Wideband massive MIMO uplink channel estimation method based on hybrid message passing

Abstract

The present invention discloses a method for estimating a broadband massive MIMO uplink channel based on hybrid message passing. Aiming at a broadband massive MIMO system in which base station antennas are configured as uniform arrays, the present invention proposes a triple beam domain channel model and establishes a probability model of a broadband massive MIMO system. Under the framework of Bethe free energy theory, the broadband massive MIMO uplink channel estimation problem is transformed into a Bethe free energy minimization problem. The hybrid message passing algorithm is obtained by solving the Bethe free energy minimization problem using the Lagrange multiplier method, and the broadband massive MIMO uplink channel estimation is implemented using the algorithm. The hybrid message passing channel estimation method in the present invention can effectively improve the accuracy of uplink channel estimation of a broadband massive MIMO system.

Description

Technical Field

[0001] The invention belongs to the technical field of wireless communications and relates to uplink channel estimation of a broadband large-scale MIMO system. Background Art

[0002] Massive Multiple-Input Multiple-Output (MIMO) technology is one of the key technologies for 5G and future wireless communication systems, which can greatly improve the system's spectrum efficiency and transmission reliability. Orthogonal Frequency Division Multiplexing (OFDM) technology can combat inter-symbol interference caused by delay spread to increase data transmission rate, and is the basic framework technology used by broadband massive MIMO systems.

[0003] In a broadband massive MIMO system, accurate channel estimation is the key to achieving effective communication. For a massive MIMO system, because the number of base station antennas increases and the channel matrix dimension increases, traditional uplink channel estimation methods, such as the least squares method (LS), are difficult to obtain accurate channels under limited pilot resources. Therefore, the present invention provides an effective broadband massive MIMO uplink channel estimation method. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a broadband massive MIMO uplink channel estimation method based on hybrid message passing to accurately estimate the uplink spatial frequency domain channel.

[0005] Technical solution: In order to achieve the above-mentioned purpose, the present invention provides the following technical solution:

[0006] The broadband massive MIMO uplink channel estimation method based on hybrid message passing comprises the following steps:

[0007] Establish triple-beam domain channel model for broadband massive MIMO systems;

[0008] Build a receive signal model for broadband massive MIMO systems;

[0009] Build probabilistic models of broadband massive MIMO systems;

[0010] Establish the Bethe free energy minimization problem for broadband massive MIMO systems;

[0011] By solving the Bethe free energy minimization problem, a hybrid message passing algorithm is obtained for channel estimation.

[0012] The base station antenna is configured as a uniform array with an antenna spacing of half a wavelength. The signal transmission of the broadband large-scale MIMO system adopts orthogonal frequency division multiplexing OFDM or its power efficiency improved modulation mode. The maximum guard interval of the system is M t system sampling interval, and the number of users is K.

[0013] The base station selects horizontal angle sampling grid points for the horizontal angle corresponding to the uniform array, the number of horizontal angle sampling grid points is greater than or equal to the number of antennas in each row of the uniform array, and each horizontal angle sampling grid point generates a corresponding horizontal angle refined steering vector.

[0014] The base station selects vertical angle sampling grid points for the vertical angle corresponding to the uniform array, the number of vertical angle sampling grid points is greater than or equal to the number of antennas in each column of the uniform array, and each vertical angle sampling grid point generates a corresponding vertical angle refined steering vector.

[0015] The base station selects a time delay sampling grid point for a one-dimensional time delay, and the number of the time delay sampling grid points is greater than or equal to M t , each delay sampling grid generates a corresponding delay refinement steering vector.

[0016] The base station sequentially arranges all horizontal angle fine steering vectors into a horizontal angle fine steering matrix, sequentially arranges all vertical angle fine steering vectors into a vertical angle fine steering matrix, and sequentially arranges all delay fine steering vectors into a delay fine steering matrix. A triple beam domain steering matrix is ​​formed using the horizontal angle fine steering matrix, the vertical angle fine steering matrix, and the delay fine steering matrix.

[0017] The triple beam domain steering matrix is ​​used to multiply a random vector on the right side of the k-th user to characterize the antenna frequency domain channel of the k-th user of the broadband massive MIMO system, wherein the random vector is a triple beam domain channel vector of the k-th user of the broadband massive MIMO system, wherein k=1, ..., K;

[0018] Each user uses non-overlapping pilot subcarriers to send a pilot signal to the base station. Each user obtains its own pilot subcarrier delay matrix from the delay refinement steering matrix according to the pilot subcarrier set. The triple beam domain perception matrix of the kth user is formed by using the pilot signal of the kth user, the pilot subcarrier delay matrix of the kth user, the horizontal angle refinement steering matrix, and the vertical angle refinement steering matrix, and it is multiplied with the triple beam domain channel vector of the kth user. The obtained vector is expressed as the noise-free received signal vector of the base station corresponding to the kth user, and the noise-free received signal of the base station corresponding to the kth user is added to a noise vector whose elements of the same dimension are independent and identically distributed complex Gaussian random variables. The final vector is used to represent the received signal vector of the base station corresponding to the kth user, where k = 1, ..., K.

[0019] For the triple beam domain channel vector of the kth user, the noise-free received signal vector of the base station corresponding to the kth user, the received signal vector and noise vector of the base station corresponding to the kth user, their elements are all unknown random variables, where k=1,…,K.

[0020] A vector with the same dimension as the triple-beam domain channel vector of the k-th user is introduced and expressed as the triple-beam domain channel support vector of the k-th user. The elements of the triple-beam domain channel support vector of the k-th user are all unknown binary random variables, where k=1,…,K.

[0021] The probability model of the broadband massive MIMO system includes the following four types of probabilities:

[0022] The conditional probability of the received signal vector of the base station corresponding to the k-th user under the noise-free received signal vector of the base station corresponding to the k-th user, where k=1, ..., K;

[0023] The conditional probability of the noise-free received signal vector of the base station corresponding to the k-th user under the triple beam domain channel vector of the k-th user, where k=1, ..., K;

[0024] The conditional probability of the triple beam domain channel vector of the k-th user under the triple beam domain channel support vector of the k-th user, where k=1, ..., K;

[0025] The prior probability of the triple-beam domain channel support vector of the k-th user, where k=1,…,K.

[0026] The four types of probabilities are all decomposed into indivisible probabilities, which are called factors. Corresponding factor belief is introduced for each factor, and corresponding variable belief is introduced for all unknown random variables. Bethe free energy is composed of the entropy and relative entropy of factor belief and variable belief. Factor belief and variable belief need to satisfy first-order moment constraints, second-order moment constraints, marginal consistency constraints and normalization constraints.

[0027] The Bethe free energy minimization problem is solved by the Lagrange multiplier method. Lagrange multipliers are introduced as constraints of the Bethe free energy minimization problem, and the Lagrange equation of the Bethe free energy minimization problem is established. By setting the first-order derivative of the Lagrange equation with respect to factor confidence and variable confidence to zero, an update process with respect to Lagrange multipliers, auxiliary variables and auxiliary probabilities is obtained, and the update process is called hybrid message passing.

[0028] The hybrid message passing algorithm comprises the following steps:

[0029] Input a triple beam domain steering matrix, a known probability parameter, a triple beam domain sensing matrix of a k-th user, and a signal vector received by a base station from the k-th user, where k=1, ..., K;

[0030] Initialize the Lagrange multipliers, auxiliary variables, auxiliary probabilities, and the mean and variance of the variable confidence corresponding to the triple beam domain channel element of the k-th user, where k = 1, ..., K;

[0031] Iteratively calculate Lagrange multipliers, auxiliary variables and auxiliary probabilities, and update the mean and variance of the variable confidence corresponding to the triple beam domain channel element of the kth user, where k = 1, ..., K;

[0032] When a preset condition (such as the maximum number of iterations) is reached, the iteration is stopped, and the mean of the variable confidences corresponding to the triple beam domain channel elements of the k-th user is used as the estimate of the triple beam domain channel elements of the k-th user, where k=1, ..., K;

[0033] The antenna frequency domain channel of the kth user is restored according to the relationship between the triple beam domain channel and the antenna frequency domain channel, where k=1, ..., K.

[0034] Beneficial effect: The triple beam domain channel model, received signal model and probability model of the broadband large-scale MIMO system established by the present invention can effectively support the Bethe free energy minimization theoretical framework to propose a broadband large-scale MIMO uplink channel estimation method based on hybrid message passing, which can effectively improve the accuracy of spatial frequency domain channel estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is a flow chart of a broadband large-scale MIMO uplink channel estimation method based on hybrid message passing of the present invention.

[0036] Figure 2 The performance comparison chart of the hybrid message passing algorithm and other algorithms. DETAILED DESCRIPTION

[0037] The technical solution provided by the present invention will be described in detail below in conjunction with specific embodiments. It should be understood that the following specific implementation methods are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0038] like Figure 1 As shown, the broadband massive MIMO uplink channel estimation method based on hybrid message passing disclosed in an embodiment of the present invention includes the following steps:

[0039] Establish triple-beam domain channel model for broadband massive MIMO systems;

[0040] Build a receive signal model for broadband massive MIMO systems;

[0041] Build probabilistic models of broadband massive MIMO systems;

[0042] Establish the Bethe free energy minimization problem for broadband massive MIMO systems;

[0043] By solving the Bethe free energy minimization problem, a hybrid message passing algorithm is obtained for channel estimation.

[0044] 1. Triple-beam-domain channel model for broadband massive MIMO systems

[0045] Consider the base station antenna configuration as a uniform array, with M antennas. v ×M h , where M v is the number of vertical antennas, M h is the number of horizontal antennas, and the antenna spacing is half a wavelength. The broadband massive MIMO system uses OFDM modulation, and the total number of subcarriers is N c , the number of active subcarriers is N p , the subcarrier spacing is Δ f , the protection interval is M t The number of users is K. Assume that there is I between the kth user and the base station. k The active subcarrier index set of the OFDM system is defined as Then the spatial frequency domain channel of the kth user in the active subcarrier can be expressed as

[0046]

[0047] in, represents the Kronecker product, α i，k represents the complex channel gain, τ i，k represents the path delay, ψ i，k is related to the pitch angle The vertical angle is φ i，k = sinθ i，k , is related to the pitch angle and azimuth The relevant horizontal angle is a h (ψ i，k ), a v (φ i，k ), a t (τ i，k ) are the horizontal angle steering vector, vertical angle steering vector and time delay steering vector respectively, which can be expressed as

[0048]

[0049]

[0050]

[0051] D is performed in the horizontal angle range [-1, 1] h Points are uniformly sampled to determine the horizontal angle sampling grid points, the dth h The horizontal angle sampling grid is expressed as The number of horizontal angle sampling grid points D h Greater than or equal to the number of antennas per row in the uniform array M h , each horizontal angle sampling grid Generate a corresponding horizontal angle refined steering vector Expressed as

[0052]

[0053] D is performed in the vertical angle range [-1, 1] v Points are uniformly sampled to determine the vertical angle sampling grid, the dth v The vertical angle sampling grid is represented by The number of vertical angle sampling grid points D v Greater than or equal to the number of antennas per row in the uniform array M v , each horizontal angle sampling grid Generate a corresponding horizontal angle refined steering vector Expressed as

[0054]

[0055] In the delay range Conduct D t The points are uniformly sampled to determine the delay angle sampling grid points. t The time-delay sampling grid is expressed as The number of delay sampling grid points D t Greater than or equal to M t , each time-delay sampling grid point Generate a corresponding delay refinement steering vector Expressed as

[0056]

[0057] The base station arranges all horizontal angle fine-grained steering vectors in sequence into a horizontal angle fine-grained steering matrix represented as

[0058]

[0059] All vertical angle refined steering vectors are arranged in sequence into a vertical angle refined steering matrix expressed as

[0060]

[0061] All delay refinement steering vectors are arranged in sequence to form a delay refinement steering matrix, which is expressed as

[0062]

[0063] The triple beam domain steering matrix formed by the horizontal angle refinement steering matrix, the vertical angle refinement steering matrix, and the delay refinement steering matrix is ​​expressed as

[0064]

[0065] Then the spatial frequency domain channel of the kth user in the active subcarrier can be re-expressed as

[0066]

[0067] in, is the triple-beam domain channel vector of the kth user in the wideband massive MIMO system.

[0068] 2. Received Signal Model of Wideband Massive MIMO System

[0069] Each user uses non-overlapping pilot subcarriers to send pilot signals to the base station. Each user obtains its own pilot subcarrier delay matrix from the delay refinement steering matrix according to its own pilot subcarrier. Assume that the pilot subcarrier set of the kth user is in, It means round down, and its pilot signal is Then the pilot subcarrier delay matrix of the kth user is expressed as

[0070]

[0071] in, is a permutation matrix used to extract the row corresponding to the pilot subcarrier set of the kth user from the delay refinement steering matrix. Then the received signal vector of the base station corresponding to the kth user is expressed as

[0072]

[0073] Where diag(·) represents diagonalization. The pilot signal of the kth user, the pilot subcarrier delay matrix of the kth user, the horizontal angle refinement steering matrix, and the vertical angle refinement steering matrix form the triple beam domain sensing matrix of the kth user. Where M = M v M h M p , D=D v D h D t ,h k =Φ k g k is the noise-free received signal vector of the base station corresponding to the kth user, n k is the noise vector, n k The elements of are independent and identically distributed complex Gaussian random variables with mean 0 and variance σ.

[0074] 3. Probabilistic Modeling of Wideband Massive MIMO Systems

[0075] For the triple beam channel vector of the kth user, the noise-free received signal vector of the kth user, the received signal vector and the noise vector of the base station corresponding to the kth user, their elements are all unknown random variables. Introduce a vector s with the same dimension as the triple beam channel vector of the kth user k ∈{-1, 1} D×1 , expressed as the triple beam domain channel support vector of the kth user, where the elements of the triple beam domain channel support vector of the kth user are all unknown binary random variables. The probability model of broadband massive MIMO system includes the following four types of probabilities:

[0076] (1) The base station receives the signal vector y corresponding to the kth user k The noise-free received signal vector h corresponding to the kth user at the base station k The conditional probability p(y k |h k ) is expressed as

[0077]

[0078] Among them, CN(h m，k ;y m，k ,σ) represents h m，k Subject to the mean y m，k Complex Gaussian distribution with variance σ, y m，k and h m，k They are y k and h k The mth element of .

[0079] (2) The conditional probability p(h) of the noise-free received signal vector of the base station corresponding to the kth user under the triple beam domain channel vector of the kth user k |g k ) is expressed as

[0080]

[0081] Where δ(·) represents the Dirac function, is Φ k The (m, d)th element of d，k Yes k The d-th element of .

[0082] (3) The triple beam domain channel vector g of the kth user k The triple-beam domain channel support vector s for the kth user k The conditional probability p(g k |s k ; k ) is expressed as

[0083]

[0084] where δ[·] represents the Kroneck function, λ k is the prior variance vector, s d，k , d，k They are k , k The dth element of . d，k =1, g d，k is 0 mean λ d，k A complex Gaussian random variable with variance , when s d，k = -1, g d，k is the zero channel element.

[0085] (4) The triple beam domain channel support vector s of the kth user k The prior probability p(s k )for

[0086]

[0087] This represents a Markov random field prior, where exp(·) represents an exponential function with the natural constant e as the base, and Z s，k is the normalization constant, β is the coupling parameter, For d，k The neighbor set of

[0088]

[0089] Here, ||·||2 represents the l2 norm.

[0090] From the above four types of probabilities, we can get the target conditional posterior probability p(g d，k |y k ; k )for

[0091]

[0092]

[0093] Among them, ∝ represents real number proportionality, g k \g d，k Indicates that from g k Remove the element g from d，k . g d，k The minimum mean square error estimate of the conditional posterior probability p(g d，k |y k ; k ). Because the conditional posterior probability p(g d，k |y k ; k ) requires a large number of integrals, which is a complex and difficult task. Therefore, we introduce Bethe free energy minimization theory to approximate the conditional posterior probability p(g d，k |y k ; k ).

[0094] 4. Bethe Free Energy Minimization Problem for Wideband Massive MIMO Systems

[0095] Decompose all four types of probabilities into indivisible probabilities, expressed as a set:

[0096] The indivisible probability is called a factor. For each factor, a corresponding factor belief is introduced, using b Y，m，k (h m，k ), b H，m，k (hm，k , g k ), b G，d，k (g d，k ,s d，k ), b S，i，d，k (s i，k ,s d，k ) represent f Y，m，k (h m，k ), f H，m，k (h m，k , g k ), f G，d，k (g d，k ,s d，k ), f S，i，d，k (s i，k ,s d，k ) is the factor confidence. All random variables are represented by a set {h m，k , g d，k ,s d，k |m=1,…,M,d=1,…,D}, introduce the corresponding variable belief for each random variable, and use q H，m，k (h m，k ),q G，d，k (g d，k ),q S，d，k (s d，k ) represent h m，k , g d，k 、s d，k The Bethe free energy is composed of the entropy and relative entropy of the factor confidence and variable confidence and is expressed as

[0097]

[0098] Where D[·||·] and H[·] represent relative entropy and entropy, respectively. We need to minimize the Bethe free energy F B，k And q H，m，k (h m，k ), b Y，m，k (h m，k ), b H，m，k (h m，k , g k ) needs to satisfy the following mean constraints and variance constraints

[0099]

[0100]

[0101] q G，d，k (g d，k ), b H，m，k (h m，k , gk ), b G，d，k (g d，k ,s d，k ) needs to satisfy the following mean constraints and variance constraints

[0102]

[0103]

[0104] q S，d，k (s d，k ), b G，d，k (g d，k ,s d，k ), b S，i，d，k (s i，k ,s d，k ) needs to satisfy the following edge consistency constraints

[0105]

[0106] At the same time, all factor confidence and variable confidence need to satisfy the following normalization constraints

[0107]

[0108]

[0109] in, Indicates h m，k In confidence The expectations below.

[0110] 5. Solve the Bethe free energy minimization problem and obtain the hybrid message passing algorithm

[0111] The Bethe free energy minimization problem is solved by the Lagrange multiplier method. Lagrange multipliers are introduced as constraints for the Bethe free energy minimization problem, and the Lagrange equation for the Bethe free energy minimization problem is established. The spatial frequency domain channel part of the Lagrange equation is

[0112]

[0113] Among them, Re{·} means taking the real part, (·) * represents conjugation, is the Lagrange multiplier, and the triple beam channel part of the Lagrangian equation is

[0114]

[0115] in, is the Lagrange multiplier, and the triple beam domain channel support part of the Lagrange equation is

[0116]

[0117] in, is the Lagrange multiplier, and the channel normalization part of the Lagrange equation is

[0118]

[0119] in, is the Lagrange multiplier. The overall Lagrange equation can be expressed as

[0120] L B，k =F B，k +L H，k +L G，k +L s，k +L N，k .

[0121] By setting the first-order derivative of the Lagrange equation with respect to factor confidence and variable confidence to zero, the update process of the auxiliary variables with respect to the Lagrange multiplier and its combination is obtained. The update process is called hybrid message passing. Finally, the mean of the variable confidence corresponding to the triple beam domain channel element of the kth user is used. As the estimate of the triple beam domain channel element for the kth user Where d = 1, ..., D, k = 1, ..., K. The hybrid message passing algorithm includes the following steps:

[0122] Step 1: Input triple beam steering matrix The known probability parameters include the noise variance σ and the Markov random field coupling parameter β, the triple beam domain sensing matrix Φ of the kth user k , the base station receives the signal vector y of the kth user k , where k = 1, ..., K;

[0123] Step 2: Initialize the Lagrange multipliers Auxiliary variables Auxiliary Probability The variable confidence corresponding to the triple beam domain channel element of the kth user The mean and variance in, d=1,...,D,k=1,...,K;

[0124] Step 3: Iteratively calculate the Lagrange multiplier, auxiliary variables and auxiliary probabilities, and update the mean and variance of the variable confidence corresponding to the triple beam domain channel element of the kth user, where m = 1, ..., M, d = 1, ..., D, k = 1, ..., K. The specific steps of step 3 iteration include:

[0125] Step 3.1: Update the Lagrange multiplier Where, d=1,…,D, k=1,…,K;

[0126] Step 3.2: Update the Lagrange multiplier Wherein, m=1, ..., M, k=1, ..., K;

[0127] Step 3.3: Update the Lagrange multiplier Wherein, m=1, ..., M, k=1, ..., K;

[0128] Step 3.4: Update auxiliary variables Wherein, m=1, ..., M, k=1, ..., K;

[0129] Step 3.5: Update auxiliary variables Where, d=1,…,D, k=1,…,K;

[0130] Step 3.6: Update the Lagrange multiplier Where, d=1,…,D, k=1,…,K;

[0131] Step 3.7: Update the Lagrange multiplier Where, d=1,…,D, k=1,…,K;

[0132] Step 3.8: Update sparse probabilities Where, d=1,…,D, k=1,…,K;

[0133] Step 3.9: Update variable confidence The mean Where, d=1,…,D, k=1,…,K;

[0134] Step 3.10: Update variable confidence Variance Where, d=1,…,D, k=1,…,K;

[0135] Step 3.11: Update the prior variance Where, d=1,…,D, k=1,…,K;

[0136] Step 3.12: Repeat steps 3.1 to 3.11 until the inner loop stop condition is reached;

[0137] Step 3.13: Update auxiliary probabilities Where, d=1,…,D, k=1,…,K;

[0138] Step 3.14: Update auxiliary probabilities in, d=1,...,D,k=1,...,K;

[0139] Step 3.15: Update auxiliary probabilities in, d=1,...,D,k=1,...,K;

[0140] Step 3.16: Update sparse probabilities Where, d=1,…,D, k=1,…,K;

[0141] Step 3.17: Update sparse probabilities Where, d=1,…,D, k=1,…,K;

[0142] Step 3.18: Update variable confidence The mean Where, d=1,…,D, k=1,…,K;

[0143] Step 3.19: Update the variable confidence q G，d，k (g d，k ) Where, d=1,…,D, k=1,…,K;

[0144] Step 3.20: Repeat steps 3.1 to 3.19 until the outer loop stop condition is reached;

[0145] Step 4: When the preset condition is reached, the iteration is stopped and the mean of the variable confidence corresponding to the triple beam domain channel element of the kth user is As the estimate of the triple beam domain channel element for the kth user Where, d=1,…,D, k=1,…,K;

[0146] Step 5: Recover the antenna frequency domain channel estimate of the kth user based on the relationship between the triple beam domain channel and the antenna frequency domain channel in, k=1,…,K,[·] T Indicates taking the transpose.

[0147] VI. Implementation Effect

[0148] In order to enable those skilled in the art to better understand the solution of the present invention, a comparison of the channel estimation performance results of the hybrid message passing algorithm in this embodiment and the existing method under a specific system configuration is given below.

[0149] Considering the half-wavelength uniform array configured in the base station of the massive MIMO system, the number of vertical antennas of the uniform array is M v=8, number of horizontal antennas M h =8, center frequency 2.4GHz, total number of subcarriers N c =512, number of active subcarriers N p =300, subcarrier spacing Δ f =15kHz, protection interval M t =36, the number of users K = 10, the number of pilot subcarriers per user M p =30, each user uses all 1 pilots, the number of vertical angle sampling grid points is D v =16, the number of horizontal angle sampling grid points D h =16, the number of delay sampling grid points D t =72. The comparison methods include the NIHT method proposed in the literature "Normalized iterative hard thresholding: Guaranteed stability and performance, in IEEE Journal of Selected Topics in Signal Processing, T. Blumensath and MEDavies, Apr. 2010, pp. 298-309" and the EM-GM-AMP method proposed in "Expectation-maximization Gaussian-mixture approximate message passing, in IEEE Transactions on Signal Processing, JPVila and P. Schniter, Oct. 2013, pp. 4658-4672". Performance comparison Figure 2 As shown, it can be seen that the hybrid message passing algorithm achieves more accurate channel estimation performance under all signal-to-noise ratios (SNRs).

[0150] In the embodiments provided in this application, it should be understood that the disclosed methods can be implemented in other ways without exceeding the spirit and scope of this application. The current embodiment is only an illustrative example and should not be used as a limitation. The specific content given should not limit the purpose of this application. For example, some features can be ignored or not executed.

[0151] The technical means disclosed in the scheme of the present invention are not limited to the technical means disclosed in the above-mentioned implementation mode, but also include technical schemes composed of any combination of the above-mentioned technical features. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications are also regarded as the protection scope of the present invention.

Claims

1. A broadband massive MIMO uplink channel estimation method based on hybrid message passing, characterized in that: The method comprises the following steps: Establish triple-beam domain channel model for broadband massive MIMO systems; Build a receive signal model for broadband massive MIMO systems; Build probabilistic models of broadband massive MIMO systems; Establish the Bethe free energy minimization problem for broadband massive MIMO systems; Solve the Bethe free energy minimization problem and obtain the hybrid message passing algorithm for channel estimation; The triple beam channel model of broadband massive MIMO system includes the base station antenna configuration as a uniform array with half-wavelength antenna spacing, and the number of antennas in each row of the uniform array is M. h , the number of antennas in each column of the uniform array is M v The signal transmission of the broadband large-scale MIMO system adopts orthogonal frequency division multiplexing OFDM or its power efficiency improved modulation method, and the maximum guard interval of the system is M t system sampling intervals, the number of users is K, and there are I k The delay of the i-th path between the base station and the k-th user is τ i，k , where k=1,…,K, i=1,…,I k ; The horizontal angle ψ of the i-th path of the k-th user corresponding to the uniform array i，k The pitch angle θ of the i-th path of the k-th user corresponding to the uniform array i，k and azimuth Regarding, specifically The vertical angle φ of the i-th path of the k-th user corresponding to the uniform array i，k The pitch angle θ of the i-th path of the k-th user corresponding to the uniform array i，k Related, specifically φ i，k = sinθ i，k , where k=1,…,K, i=1,…,I k ; The triple beam channel model of broadband massive MIMO system includes the horizontal angles {ψ i，k ,k=1,…,K,i=1,…,I k }Select D from the union of the value ranges of h horizontal angle sampling grid Among them, d h =1,…,D h , horizontal angle sampling grid The number of D h Greater than or equal to the number of antennas per row in the uniform array M h , each horizontal angle sampling grid Generate a corresponding horizontal angle refined steering vector The vertical angle {φ i，k ,k=1,…,K,i=1,…,I k }Select D from the union of the value ranges of v Vertical angle sampling grid Vertical angle sampling grid The number of D v Greater than or equal to the number of antennas in each column of the uniform array M v , each vertical angle sampling grid Generate a corresponding vertical angle refined steering vector Among them, d v =1,…,D v ; The delay of the base station on all user paths {τ i，k ,k=1,…,K,i=1,…,I k The delay sampling grid point is selected within the union of the value range of Delayed sampling grid The number of D t Greater than or equal to M t , each time-delay sampling grid point Generate a corresponding delay refinement steering vector Among them, d t =1,…,D t ; The base station refines all horizontal angles into vectors Arranged as horizontal angle refined steering matrix A h ; The base station refines all vertical angles into vectors Arranged as vertical angle refined guide matrix A v ; The base station will fine-tune the entire delay to the vector Arranged in sequence as the delay refinement steering matrix A t ; Use the horizontal angle to refine the steering matrix A h , vertical angle refinement steering matrix A v , Delay refinement steering matrix A t The Kroneck product forms a triple beam-domain steering matrix The triple beam steering matrix Multiplying the k-th user by a random vector on the right to represent the antenna frequency domain channel of the k-th user of the broadband massive MIMO system, wherein the random vector is a triple beam domain channel vector of the k-th user of the broadband massive MIMO system, wherein k=1, ..., K; The receiving signal model of the broadband massive MIMO system includes each user sending a pilot signal to the base station using non-overlapping pilot subcarriers; each user refines the steering matrix A from the delay according to the pilot subcarrier set t Obtaining respective pilot subcarrier delay matrices; using the pilot signal of the kth user, the pilot subcarrier delay matrix of the kth user and the horizontal angle refinement steering matrix A h , vertical angle refinement steering matrix A v Form a triple beam domain sensing matrix of the kth user, and multiply it by the triple beam domain channel vector of the kth user, express the obtained vector as the noise-free received signal vector of the base station corresponding to the kth user, and add the noise-free received signal of the base station corresponding to the kth user to a noise vector of independent and identically distributed complex Gaussian random variables of the same dimension, and use the finally obtained vector to represent the received signal vector of the base station corresponding to the kth user, wherein k=1, ..., K; In the probability model of the broadband massive MIMO system, a vector with the same dimension as the triple beam domain channel vector of the k-th user is introduced, which is expressed as the triple beam domain channel support vector of the k-th user. The elements of the triple beam domain channel support vector of the k-th user are all unknown binary random variables, where k=1, ..., K. The probability model of the broadband massive MIMO system includes the following four types of probabilities: The conditional probability of the received signal vector of the base station corresponding to the k-th user under the noise-free received signal vector of the base station corresponding to the k-th user, where k=1, ..., K; The conditional probability of the noise-free received signal vector of the base station corresponding to the k-th user under the triple beam domain channel vector of the k-th user, where k=1, ..., K; The conditional probability of the triple beam domain channel vector of the k-th user under the triple beam domain channel support vector of the k-th user, where k=1, ..., K; The prior probability of the triple beam domain channel support vector of the kth user, where k = 1, ..., K; The Bethe free energy minimization problem is specifically to decompose all the four types of probabilities into indivisible probabilities, which are called factors. A corresponding factor belief is introduced for each factor, and a corresponding variable belief is introduced for all unknown random variables. The Bethe free energy is composed of the entropy and relative entropy of the factor belief and the variable belief. The factor belief and the variable belief need to satisfy the first-order moment constraint, the second-order moment constraint, the marginal consistency constraint and the normalization constraint. The Bethe free energy minimization problem is solved by the Lagrange multiplier method; Lagrange multipliers are introduced as constraints of the Bethe free energy minimization problem, and the Lagrange equation of the Bethe free energy minimization problem is established; by setting the first-order derivative of the Lagrange equation with respect to factor confidence and variable confidence to zero, an update process with respect to the Lagrange multiplier, auxiliary variables and auxiliary probabilities is obtained, and the update process is called hybrid message passing; The hybrid message passing algorithm consists of the following steps: Input a triple beam domain steering matrix, a known probability parameter, a triple beam domain sensing matrix of a k-th user, and a signal vector received by a base station from the k-th user, where k=1, ..., K; Initialize the Lagrange multipliers, auxiliary variables, auxiliary probabilities, and the mean and variance of the variable confidence corresponding to the triple beam domain channel element of the k-th user, where k = 1, ..., K; Iteratively calculate Lagrange multipliers, auxiliary variables and auxiliary probabilities, and update the mean and variance of the variable confidence corresponding to the triple beam domain channel element of the kth user, where k = 1, ..., K; When a preset condition (such as a maximum number of iterations) is reached, the iteration is stopped, and the mean of the variable confidences corresponding to the triple beam domain channel elements of the k-th user is used as the estimate of the triple beam domain channel elements of the k-th user, where k=1, ..., K; The antenna frequency domain channel of the kth user is restored according to the relationship between the triple beam domain channel and the antenna frequency domain channel, wherein k=1, ..., K.

2. The method for broadband massive MIMO uplink channel estimation based on hybrid message passing according to claim 1, characterized in that: For the triple beam domain channel vector of the kth user, the noise-free received signal vector of the base station corresponding to the kth user, the received signal vector and noise vector of the base station corresponding to the kth user, their elements are all unknown random variables, where k=1,…,K.

Images