has a very high dependence on the environment, and it is difficult to ensure the universality of the positioning technology. Summary of the Invention

[0005] To solve the above technical problems, the present invention provides a global positioning method for constructing a channel map based on deep learning difference matrix measurement. By means of a difference measurement matrix based on deep learning, a channel map is constructed using multi-dimensional scaling manifold learning technology, and global positioning is achieved through transformation means, ultimately achieving the accurate positioning goal of a general-purpose user equipment (UE).

[0006] The global positioning method for constructing a channel map based on deep learning difference matrix measurement according to the present invention includes the following steps:

[0007] Step 1: Collect channel state data, and perform time-domain transformation on the collected channel state data to obtain time-domain channel state information data;

[0008] Step 2: Perform feature engineering on the time-domain channel state information data, and construct a multi-base station difference matrix based on the differences of deep learning;

[0009] Step 3: Based on the multi-base station difference matrix, use multi-dimensional scaling manifold learning to construct a channel map;

[0010] Step 4: Perform rotation transformation, common transformation, and affine transformation on the channel map to achieve global positioning.

[0011] Further, Step 1 is specifically as follows:

[0012] Step 1-1: Obtain channel state information data;

[0013] Given a base station BS composed of B distributed antenna arrays mMIMO, with each antenna array equipped with M antennas, and a user equipment UE with a single antenna, the base station BS includes B×M antennas, and the channel coefficients of the N sub orthogonal frequency division multiplexing OFDM subcarriers measured between the B×M antennas of the base station BS and the single antenna of the user equipment UE at the l-th discrete time are expressed as a frequency-domain tensor H (l) , and the formula is:

[0014]

[0015] Among them, the set of channel coefficients measured at L discrete times and the corresponding measurement times constitute a frequency-domain channel state information data set CSI Fre , and the formula is:

[0016] CSI Fre ={(H (l) ,t (l) )}, l = 1,..., L (2)

[0017] Among them, t (l) ∈R is the discrete time stamp of the channel coefficient measurement, in milliseconds; R is the set of real numbers;

[0018] Step 1-2, Time-domain transformation of the channel state information data set:

[0019] For channel graph-based positioning applications, time-domain transformation of the channel state data set is required. Perform the inverse discrete Fourier transform of H (l) along the frequency axis to obtain the corresponding time-domain tensor The formula is:

[0020]

[0021] where, n is the order of τ, im is the imaginary unit, τ is the time tap, τ = 1 corresponds to the first time tap, and τ = N sub corresponds to the last time tap;

[0022] Obtain the corresponding time-domain channel state information data CSI Tim , and the formula is:

[0023]

[0024] Furthermore, in Step 2, perform feature engineering on the time-domain channel state information data, specifically:

[0025] Step 2-1-1, Calculate the autocorrelation features of the time-domain channel state information data;

[0026] Given a base station BS composed of B distributed antenna arrays mMIMO, with each antenna array equipped with M antennas, and the minimum time tap τ min and the maximum time tap τ max , calculate a set of index tuples J based on the Cartesian product, and the formula is:

[0027] J = {1,..., B} 2 × {1,..., M} 2 × {τ min ,..., τ max}(5)

[0028] where, 1 << τ min < N sub / 2 < τ max << N sub , that is, only consider the time taps of multiple channel impulse responses CIRs containing the visible path and the first multiple path components; τ min and τ max The selection of depends on the expected maximum delay distribution;

[0029] For each pair of antenna arrays b i , b j, for each pair of antennas m p , m q and for each time tap τ, the time-domain tensor of l discrete-time measurements between the sample autocorrelation features c (l) is calculated as follows:

[0030]

[0031] where (b i , b j , m p , m q , τ) ∈ J; i, j, p, q ∈ R, where R is the set of real numbers; * denotes taking the conjugate ;

[0032] Step 2-1-2: Vectorize the autocorrelation features to obtain a feature vector;

[0033] By separating the real and imaginary parts of c (l) to achieve vectorization, the corresponding feature vector f (l) is obtained, and the formula is:

[0034]

[0035] where Re{} and Im{} respectively denote the real and imaginary parts of the complex number c (l) .

[0036] Furthermore, in Step 2, a multi-base station difference matrix is constructed based on the differences of deep learning, specifically:

[0037] Step 2-2-1: Train and optimize a difference metric model based on the deep neural network DNN;

[0038] The feature vectors f (i) and f (j) of any two discrete times i, j are used as model inputs, where 1 ≤ i, j ≤ L and i ≠ j; through the processing of multiple hidden layers, the estimated value of the difference in the channel state information data acquisition times i, j is obtained from the output layer Calculate the output estimated value and the error between the true measurement value d time,i,j = |t (i) - t (j) |

[0039] Define the loss function by the normalized mean square error NMSE, and the formula is:

[0040]

[0041] where dtime,i,j = |t (i) -t (j) | represents the true difference in the channel state information data acquisition times i and j, and β > 0 is a hyperparameter;

[0042] Finally, calculate the gradient through the loss function, and obtain the optimized difference metric model DΘ based on backpropagation calculation and stochastic gradient descent;

[0043] Step 2-2-2: Predict and output the difference values of all channel state information data acquisition times i and j based on the trained optimized difference metric model DΘ and Average the results based on the symmetry of the dissimilarity to obtain the difference metric d of the channel state information data at acquisition times i and j DL,i,j , and the formula is:

[0044]

[0045] Step 2-2-3: Construct a multi-base station difference matrix based on visibility;

[0046] Based on the difference values of all channel state information data acquisition times i and j, obtain the difference metric matrix D for L time periods of the base station DL , and the formula is:

[0047]

[0048] Given n base stations BS and a user equipment UE, at two discrete measurement times i and j, the user equipment UE is visible to the base station k, 1 ≤ k ≤ n, then the visibility of the user equipment UE to the base station k is defined as Otherwise

[0049] Based on the weighted average, combine the difference metric matrices D of n base stations BS for the user equipment UE DL to obtain the global difference metric matrix D DLG , and the formula is:

[0050]

[0051] where is the difference value of the channel state information data acquisition times i and j based on the output of the trained model; is the difference value of the channel state information data acquisition times j and i based on the output of the trained model; is the difference value of the channel state information data between the base station k and the user equipment UE at two discrete measurement times i and j; is the visibility of the user equipment UE to the base station k at two discrete measurement times i and j; d DLG,i,jis the time-domain tensor of the channel state information data at collection times i and j in the high-dimensional space the medium distance

[0052] Further, step 3 is specifically as follows:

[0053] Given the time-domain tensor of the time-domain channel state data at any two discrete collection times i and j Using the multi-dimensional scaling manifold learning method to obtain their mapped positions z in the reduced-dimensional space R D＝2 in which (i) , z (j) , that is:

[0054]

[0055] The mapping calculation satisfies the limiting condition:

[0056] ||z (i) - z (j) ||2 ∝ d DLG,i,j (13)

[0057] where ||z (i) - z (j) ||2 is the distance between the mapped positions z in the corresponding reduced-dimensional space R D＝2 in which (i) , z (j) , and ∝ means to approach to the greatest extent;

[0058] For the time-domain channel state data at all l = 1, …, L discrete times, the pairwise distance d DLG at the points in D DLG,i,j and the corresponding ||z (i) - z (j) ||2 should satisfy the minimization of the mean square distance, that is:

[0059]

[0060] All z that satisfy the conditions (l) constitute the channel state data channel graph Z in the reduced-dimensional space R D＝2 , that is: cc , that is:

[0061]

[0062] Further, to construct the channel state data channel graph in the reduced-dimensional space, the specific steps are as follows:

[0063] (1) Calculate using the element data in D DLG in which

[0064] (2) Calculate the inner product and further obtain the inner product matrix after dimensionality reduction wherein, is the mean value of all elements in the i-th row of D DLG , is the mean value of all elements in the j-th row of D DLG , is the global mean value of D DLG ;

[0065] (3) Perform eigenvalue decomposition on the matrix IPM, IPM = VΛV T , where T represents the transpose of the eigenvector matrix V, is the diagonal matrix composed of eigenvalues, When V has two non-zero eigenvalues, V = V * , Λ = Λ * = diag(λ1, λ2), then the time-domain channel state data channel graph is obtained

[0066] Furthermore, step 4 is specifically as follows:

[0067] Step 4-1: When the base station collects the channel state data of the user equipment at L discrete times, synchronously obtain the reference signal received power data set and estimate the position of the base station in the channel graph;

[0068] Step 4-2: Given the map spatial position of the center translation of each base station, calculate its rotation angle relative to the estimated position in the channel graph, and use the circular average method to estimate the global rotation angle;

[0069] Step 4-3: Perform conformal transformation using elliptical grid mapping to map the circle to a square to eliminate the circularity in the channel graph;

[0070] Step 4-4: Construct a homogeneous coordinate matrix, obtain an affine transformation matrix based on a linear equation system, and solve it;

[0071] Step 4-5: Perform global positioning based on the affine matrix.

[0072] The beneficial effects of the present invention are as follows: By constructing a difference metric matrix based on deep learning, the method of the present invention accurately captures specific signal features in complex environments, and then uses multi-dimensional scaling manifold learning technology to construct a channel graph, and combines means such as rotation, conformal and affine transformations to achieve global positioning. In the present invention, a deep neural network (DNN) difference metric model is designed with the time-domain channel state information at discrete times as the model input, and optimized by backpropagation calculation and stochastic gradient descent. It uses the time difference of real measurement values that do not depend on specific scene features as the verification value, ensuring the universality of the method for constructing a channel graph and performing global positioning based on the predicted values of the difference metric model. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 is a flowchart of the method of the present invention.

[0074] Figure 2 is a schematic diagram of the channel coefficient between a base station (BS) and a user equipment (UE);

[0075] Figure 3 is a flowchart of training the difference metric model based on a deep neural network;

[0076] Figure 4 is a schematic diagram of the optimized model structure based on backpropagation calculation and stochastic gradient descent;

[0077] Figure 5 is a schematic diagram of the visibility of a user equipment (UE) to a single base station (BS);

[0078] Figure 6 is a schematic diagram of the visibility of a user equipment (UE) to a base station (BS1) and a base station (BS2);

[0079] Figure 7 is a channel graph constructed based on multi-dimensional scaling manifold learning;

[0080] Figure 8 is a schematic diagram of the estimated corresponding positions of the spatial positions of 4 base stations in the channel graph;

[0081] Figure 9 is a schematic diagram of global positioning of conformal transformation points in the channel graph in the map space based on an affine transformation matrix. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0082] In order to make the content of the present invention easier to be clearly understood, the present invention will be further described in detail below according to specific embodiments and in conjunction with the accompanying drawings.

[0083] As Figure 1 shown, the global positioning method for constructing a channel graph based on deep learning difference matrix measurement of the present invention includes the following steps:

[0084] Step 1: Collect channel state data, perform time-domain transformation on the collected channel state data to obtain time-domain channel state information data;

[0085] Step 2: Perform feature engineering on the time-domain channel state information data, and construct a multi-base station difference matrix based on the differences of deep learning;

[0086] Step 3: Based on the multi-base station difference matrix, use multi-dimensional scaling manifold learning to construct a channel graph;

[0087] Step 4: Perform rotation transformation, common transformation, and affine transformation on the channel graph to achieve global positioning.

[0088] Among them, Step 1 is specifically as follows:

[0089] Step 1-1: Obtain channel state information data;

[0090] Given a base station (BS) composed of B distributed antenna arrays (mMIMO) with M antennas equipped in each antenna array and a single-antenna user equipment (UE), the BS contains B×M antennas, and the channel coefficients of the N sub orthogonal frequency division multiplexing (OFDM) subcarriers measured between the B×M antennas of the BS and the single antenna of the user equipment (UE) at the l-th discrete time are expressed as a frequency-domain tensor H (l) , as Figure 2 shown, the formula is:

[0091]

[0092] Among them, the set of channel coefficients measured at L discrete times and the corresponding measurement times constitute a frequency-domain channel state information data set CSI Fre , the formula is:

[0093] CSI Fre ={(H (l) ,t (l) )}, l = 1,..., L (2)

[0094] Among them, t (l) ∈R is the discrete time stamp of the channel coefficient measurement, in milliseconds; R is the set of real numbers;

[0095] Step 1-2: Time-domain transformation of the channel state information data set:

[0096] Perform the inverse discrete Fourier transform of unit along the frequency axis on H (l) to obtain the corresponding time-domain tensor The formula is:

[0097]

[0098] Among them, n is the order of τ, im is the imaginary unit, τ is the time tap, τ = 1 corresponds to the first time tap, and τ = N sub corresponds to the last time tap;

[0099] The corresponding channel state information data CSI in the time domain is obtained Tim , and the formula is:

[0100]

[0101] In step 2, a differential metric model is trained based on a deep neural network, and the basic process is as Figure 3 shown.

[0102] (1) Feature engineering of channel state information data;

[0103] 1) Calculate the autocorrelation feature of the channel state information data in the time domain

[0104] Given a base station (BS) composed of B distributed antenna arrays (mMIMO) with M antennas in each antenna array, and the minimum time tap τ min and the maximum time tap τ max , a set of index tuples J is calculated based on the Cartesian product, and the formula is:

[0105] J = {1,..., B} 2 ×{1,..., M} 2 ×{τ min ,..., τ max}(5)

[0106] where 1 << τ min < N sub / 2 < τ max << N sub , that is, only the time taps of multiple channel impulse responses (CIRs) including the visible path and the first multiple path components are considered. τ min and τ max The selection of depends on the expected maximum delay distribution.

[0107] For each pair of antenna arrays b i , b j , each pair of antennas m p , m q and each time tap τ in the base station (BS) of the antennas, the sample autocorrelation feature between the l discrete-time measured time-domain tensors is c (l) , and the formula is:

[0108]

[0109] wherein (b i , b j , m p , m q , τ) ∈ J; i, j, p, q ∈ R, where R is the set of real numbers; * represents taking the conjugate ;

[0110] 2) Obtain the feature vector by vectorizing the autocorrelation feature

[0111] By separating the real and imaginary parts of c (l) to achieve vectorization, the corresponding feature vector f (l) is obtained, and the formula is

[0112]

[0113] Re{} and Im{} respectively represent the real and imaginary parts of the complex number c (l) .

[0114] (2) Training of the difference metric model based on the deep neural network (DNN)

[0115] 1) Optimization of model training based on stochastic gradient descent and backpropagation calculation

[0116] For any two discrete times i, j (1 ≤ i, j ≤ n and i ≠ j), the feature vectors f (i) and f (j) are used as model inputs. Through the processing of multiple hidden layers, the output layer obtains an estimated value of the difference between the channel state information data acquisition times i and j Calculate the output estimated value and the error between the true measurement value d time,i,j = |t (i) - t (j) |

[0117] Define the loss function through the normalized mean square error NMSE, and the formula is

[0118]

[0119] where d time,i,j = |t (i) - t (j) | represents the true difference between the channel state information data acquisition times i and j, and β > 0 is a hyperparameter

[0120] Finally, calculate the gradient through the loss function, and obtain the optimized model DΘ based on backpropagation calculation and stochastic gradient descent. The basic network structure is as Figure 4as shown

[0121] In this example, the dissimilarity metric model DΘ includes: 1 input layer, 3 hidden layers, and 1 output layer; the number of neuron nodes in the 3 hidden layers is 128, 64, and 32 respectively, all using the ReLU activation function, and a fully connected method is adopted between the nodes of each layer.

[0122] 2) Predict and output the difference value of all channel state information data acquisition times i and j based on the trained optimized model DΘ and And based on the symmetry of the dissimilarity, average the results to obtain the dissimilarity metric d of the channel state information data at acquisition times i and j DL,i,j , the formula is:

[0123]

[0124] (3) Construction of the multi-base station dissimilarity matrix based on visibility;

[0125] 1) Based on the difference values of all channel state information data acquisition times i and j, obtain the dissimilarity metric matrix D of the base station in L time periods DL , the formula is:

[0126]

[0127] Given n base stations and a user equipment (UE), at two discrete measurement times i and j, the user equipment (UE) is visible (LoS) to base station k, 1 ≤ k ≤ n, then the visibility of the user equipment (UE) to base station (BS) k is defined as otherwise

[0128] In this example, Figure 5 is an example of the visibility of the user equipment (UE) to a single base station (BS), where Figure 5 in (a) of means that the user equipment (UE) is visible to the base station (BS) at both discrete measurement times i and j; Figure 5 (b) of means that the user equipment (UE) is visible to the base station (BS) at discrete measurement time i, but not visible to the base station (BS) at discrete measurement time j.

[0129] Figure 6 is an example of the visibility of the user equipment (UE) to multiple base stations (BS), where Figure 6 in (a) of means that the user equipment (UE) is visible to the base station (BS1) at discrete measurement time i, but not visible to the base station (BS1) at discrete measurement time j; It means that the user equipment (UE) is visible to the base station (BS2) at discrete measurement times i and j; Figure 6 in (b) of It means that the user equipment (UE) is visible to the base station (BS1) at discrete measurement time i and invisible to the base station (BS1) at discrete measurement time j; It means that the user equipment (UE) is invisible to the base station (BS2) at discrete measurement time i and visible to the base station (BS2) at discrete measurement time j.

[0130] 2) Based on the weighted average, combine the differential metric matrices D of n base stations (BS) for the user equipment (UE) DL to obtain the global differential metric matrix D DLG , and the formula is:

[0131]

[0132] where is the difference value of the channel state information data acquisition times i and j based on the output of the training model; is the difference value of the channel state information data acquisition times j and i based on the output of the training model; is the difference value of the channel state information data between base station k and user equipment UE at two discrete measurement times i and j; is the visibility of user equipment UE to base station k at two discrete measurement times i and j; d DLG,i,j is the time-domain tensor of the channel state information data at acquisition times i and j in the high-dimensional space distance.

[0133] In step 3, given the time-domain tensor of the time-domain channel state data at any two discrete acquisition times i and j adopt the multi-dimensional scaling manifold learning method to obtain their mapped positions z in the reduced-dimensional space R D ＝2 , that is: (i) z (j) , that is:

[0134]

[0135] The mapping calculation satisfies the limiting condition:

[0136] ||z (i) -z (j) ||2 ∝ d DLG,i,j (13)

[0137] where ||z (i) -z (j) ||2 is the time-domain tensor in the corresponding reduced-dimensional space RD＝2 The mid-mapping position z (i) , z (j) The distance between them, ∝ indicates the closest approximation.

[0138] For all time-domain channel state data at discrete times l = 1, …, L, D DLG The midpoint pair distance d DLG,i,j corresponding to ||z (i) - z (j) ||2 should satisfy the minimization of the mean square distance, that is:

[0139]

[0140] All z that satisfy the condition (l) constitute the channel state data channel graph Z in the reduced-dimensional space R D＝2 , that is: cc , that is:

[0141]

[0142] The specific implementation steps are as follows:

[0143] (1) Use the element data in D DLG to calculate

[0144] (2) Calculate the inner product and further obtain the inner product matrix after dimensionality reduction where, is the mean of all elements in the i-th row of D DLG , is the mean of all elements in the j-th row of D DLG , is the global mean of D DLG .

[0145] (3) Perform eigenvalue decomposition on the matrix IPM, IPM = VΛV T , where T represents the transpose of the eigenvector matrix V, is the diagonal matrix composed of eigenvalues, When V has 2 non-zero eigenvalues, V = V * , Λ = Λ * = diag(λ1, λ2), then the time-domain channel state data channel graph

[0146] In this example, Figure 7 is the channel graph obtained from the channel state data measured at 10000 measurement times (in milliseconds) based on 4 base stations and 1 user equipment (UE) through steps 1 to 3.

[0147] In step 4, the channel graph Z obtained based on the channel state data cc It does not have a true scale and shape and cannot be used for map spatial positioning. Rotation transformation, conformal transformation, and affine transformation are required. The specific steps include:

[0148] (1) Position estimation of the base station BS in the channel graph;

[0149] 1) When the base station (BS) collects the channel state data of the user equipment (UE) at L discrete times, it synchronously obtains the reference signal received power data set RSRQs, which is defined as:

[0150] RSRQs = {power1, power2,..., power L}(16)

[0151] where power i , 1 ≤ i ≤ L, represents the signal received power at the i-th discrete time.

[0152] 2) Sort the elements in CSI-RSRQs in descending order to obtain a subset RSRQs t composed of the first T elements, which is defined as:

[0153] RSRQs T = {por1, por2,..., por T}(17)

[0154] 3) Based on the collection times of the elements in RSRQst, obtain a set of T corresponding mapped positions from the time-domain channel state data channel graph generated by the base station (BS) and the user equipment (UE) Its definition is:

[0155]

[0156] 4) Based on the mean value of the elements in, estimate the position z of the base station (BS) in the channel graph Z cc Its definition is: (bs)

[0157]

[0158] In this example, Figure 8

[0159] is a schematic diagram of the position estimation of the spatial positions of 4 base stations corresponding in the channel graph. (2) Rotation transformation;

[0160] 1) Central translation of the map spatial position of the base station BS;

[0161] Calculate the average central position of n base stations (BSs), and subtract the coordinates of the average central position from the coordinates of each base station (BS) to achieve the central translation of the map space position of the base station (BS). The calculation formula is as follows:

[0162]

[0163] where, is the map space position of base station k, and BSs cen is the average central position of the map space positions of n base stations (BSs), is the position translated to the average center.

[0164] 2) Calculation of the rotation angle between the map position after the central translation of base station BS and the estimated position of base station BS in the channel map

[0165] Given the map space position of base station k after central translation and the estimated position of base station k in the channel map Calculate the rotation angle θ between two points based on the two-parameter arctangent function. The formula is as follows:

[0166]

[0167] where, are the estimated positions of base station (BS) k in the channel map of the horizontal and vertical coordinates respectively, are the horizontal and vertical coordinates of the map space position of base station (BS) k after central translation respectively.

[0168] Furthermore, based on the position information of n base stations, for the n calculated rotation angles, use the circular mean method to estimate the global rotation angle The formula is as follows:

[0169]

[0170] where, θ j is the rotation angle calculated based on the position information of base station j.

[0171] 3) Use the globally estimated rotation angle For each mapped position z (l) perform rotation to obtain the corresponding rotated point The formula is as follows:

[0172]

[0173] Furthermore, obtain the set of all rotated points

[0174] (3) Conformal transformation;

[0175] Adopt elliptical grid mapping for conformal transformation to map a circle to a square to eliminate the circularity in the channel graph;

[0176] 1) For in scale according to the maximum distance from a point to the center origin of the channel graph, and the calculation formula is:

[0177]

[0178] where, represents the distance from the rotation point to the center origin of the channel graph, represents the maximum value of the distances from L rotation points to the origin, represents the scaled point;

[0179] 2) Perform elliptical grid conformal mapping on the scaled point to obtain the conformal point The calculation formula is:

[0180]

[0181] where, are respectively the horizontal and vertical coordinates of the scaled point , correspond respectively to the horizontal and vertical coordinates of the point after conformal transformation.

[0182] Furthermore, obtain the set of all conformal points

[0183] (4) Affine transformation;

[0184] Given the set of n base station (BS) map spatial positions and their corresponding positions in the channel graph the set of positions after rotation, scaling, and conformal processing solve for the affine transformation parameters, specifically:

[0185] 1) Construct the homogeneous coordinate matrix;

[0186] For each point and the corresponding target point BS k ·Maplocs, 1 ≤ k ≤ n are extended to homogeneous coordinates and composed into the following matrix by rows:

[0187]

[0188] 2) Construct the linear equations;

[0189] For each point and the corresponding target point BS k · Maplocs, 1 ≤ k ≤ n, satisfy the following relationship:

[0190]

[0191] For all n pairs of point sets, their relationship is expressed as a matrix equation:

[0192]

[0193] where M is the affine transformation matrix.

[0194] 3) Solve the affine transformation matrix;

[0195] Solve the affine transformation matrix M based on the following matrix equation:

[0196] M = (X T X) -1 X T Y(29)

[0197] where X T is the transpose matrix.

[0198] (5) Perform global positioning based on the affine matrix;

[0199] Given the set of conformal transformation points in the channel graph and the affine transformation matrix M, for any conformal transformation point Construct homogeneous coordinates:

[0200]

[0201] Furthermore, based on the homogeneous coordinates and the affine transformation matrix M, calculate the global positioning point Maploc in the corresponding map space l , and the calculation formula is:

[0202]

[0203] Finally, obtain the set of global positioning points of all conformal transformation points in the channel graph in the map space

[0204]

[0205] In this example, Figure 9 is a schematic diagram of the transformation of a 30 * 75 fixed-point channel graph to the map space based on the conformal transformation affine transformation matrix in the channel graph.

[0206] Perform error analysis on the coordinates of 30 * 75 in the map space obtained by affine transformation of the channel graph and the true coordinates. The results are shown in Table 1. From the experimental results of the sampling point residuals and total errors in Table 1: The channel graph affine transformation model obtained from 10 sample points has high positioning accuracy when applied to 30 * 75 test points, verifying the universality of the global positioning method based on constructing the channel graph by measuring the difference matrix of deep learning proposed in the present invention.

[0207] Table 1 Error analysis of the map space based on the affine transformation of the channel graph

[0208]

[0209] The above is only the preferred solution of the present invention and is not intended as a further limitation of the present invention. All equivalent changes made by using the content of the specification and drawings of the present invention are within the protection scope of the present invention.

Claims

1. A global positioning method for constructing a channel graph based on deep learning differential matrix measurement, characterized in that It includes the following steps: Step 1: Collect channel state data, perform time-domain transformation on the collected channel state data, and obtain time-domain channel state information data; Step 2: Perform feature engineering on the time-domain channel state information data, and construct a multi-base station difference matrix based on the differences of deep learning; Step 3: Based on the multi-base station difference matrix, construct a channel graph using multi-dimensional scaling manifold learning; Step 4: Perform rotation transformation, commonality transformation, and affine transformation on the channel graph to achieve global positioning.

2. The global positioning method for constructing a channel graph based on deep learning difference matrix metric according to claim 1, characterized in that, Specifically, Step 1 is as follows: Step 1-1: Obtain channel state information data; Given a base station BS consisting of B distributed antenna arrays mMIMO with each antenna array equipped with M antennas and a user equipment UE with a single antenna, the base station BS includes B×M antennas, and the channel coefficients of N sub orthogonal frequency division multiplexing OFDM subcarriers measured between the B×M antennas of the base station BS and the single antenna of the user equipment UE at the l-th discrete time are expressed as a frequency domain tensor H (l) , and the formula is: Among them, the set of channel coefficients of L discrete-time measurements and the corresponding measurement times constitute the frequency-domain channel state information data set CSI Fre , and the formula is as follows: CSI Fre = {(H (l) , t (l) ), l = 1, ..., L (2) where, t (l) ∈R is the discrete timestamp for channel coefficient measurement, in milliseconds; R is the set of real numbers; Step 1-2: Time-domain transformation of the channel state information data set: Perform the inverse discrete Fourier transform of H (l) along the frequency axis to obtain the corresponding time-domain tensor The formula is as follows: wherein, n is the order of τ, im is the imaginary unit, τ is the time tap, τ = 1 corresponds to the first time tap, and τ = N sub corresponds to the last time tap; Obtain the corresponding time-domain channel state information data CSI Tim , and the formula is:

3. The global positioning method for constructing a channel graph based on deep learning difference matrix metric according to claim 2, wherein In Step 2, when performing feature engineering on the time-domain channel state information data, specifically: Step 2-1-1: Calculate the autocorrelation features of the time-domain channel state information data; Given a base station BS consisting of B distributed antenna arrays mMIMO, where each antenna array is equipped with M antennas, and a minimum time tap τ min and a maximum time tap τ max , a set of index tuples J is calculated based on the Cartesian product, and the formula is: J = {1,..., B} 2 × {1,..., M} 2 × {τ min ,..., τ max}} (5) where 1 << τ min < N sub / 2 < τ max << N sub , that is, only the time taps of multiple channel impulse responses (CIRs) including the visual path and the first several path components are considered; the selection of τ min and τ max depends on the expected maximum delay profile; For each pair of antenna arrays b i , b j in the base station BS of the antenna, each pair of antennas m p , m q and each time tap τ, the sample autocorrelation feature c between the l discrete-time measurements of the time-domain tensor (l) is calculated as follows: Among them, (b i ,b j ,m p ,m q ,τ) ∈ J; i, j, p, q ∈ R, where R is the set of real numbers; * denotes taking the conjugate; taking the conjugate; Step 2-1-2: Vectorize the autocorrelation features to obtain feature vectors; By separating the real and imaginary parts of c (l) vectorization is achieved to obtain the corresponding eigenvector f (l) , and the formula is: where Re{} and Im{} respectively represent the real part and the imaginary part of the complex number c (l) ​ 4. The global positioning method for constructing a channel graph based on deep learning difference matrix metric according to claim 3, characterized in that, In Step 2, when constructing a multi-base station difference matrix based on the differences of deep learning, specifically: Step 2-2-1: Train and optimize a difference metric model based on the deep neural network DNN; The eigenvectors f for any two discrete times i, j (i) , f (j) are used as model inputs, where 1 ≤ i, j ≤ L and i ≠ j; through the processing of multiple hidden layers, the output layer obtains an estimated value of the difference in the channel state information data acquisition times i, j Calculate the output estimated value and the error with the true measurement value d time,i,j = |t (i) - t (j) | Define the loss function through the normalized mean square error NMSE, and the formula is: where d time,i,j = |t (i) - t (j) | represents the true difference between the channel state information data acquisition times i and j, and β > 0 is a hyperparameter; Finally, calculate the gradient through the loss function, and obtain the optimized difference metric model DΘ based on backpropagation calculation and stochastic gradient descent; Step 2-2-2: Predict and output the difference value of the acquisition times i and j of all channel state information data based on the trained optimized dissimilarity metric model DΘ and Based on the symmetry of the dissimilarity, average the results to obtain the dissimilarity metric d of the channel state information data at acquisition times i and j DL,i,j , the formula is: Step 2-2-3: Construct a multi-base station difference matrix based on visibility; Based on the difference values of the data acquisition times \(i\) and \(j\) of all channel state information, the differential metric matrix \(D\) of the base station in \(L\) time periods is obtained DL , and the formula is: Given n base stations BS and a user equipment UE, at two discrete measurement times i, j, if the user equipment UE is visible to base station k, 1 ≤ k ≤ n, then the visibility of the user equipment UE to base station k is defined as Otherwise Based on the weighted average combination of the difference metric matrices D of n base stations BS for the user equipment UE DL to obtain the global difference metric matrix D DLG , the formula is: Among them, is the difference value of the channel state information data acquisition times i and j output based on the training model; is the difference value of the channel state information data acquisition times j and i output based on the training model; is the difference value of the channel state information data between the base station k and the user equipment UE at two discrete measurement times i and j; is the visibility of the user equipment UE for the base station k at two discrete measurement times i and j; d DLG,i,j is the time-domain tensor of the channel state information data at the acquisition times i and j in the high-dimensional space distance.

5. The global positioning method for constructing a channel graph based on deep learning difference matrix metric according to claim 4, characterized in that, Specifically, Step 3 is as follows: Time-domain tensor of time-domain channel state data at any two discrete acquisition times i and j Using the multi-dimensional scaling manifold learning method to obtain their mapped positions z in the reduced-dimensional space R D＝2 where z (i) , z (j) , that is: The mapping calculation satisfies the specified conditions: ||z (i) -z (j) ||2∝d DLG,i,j (13) where, ||z (i) -z (j) ||2 is a time-domain tensor at the corresponding mapped position z in the reduced-dimensional space R D＝2 , z (i) , z (j) the distance between them, and ∝ indicates the closest approximation; For the time-domain channel state data at all discrete times \(l = 1,\ldots,L\), \(D\) DLG Midpoint pair distance \(d\) DLG,i,j And the corresponding \(\|\mathbf{z}\) (i) -\mathbf{z}\) (j) The distance between \(\|_2\) should satisfy the minimization of the mean square distance, that is: All z that satisfy the conditions (l) constitute the channel state data channel graph Z D＝2 in the reduced-dimensional space R cc , that is:

6. The global positioning method for constructing a channel graph based on deep learning differential matrix metric according to claim 5, characterized in that Based on all z that meet the conditions (l) Construct a channel state data channel graph in the dimensionality-reduced space. The specific steps are as follows: (1) Calculate using the element data in D DLG ​ (2) Calculate the inner product and further obtain the inner product matrix after dimensionality reduction where, is the mean of all elements in the i-th row of D DLG , is the mean of all elements in the j-th row of D DLG , is the global mean of D DLG . (3) Perform eigenvalue decomposition on the matrix IPM, IPM = VΛV T , where T represents the transpose of the eigenvector matrix V, and it is a diagonal matrix composed of eigenvalues, When V has 2 non-zero eigenvalues, V = V * , Λ = Λ * = diag(λ1, λ2), then the time-domain channel state data channel graph is obtained 7. The global positioning method for constructing a channel graph based on the deep learning difference matrix metric according to claim 6, characterized in that, Specifically, Step 4 is as follows: Step 4-1: When the base station collects the channel state data of the user equipment at L discrete times, synchronously obtain the reference signal received power data set, and estimate the position of the base station in the channel graph; Step 4-2: Given the map space position of the center translation of each base station, calculate its rotation angle compared with the estimated position in the channel graph, and estimate the global rotation angle using the circular average method; Step 4-3: Perform conformal transformation using elliptical grid mapping to map the circle to a square to eliminate the circularity in the channel graph; Step 4-4: Construct a homogeneous coordinate matrix, obtain an affine transformation matrix based on a system of linear equations, and solve it; Step 4-5: Perform global positioning based on the affine matrix.

Images