A channel estimation method empowered by real-time channel knowledge graph

Abstract

The application discloses a kind of real-time channel knowledge graph empowerment's channel estimation method, belong to information and communication technology field.The application is by low cost and continuous communication receiving signal, extracts the channel statistical information including PDP and PAP relevant to user position, which reflects the local wireless propagation characteristics of user customization, corresponding frequency domain and spatial covariance matrix has low-dimensional structured subspace feature;The application fills the blank of the existing channel knowledge graph technology neglecting parameter extraction;Further, the application uses the low-dimensional structured frequency-space covariance matrix, designs two kinds of serial MMSE channel estimation scheme, which compared with the existing non-customized statistical prior channel estimation method, realizes the substantial performance improvement.And, experiments show that the scheme of the application still has performance advantage compared with the existing MMSE scheme based on historical sample covariance when pilot overhead is 1 / 4.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information and communication technology, and in particular to a channel estimation method empowered by real-time channel knowledge graph. BACKGROUND

[0002] Wireless communication networks are evolving towards larger scale antennas (such as extra-large scale multiple-input multiple-output, i.e., XL MIMO), higher frequency band transmissions (such as millimeter wave terahertz communications), and denser connections, providing larger scale access and higher transmission rates, while also causing a significant increase in the dimension of wireless channels and the intensification of signal interference problems, posing great challenges to the performance and pilot overhead of wireless air interface algorithms such as channel estimation and beamforming. On the other hand, large-dimensional wireless channels and dense access nodes also bring new opportunities, as they carry a large amount of wireless data that can be utilized. These data reflect the local characteristics of the wireless propagation environment, which has led to a leap from non-environmental perception to environmental perception in wireless communications.

[0003] Existing environmental perception technologies include a physical environment map + ray tracing platform “The design and applications of high-performance ray-tracing simulation platform for 5g and beyond wireless communications: A tutorial”, which first depicts the three-dimensional model of scatterers in the environment, the material electromagnetic parameters, and the propagation mode of electromagnetic waves, etc. Based on this information, a ray tracing platform is established to simulate the deterministic propagation of electromagnetic waves in a specific environment. However, due to the intensive computation involved in ray tracing, the physical environment map is mainly used for off-network system evaluation and optimization.

[0004] Recently, the concept of channel knowledge map (CKM) is proposed, which can be regarded as a mapping of user location to channel statistics, where the input user location information can be obtained according to the integration technology of sensing, and the output channel statistical characteristics can be customized according to the requirements of wireless air interface algorithms, such as the existence of line-of-sight / non-line-of-sight path to assist beamforming, channel multipath angle-time delay information to assist channel estimation. And since CKM is regarded as a mapping function or a lookup table, it can not only be used for offline system optimization, but also assist real-time signal processing. The existing CKM construction work "Channel knowledge map for environment-aware communications: EM algorithm for map construction" focuses on how to predict or interpolate the channel characteristics of the entire communication area based on the channel statistical characteristics obtained by the probe location, ignoring the process of extracting channel characteristics from data. And the existing literature lacks research on how to extract channel characteristics from real-time received signals and assist channel estimation. Therefore, it is necessary to propose a technical solution focusing on real-time channel knowledge map construction and channel estimation. SUMMARY

[0005] The present application aims at the problem that the existing channel estimation technology lacks the extraction of channel characteristics from real-time received signals, and proposes a real-time channel knowledge map enabled channel estimation method to realize real-time channel knowledge map (CKM) construction and CKM enabled high-performance channel estimation in MIMO-OFDM system. The method extracts the power delay profile (PDP) and power angle profile (PAP) of the channel from the uplink real-time received signal based on the matching principle, and associates the information with the user location as the user customized channel prior information. Then, two frequency-space serial MMSE (minimum mean square error) channel estimations are proposed using the customized prior information, which realizes a significant performance improvement compared with the existing non-customized statistical prior channel estimation method.

[0006] The technical scheme adopted by the present application is:

[0007] A real-time channel knowledge map enabled channel estimation method, in a communication system in which the receiver is configured as a uniform planar array antenna and performs uplink OFDM (orthogonal frequency division multiplexing) communication on multiple equally spaced subcarriers, performs the following steps:

[0008] Step 1, constructing a communication system model based on channel knowledge map (CKM);

[0009] The physical range served by the receiver is divided into a grid, and a signal model of the frequency-space domain received signal of the probe signal of each grid is constructed.

[0010] Step 2, based on the multiple uniform samplings of the line spectrum corresponding to the frequency domain response of each channel path about the time delay τ, a frequency domain dictionary matrix for extracting the power delay profile (PDP) of the channel is constructed;

[0011] Step 3, based on the matching results of the column vectors of the frequency domain dictionary matrix and the frequency domain residual error of each grid of the physical range served by the receiver, the frequency domain steering vector matrix and the time delay are iteratively updated, and based on the column index of the current matching, the column vector corresponding to the frequency domain dictionary matrix is set to a zero vector; based on the frequency domain steering vector matrix updated after each iteration, the frequency domain channel complex coefficient of each iteration operation is estimated, and the frequency domain channel complex coefficient obtained in each iteration operation is recorded;

[0012] When the maximum number of iterations reaches the preset maximum frequency domain iteration number, the iteration operation is stopped, each iteration is taken as a path, the frequency domain energy value of each path is obtained based on the frequency domain channel complex coefficients, and the power delay profile characteristic information of the channel is obtained based on the frequency domain energy values of all paths and the frequency domain steering vector matrix obtained after the last update;

[0013] Step 4, based on the multiple uniform samplings of the horizontal angle component and the vertical angle component in the angle of the line spectrum corresponding to the spatial domain response of each channel path, a line spectrum sequence about the horizontal angle component and a line spectrum sequence about the vertical angle component are obtained; then, the Kronecker product operation is performed on the two line spectrum sequences to obtain a spatial domain dictionary matrix for extracting the power angle profile (PAP) of the channel;

[0014] Step 5, based on the matching results of the column vectors of the spatial domain dictionary matrix and the spatial domain residual error of each grid of the physical range served by the receiver about the received signal, the spatial domain steering vector matrix and the angle are iteratively updated, and based on the column index of the current matching, the column vector corresponding to the spatial domain dictionary matrix is set to a zero vector; then, based on the spatial domain steering vector matrix updated after each iteration, the spatial domain channel complex coefficient of each iteration operation is estimated, and the spatial domain channel complex coefficient obtained in each iteration operation is recorded;

[0015] When the maximum number of iterations reaches the preset maximum spatial domain iteration number, the iteration operation is stopped, each iteration is taken as a path, the spatial domain energy value of each path is obtained based on the spatial domain channel complex coefficients, and the power angle profile characteristic information of the channel is obtained based on the spatial domain energy values of all paths and the spatial domain steering vector matrix obtained after the last update;

[0016] Step 6, in the channel estimation pilot transmission stage, K users transmit pilot signals at the same frequency resource point, and based on the grid index of the user k, the power delay profile characteristic information and the power angle profile characteristic information of each user are obtained, so as to obtain the frequency domain covariance and the spatial domain covariance of each user;

[0017] Step 7, based on the frequency domain covariance and spatial covariance of each user, the channel estimation result is obtained by using the serial MMSE channel estimation method of frequency domain first and spatial domain second.

[0018] Further, step 7 can be replaced by: based on the frequency domain covariance and spatial covariance of each user, the channel estimation result is obtained by using the serial MMSE channel estimation method of spatial domain first and frequency domain second.

[0019] Further, in step 1, the signal model is specifically:

[0020] For any single grid, the frequency-space domain received signal of the probe signal is represented as:

[0021] Y=XG+ζ

[0022] Where Y represents the frequency-space domain received signal in the grid, X represents the signal diagonal matrix, in the present application, the receiver knows the signal matrix (i.e. X is known to the receiver), Y represents the frequency-space domain received signal, ζ represents an additive white Gaussian noise (AWGN) matrix, the elements of which are independent and subject to Gaussian distribution γ represents the noise variance, C represents the complex domain, M=M x ×M y M represents the number of elements of the uniform planar array antenna, M x , M y are the number of elements in the horizontal and vertical directions of the planar array antenna respectively, and N represents the number of subcarriers.

[0023] Further, in step 2, the construction of the frequency domain dictionary matrix specifically includes:

[0024] Based on the equal interval arrangement property of OFDM frequency domain carriers, each path in the multipath channel presents an equal phase shift characteristic in response to adjacent carriers, and the corresponding line spectrum can be represented as:

[0025]

[0026] Where i is an imaginary symbol, τ∈[0,2π) is a normalized delay, the obtained line spectrum is arranged into a frequency domain dictionary matrix by uniformly sampling the delay τ T (a preset value) times:

[0027]

[0028] Further, in step 3, the matching method of the column vector of the frequency domain dictionary matrix and the residual of each grid with respect to the received signal is:

[0029]

[0030] where d f,n represents the n-th column of the frequency domain dictionary matrix D f f represents the frequency domain residual, whose initial value is R f = Q = X -1 Y, where Q represents the frequency-space domain received signal corresponding to the grid, if the grid has multiple frequency-space domain received signals, then the frequency-space domain received signal corresponding to the grid is obtained by splicing all the frequency-space domain received signals of the current grid.

[0031] Further, in step 3, each iteration operation includes:

[0032] based on the column vector d f of the frequency domain dictionary matrix D f,n matched at present, the frequency domain steering vector matrix is updated as: A f = [A f , d f,n ], that is, the updated frequency domain steering vector matrix is obtained by splicing A f , d f,n , wherein the initial value of the frequency domain steering vector matrix A f is an empty matrix, and n is the column index of the frequency domain dictionary matrix D f ; and the time delay is updated as: wherein the subscript t represents the iteration number; and the n-th column of the frequency domain dictionary matrix D f is set as a zero vector.

[0033] based on the current updated frequency domain steering vector matrix A f , the channel complex coefficient is estimated as: wherein represents the pseudo-inverse operation.

[0034] based on the current updated frequency domain steering vector matrix A f and the current estimated channel complex coefficient (also referred to as channel matrix), the frequency domain residual is updated as: wherein Q represents the frequency-space domain received signal corresponding to the grid.

[0035] Further, in step 3, the frequency domain energy value of each path is: wherein represents the (l, m) element of the estimated channel complex coefficient at the l-th iteration.

[0036] Further, in step 4, the spatial domain dictionary matrix for extracting the power angular spectrum (PAP) of the channel is specifically constructed as:

[0037] The spatial domain response corresponding to the line spectrum of each channel path is:​

[0038]

[0039] where θ and φ represent horizontal and vertical angle components, respectively, M1 and M2 are the number of uniform sampling for θ and φ, respectively, and Dθ and Dφ are the spatial dictionary matrix.

[0040]

[0041] where denotes Kronecker product, and M1 and M2 are preset values.

[0042] Further, in step 5, the column vector of the spatial dictionary matrix and the matching manner of the residual of each grid with respect to the received signal are as follows:

[0043]

[0044] where d s,m denotes the mth column of the frequency domain dictionary matrix D s , and R s denotes the spatial residual, the initial value of which is R s = Q T , where Q denotes the frequency-space domain received signal after removing the pilot corresponding to the grid, and if the grid has multiple frequency-space domain received signals, the frequency-space domain received signal corresponding to the grid is obtained by splicing all the frequency-space domain received signals of the current grid.

[0045] Further, in step 5, each iteration operation includes:

[0046] updating the spatial steering vector matrix A s = [A s , d s,m ] based on the column vector d s,m of the frequency domain dictionary matrix D s matched in the current iteration, and updating the angle where θ and φ represent horizontal and vertical angle components, respectively, subscript t represents the iteration number, and m is the column index of the frequency domain dictionary matrix D s ; and setting the mth column of the spatial dictionary matrix D s to a zero vector.

[0047] updating the spatial residual based on the updated spatial steering vector matrix A s and the estimated channel complex coefficient (also referred to as channel matrix) in the current iteration: where Q denotes the frequency-space domain received signal after removing the pilot corresponding to the grid, denotes pseudo-inverse operation.

[0048] Further, in step 5, the spatial energy value of each path wherein, denotes the estimated channel complex coefficient at the lth iteration of the (l, m)th element of the estimated channel complex coefficient.

[0049] Further, in step 6, the frequency domain covariance and the spatial covariance of each user are specifically:

[0050]

[0051] wherein, A f,k , A s,k are respectively the frequency domain steering vector matrix in the power delay spectrum feature information of user k and the spatial steering vector matrix in the power angle spectrum feature information of user k, ρ f,l,k , ρ s,l,k are respectively the frequency energy value and the spatial energy value of the lth path of user k.

[0052] Further, in step 7, the serial MMSE channel estimation method of frequency domain first and then spatial domain is adopted to obtain the channel estimation result, which specifically includes:

[0053] Five auxiliary quantities are calculated:

[0054]

[0055] The channel estimation of each user is calculated:

[0056]

[0057] wherein, the frequency-space domain pilot receiving signal is denoted as X, the pilot signal diagonal matrix X = diag(X1,..., X K ), the frequency-space domain multi-user channel matrix X k denotes the signal diagonal matrix of user k, G k denotes the frequency-space domain receiving signal of user k, ζ' denotes the thermal noise and interference term, the covariance matrix estimation of which is defined as R I , the energy is defined as (i.e., the variance of ζ'), I denotes the unit matrix, P k denotes the pilot energy of user k, the frequency domain covariance C f = (C f,1 ,..., C f,K ), denotes the operator of taking the diagonal elements of a matrix to form a vector, represents the channel prior variance, the intermediate quantity ||·||1 represents the vector 1 norm.

[0058] Furthermore, in step 7, obtaining a channel estimation result by using a serial MMSE channel estimation method of first spatial domain and then frequency domain specifically includes:

[0059] Calculate four auxiliary quantities:

[0060]

[0061] Calculate the channel estimation matrix:

[0062]

[0063] Among them, the frequency-space domain pilot reception signal S = XG + ζ′, the pilot signal diagonal matrix X = diag (X1, ..., X K ), channel matrix ζ′ represents thermal noise and interference, and its covariance matrix estimate is defined as R I , energy is defined as I represents the identity matrix, P k represents the pilot energy of user k, Z post represent The estimated value of represent The prior covariance of the frequency domain covariance C f =(C f,1 ,...,C f,K ), represents an operator that takes the diagonal elements of a matrix to form a vector,

[0064] The technical solution provided by the present invention brings at least the following beneficial effects:

[0065] This invention extracts user location-related channel statistics, including PDP and PAP, through low-cost and continuous communication signal reception. These statistics reflect user-customized local wireless propagation characteristics, and the corresponding frequency and spatial covariance matrices possess low-dimensional structured subspace features. This invention fills the gap in parameter extraction that is neglected in existing channel knowledge graph technologies. Furthermore, this invention utilizes low-dimensional structured frequency and spatial covariance matrices to design two serial MMSE channel estimation schemes, which achieve significant performance improvements compared to existing non-customized statistical prior channel estimation methods. Furthermore, experiments show that the proposed scheme still has performance advantages over existing MMSE schemes based on historical sample covariance when the pilot overhead is 1 / 4. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments description. Obviously, the drawings in the following description only represent some of the embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative effort.

[0067] Figure 1 Channel knowledge graph for regional grid division;

[0068] Figure 2 MSE curve of channel estimation of different schemes with SINR change;

[0069] Figure 3 MSE curve of channel estimation of different grid sizes under the scheme proposed in the embodiments of the present application with SINR change. DETAILED DESCRIPTION

[0070] In order to make the purpose, technical solutions and advantages of the embodiments of the present application more clear, the following will combine the drawings in the embodiments of the present application to describe the technical solutions in the embodiments of the present application in detail and completely. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Generally, the components of the embodiments of the present application described and shown in the drawings can be arranged and designed using different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not only to limit the scope of the claimed present application, but only to represent selected embodiments of the present application.

[0071] The embodiments of the present application provide a real-time channel knowledge graph enabled channel estimation method to realize real-time channel knowledge graph construction and its enabled high-performance channel estimation under MIMO-OFDM system. In the embodiments of the present application, channel data is obtained by field measurement and collection, containing 192 OFDM subcarriers, carrier spacing 240KHz, carrier frequency 4.9G, transmitting two signals, receiving end antenna number 32 (4x8 uniform surface array), antenna spacing 1 / 2 wavelength. 3000 channel samples are collected along a straight line at pedestrian speed and 5ms interval, about 10 meters of sampling data, a total of two straight line route data are collected, the first segment is used as a double user channel, and the second segment is used as a double interference channel. The 1m grid granularity of the graph corresponds to 300 channel samples. The pilot transmission uses a pseudo-random sequence, the double user uses a random sequence under the same random seed number, and the pilot of user 2 is additionally cyclically shifted by 1 / 4 OFDM period. The pilot generation method of the double interference is similar, and the seed number is different from that of the user. The pilot occupies 1 / 2 of the subcarriers, i.e. the odd bits of the subcarriers.

[0072] According to the above parameter settings, the specific steps of this implementation example are as follows:

[0073] S1, CKM constructs a communication system model: the receiver is equipped with M=4x8 uniform planar antenna, and performs uplink OFDM communication on N=192 equally spaced subcarriers. The receiver service physical range is divided into a grid, as shown in Figure 1 , for a single grid, considering the uplink probe signal in the grid, the frequency-space domain received signal model is:

[0074] Y=XG+N,

[0075] wherein represents the signal diagonal matrix of the grid, and it is assumed that the receiver knows the signal matrix. represents the frequency-space domain received signal, represents an additive white Gaussian noise (AWGN) matrix, and the elements thereof are independent and subject to Gaussian distribution The equivalent received model is obtained by multiplying the received signal Y by the inverse matrix of the pilot matrix, that is:

[0076] Q=X -1 Y=G+X -1 N,

[0077] That is, Q is the frequency-space domain received signal after removing the pilot corresponding to a single network.

[0078] S2, frequency domain dictionary matrix construction for extracting PDP:

[0079] Based on the equally spaced arrangement property of OFDM frequency domain carriers, each path in the multipath channel presents an equal phase shift characteristic in response to adjacent carriers, corresponding to a line spectrum wherein i is an imaginary symbol, τ∈[0,2π) is a normalized delay, the delay τ is uniformly sampled T=2N times, and the obtained line spectrum is arranged into a frequency domain dictionary matrix:

[0080]

[0081] S3, initialize the iteration number t=0, and the frequency domain steering vector matrix A f =[]. When there are multiple received signals in the same grid, the multiple received signals are spliced, and the spliced Q is obtained in the same way as S1, and the residual R f =Q.

[0082] S4, select the line spectrum with the highest matching degree in the frequency domain dictionary matrix and obtain its index as follows:

[0083]

[0084] wherein d f,n represents the nth column of the frequency domain dictionary matrix D f , and ||·||2 represents the 2-norm of the vector.

[0085] S5, update the frequency domain steering vector matrix and time delay:

[0086] A f ←[A f ,d f,n ],

[0087]

[0088] Set the nth column of the frequency domain dictionary matrix D f to be a zero vector.

[0089] S6, obtain the channel complex coefficient estimate:

[0090]

[0091] where represents the pseudo-inverse operation.

[0092] S7, update the frequency domain residual:

[0093]

[0094] And let the iteration number t increase by 1, that is, t = t + 1;

[0095] S8, when the iteration number t reaches the maximum frequency domain iteration number L f = 50, stop the iteration, and output A f as the frequency domain steering vector matrix, output as the energy of the lth path, where represents the (l, m) element of the matrix A f and ρ f,l constitute the PDP. If the iteration is not stopped, continue to execute steps S4-S7 based on the updated iteration number t.

[0096] S9, spatial domain dictionary matrix construction for extracting PAP:

[0097] The spatial domain response of each channel path corresponds to the line spectrum and θ and φ represent the horizontal and vertical angle components, respectively, and θ and φ are uniformly sampled M1 = 2M x and M2 = 2M y times, respectively, to obtain the spatial domain dictionary matrix

[0098]

[0099] where represents the Kronecker product.

[0100] S10, initialize iteration number t = 1, and the spatial domain steering vector matrix A = []. s When there are multiple received signals in the same grid, the multiple received signals are spliced, and the spliced Q and the residual R s = Q T are obtained according to the processing manner of S1.

[0101] S11, select the line spectrum with the highest matching degree with the residual in the spatial domain dictionary matrix, and obtain the index as follows

[0102]

[0103] Wherein, d s,m represents the mth column of the spatial domain dictionary matrix D s .

[0104] S12, update the spatial domain steering vector matrix and the angle:

[0105] A s ← [A s , d s,m ],

[0106]

[0107] The mth column of the spatial domain dictionary matrix D s is set to a zero vector.

[0108] S13, obtain the channel complex coefficient estimation:

[0109]

[0110] S14, update the spatial domain residual:

[0111]

[0112] And let the iteration number t increase by 1, that is, t = t + 1

[0113] S15, when the iteration number t reaches the maximum spatial domain iteration number L s = 20, the iteration is stopped. Output A s as the frequency domain steering vector matrix, and output as the energy of the lth path, wherein represents the (l, n)th element of the matrix . A s and ρ s,l constitute PAP. If the iteration is not stopped, continue to execute steps S11-S14 based on the updated iteration number t.

[0114] S16, considering the pilot transmission stage of channel estimation, unlike step S1, at this time K=2 users transmit pilot signals at the same frequency resource point, and the frequency-space domain pilot received signal is represented as:

[0115]

[0116] wherein the pilot diagonal matrix X = diag (X1,..., X K ), X k represents the pilot diagonal matrix of user k, and the frequency-space domain multi-user channel matrix diag (·) represents diagonalization operation, and the pilot energy of user k is P k ; ζ' represents thermal noise + interference term, and the covariance matrix estimation is R I , and the energy is

[0117] S17, according to the grid index where user k is located, S1-S15 are called based on the PDP and PAP information obtained from the historical probe signals in the same grid index, and the PDP information of user k is represented as A f,k and ρ f,l,k , and the PAP information of user k is represented as A s,l and ρ s,l,k . The frequency domain covariance and the spatial domain covariance of user k are represented as

[0118]

[0119] S18, serial MMSE channel estimation (frequency domain first and then spatial domain) is adopted:

[0120]

[0121] wherein C f = (C f,1 ,..., C f,K ), represents an operator for taking the diagonal elements of a matrix to form a vector, represents the channel prior variance, ||·||1 represents the vector 1 norm.

[0122] S19, the matrix is output as the serial MMSE (frequency domain first and then spatial domain) channel estimation.

[0123] In addition, the embodiment of the application can also adopt serial MMSE channel estimation (spatial domain first and then frequency domain), that is, step S18 and step S19 are replaced by the following two steps:

[0124] (1) serial MMSE channel estimation (spatial domain first and then frequency domain) is adopted:

[0125]

[0126] wherein,

[0127] (2) Output matrix As a serial MMSE (spatial domain first and frequency domain second) channel estimation.

[0128] Figure 2 The curve of the channel estimation MSE changing with SINR is shown, and the real-time channel knowledge graph enabled channel estimation method proposed in the embodiment of the application adopts a 1-meter granularity graph grid, the traditional MMSE scheme adopts a truncation zero method for time domain noise reduction, and the covariance matrix used by the spatial domain MMSE is obtained based on the channel estimation value of the last adjacent sampling position. Figure 2 It can be seen that the scheme proposed in the embodiment of the application is still better than the traditional MMSE scheme in terms of channel estimation performance under the condition of 1 / 4 pilot overhead.

[0129] Figure 3 The performance change of the scheme proposed in the embodiment of the application under different grid granularities is shown, and the grid granularity is in the range of 0-3 meters, and the scheme proposed in the embodiment of the application has a performance difference of about 1dB.

[0130] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

[0131] The above is only some embodiments of the present application. For those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the protection scope of the present application.

Claims

1. A channel estimation method enabled by a real-time channel knowledge graph, characterized in that: In a communication system in which a receiver is configured as a uniform array antenna and performs uplink orthogonal frequency division multiplexing communication on a plurality of equally spaced subcarriers, the following steps are performed: Step 1: Build a communication system model based on the channel knowledge graph; Divide the physical range of the receiver service into grids and construct a signal model of the frequency-space domain received signal of the detection signal of each grid; Step 2: Based on multiple uniform samplings of the line spectrum corresponding to the frequency domain response of each channel path with respect to the time delay τ, a frequency domain dictionary matrix for extracting the power delay spectrum of the channel is constructed; Step 3: Iteratively update the frequency domain steering vector matrix and the delay based on the matching results of the column vectors of the frequency domain dictionary matrix and the frequency domain residual of the received signal for each grid in the physical range served by the receiver. At the same time, based on the current matching column index, set the column vector corresponding to the frequency domain dictionary matrix to the zero vector; then, estimate the frequency domain channel complex coefficients of each iterative operation based on the frequency domain steering vector matrix after each iterative update, and record the frequency domain channel complex coefficients obtained during each iterative operation; When the maximum number of iterations reaches the preset maximum number of frequency-domain iterations, the iteration operation is stopped. Each iteration is regarded as a path, and the frequency-domain energy value of each path is obtained based on the complex coefficients of each frequency-domain channel. Based on the frequency-domain energy values ​​of all paths and the frequency-domain steering vector matrix obtained from the last update, the power delay spectrum characteristic information of the channel is obtained. Step 4: Based on the line spectrum corresponding to the spatial response of each channel path, the horizontal angle component and the vertical angle component of the angle are uniformly sampled multiple times to obtain a line spectrum sequence for the horizontal angle component and a line spectrum sequence for the vertical angle component. The two line spectrum sequences are then subjected to a Kronecker product operation to obtain a spatial dictionary matrix for extracting the power angle spectrum of the channel. Step 5: Iteratively update the spatial steering vector matrix and angle based on the matching results between the column vectors of the spatial dictionary matrix and the spatial residual of the received signal for each grid in the physical range served by the receiver. At the same time, based on the current matching column index, set the column vector corresponding to the spatial dictionary matrix to the zero vector. Then, estimate the spatial channel complex coefficients for each iterative operation based on the spatial steering vector matrix after each iterative update, and record the spatial channel complex coefficients obtained during each iterative operation. When the maximum number of iterations reaches the preset maximum number of spatial iterations, the iterative operation is stopped, and each iteration is regarded as a path. The spatial energy value of each path is obtained based on the complex coefficients of each spatial channel; Based on the spatial energy values ​​of all paths and the spatial steering vector matrix obtained from the last update, the power angle spectrum characteristic information of the channel is obtained; Step 6: During the channel estimation pilot transmission phase, multiple users transmit pilot signals at the same frequency resource point. Based on the grid index where the user is located, the power delay spectrum characteristic information and power angle spectrum characteristic information of each user are obtained, thereby obtaining the frequency domain covariance and spatial domain covariance of each user. Step 7: Based on the frequency domain covariance and spatial domain covariance of each user, a channel estimation result is obtained by using a serial minimum mean square error channel estimation method of frequency domain first and then spatial domain; or a serial minimum mean square error channel estimation method of spatial domain first and then frequency domain is used to obtain the channel estimation result; In step 6, the frequency domain covariance and spatial domain covariance of each user are specifically: Among them, A f,k 、A s,k are the frequency domain steering vector matrix in the power delay spectrum feature information of user k and the spatial domain steering vector matrix in the power angle spectrum feature information, respectively, f,l,k , ρ s,l,k are the frequency domain energy value and spatial domain energy value of the l-th path of user k respectively; Step 7 is as follows: (1) The channel estimation results are obtained by using a serial minimum mean square error channel estimation method that first uses the frequency domain and then the spatial domain. Specifically, the following are included: Calculate five auxiliary quantities: Compute the channel estimate for each user: Among them, the pilot signal received in the frequency and space domain is Signal diagonal matrix X = diag(X1,...,X K ), frequency-space domain multi-user channel matrix X k represents the signal pilot diagonal matrix of user k, G k represents the frequency-space domain received signal of user k, ζ ′ Represents thermal noise and interference terms, and its covariance matrix estimate is defined as R I , energy is defined as I represents the identity matrix, P k represents the pilot energy of user k, and the frequency domain covariance C f =diag(C f,1 ,...,C f,K ), K represents the number of users, represents an operator that takes the diagonal elements of a matrix to form a vector, Represents the channel prior variance, the intermediate quantity ||·||1 represents the vector 1 norm; (2) The channel estimation results are obtained by using a serial minimum mean square error channel estimation method in the spatial domain first and then in the frequency domain, specifically including: Calculate four auxiliary quantities: Calculate the channel estimation matrix: in, represents the channel estimate of user k, k = 1, 2, ..., K, where K represents the number of users.

2. The method according to claim 1, wherein In step 1, the signal model is specifically: For any single grid, the frequency-space domain received signal of the detection signal is expressed as: Y=XG+ζ Among them, Y represents the frequency-space domain received signal within the grid, represents the detection signal diagonal matrix, represents the frequency-space domain channel matrix within the grid, Represents the additive Gaussian white noise matrix, whose elements are independent and obey Gaussian distribution γ represents the noise variance, represents the complex domain, M represents the number of array elements of the uniform array antenna, and N represents the number of subcarriers.

3. The method according to claim 1, wherein In step 3, the column vectors of the frequency domain dictionary matrix are matched with the residual of each grid with respect to the received signal as follows: Among them, d f,n Denotes the frequency domain dictionary matrix D f The nth column, R f Represents the frequency domain residual, whose initial value is R f =Q=X -1 Y, where Q represents the frequency-space domain received signal corresponding to the grid after removing the pilot. If there are multiple frequency-space domain received signals in the grid, all the frequency-space domain received signals of the current grid are spliced ​​together to obtain the corresponding frequency-space domain received signal.

4. The method according to claim 1, wherein In step 3, each iteration operation includes: Based on the frequency domain dictionary matrix D on the current match f The column vector d f,n The iterative update frequency domain steering vector matrix is: A f =[A f ,d f,n ], where the frequency domain steering vector matrix A f The initial value of is an empty matrix, and n is the frequency domain dictionary matrix D f Column index of Update delay Wherein, the subscript t represents the number of iterations; The frequency domain dictionary matrix D f The nth column of is set to zero vector; Based on the current updated frequency domain steering vector matrix A f Estimate the complex channel coefficients: in, represents pseudo-inverse operation; Based on the current updated frequency domain steering vector matrix A f and the current estimated channel complex coefficients Update the frequency domain residual: Where Q represents the frequency-space domain received signal corresponding to the grid.

5. The method according to claim 4, wherein In step 3, the frequency domain energy value of each path in, represents the channel complex coefficient estimated at the lth iteration The (l,m)th element of .

6. The method according to claim 1, wherein In step 4, the spatial domain dictionary matrix for extracting the power angle spectrum of the channel is constructed as follows: The line spectrum corresponding to the spatial response of each channel path is: Among them, θ and φ represent the horizontal and vertical angle components respectively, M x 、M y are the number of array elements in the horizontal and vertical directions of the array antenna, respectively, and the superscript i is the imaginary unit; Uniformly sample θ and φ M1 and M2 times respectively, and get the spatial domain dictionary matrix: in, Represents Kronecker product, where the sampling numbers M1 and M2 are preset values.

7. The method according to claim 1, wherein In step 5, the column vectors of the spatial dictionary matrix are matched with the residual of each grid with respect to the received signal as follows: Among them, d s,m Denotes the frequency domain dictionary matrix D s The mth column, R s Represents the spatial residual, whose initial value is R s =Q T , where Q represents the frequency-space domain received signal corresponding to the grid after removing the pilot. If there are multiple frequency-space domain received signals in the grid, all the frequency-space domain received signals of the current grid are spliced ​​to obtain the corresponding frequency-space domain received signal.

8. The method according to claim 1, wherein In step 5, each iteration operation includes: Based on the frequency domain dictionary matrix D on the current match s The column vector d s,m Iteratively update the spatial steering vector matrix: A s =[A s ,d s,m ], and update the angle Where θ and φ represent the horizontal and vertical angle components respectively, the subscript t represents the number of iterations, and m is the frequency domain dictionary matrix D s The column index of , M1 and M2 are the uniform sampling numbers about θ and φ respectively; The spatial dictionary matrix D s The mth column of is set to zero vector; Based on the currently updated spatial steering vector matrix A s and the current estimated channel complex coefficients Update the spatial residual: Where Q represents the frequency-space domain received signal after removing the pilot signal corresponding to the grid. Represents a pseudo-inverse operation.

Images