A Dictionary Learning-Based Channel Estimation Method for Ultra-Large MIMO Mixed Fields

Abstract

The application provides a super-large-scale MIMO mixed field channel estimation method based on dictionary learning, establishes a mixed field channel scene, the scene comprises a base station, at least one near-field scatterer, at least one far-field scatterer and K users; a mixed field channel model is constructed according to the mixed channel scene; and a mixed field channel recovery problem is established according to the mixed field channel model and solved. The application can well capture the channel characteristics of the mixed field channel and realize accurate estimation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless communication, and more particularly to a super large scale MIMO mixed field channel estimation method based on dictionary learning. BACKGROUND

[0002] In the existing far field and near field channel models, it is assumed that all scatterers are in the far field or near field region. However, in actual scenarios, the scatterers usually appear partly in the far field region and partly in the near field region. The existing super large scale MIMO mixed field channel estimation mostly uses the far field dictionary and the near field dictionary for channel estimation respectively for the far field channel and the near field channel, but these methods are not suitable for this mixed field communication scenario. In addition, in this mixed field channel model, the far field dictionary and the near field dictionary only use the characteristics of the far field channel or the near field channel, and cannot accurately represent the characteristics of the mixed field channel. SUMMARY

[0003] The present application provides a super large scale MIMO mixed field channel estimation method based on dictionary learning to overcome the defects of the existing mixed field channel estimation methods that cannot well estimate the channel characteristics.

[0004] To solve the above technical problems, the technical solutions of the present application are as follows:

[0005] The present application provides a super large scale MIMO mixed field channel estimation method based on dictionary learning, comprising the following steps:

[0006] A mixed field channel scene is established, which includes a base station, at least one near field scatterer, at least one far field scatterer and K users;

[0007] A mixed field channel model is constructed according to the mixed channel scene;

[0008] A mixed field channel recovery problem is established according to the mixed field channel model and is solved.

[0009] Preferably, constructing a mixed field channel model according to the mixed channel scene comprises:

[0010] y=Ph hybrid +n

[0011] wherein y represents the received signal of the user, wherein N RF represents the number of base station radio frequency chains, N RF <<N, wherein N represents the number of antennas equipped by the base station, S represents the number of time periods, (·) T represents transposition, represents the overall observation matrix, p S represents the transmission pilot sequence of the user in the S time slot, denotes the configuration model matrix of the base station in the S-th time slot, h hybrid denotes the mixed field channel, denotes the noise, h hybrid The calculation formula is:

[0012]

[0013] wherein, L f and L n are the number of far field and near field path components, respectively, L = L f + L n is the total number of path components, and denote the gain and angle of the l f th far field path component, respectively, d and λ c denote the antenna spacing and carrier wavelength, respectively, d = λ c / 2, is the actual physical angle, denotes the steering vector of the l f th far field path component, and is specifically:

[0014]

[0015] wherein, (·) H denotes the conjugate transpose;

[0016] β ln denotes the gain of the l n th path component, θ ln and r ln denote the angle and distance of the l n th near field scattering distance from the center of the antenna array, denotes the distance from the l n th near field scatterer to the n th antenna of the base station,

[0017] Preferably, the mixed field channel recovery problem is established according to the mixed field channel model and solved, comprising:

[0018] establishing a mixed field channel recovery problem;

[0019] transforming the mixed field channel recovery problem into a dictionary learning problem P1;

[0020] transforming the dictionary learning problem P1 into a dictionary learning problem P2;

[0021] transforming the dictionary learning problem P2 into a dictionary learning problem P3;

[0022] Solving the dictionary learning problem P3.

[0023] Preferably, the establishing the mixed field channel recovery problem comprises:

[0024] Converting the mixed field channel h hybrid into a sparse representation form:

[0025]

[0026] wherein, is the sparse representation of h hybrid , is to be estimated, and D is the corresponding mixed dictionary.

[0027] The mixed field channel sparse recovery problem is represented as:

[0028]

[0029] wherein, represents the L0 norm, and represents the number of non-zero items of a vector .

[0030] Preferably, the converting the mixed field channel recovery problem into the dictionary learning problem P1 comprises:

[0031]

[0032] wherein,

[0033] Preferably, the converting the dictionary learning problem P1 into the dictionary learning problem P2 comprises:

[0034] According to the separable structure property of , we have wherein represents the Kronecker product, and the dictionary learning problem P1 is further converted into:

[0035]

[0036] wherein, Y = [y1, …, y K ], y K represents the received signal of the Kth user, represents the sparse representation of the mixed field channel of the Kth user.

[0037] Preferably, the converting the dictionary learning problem P2 into the dictionary learning problem P3 comprises:

[0038]

[0039] wherein,

[0040] Preferably, solving the dictionary learning problem P3 comprises:

[0041] The dictionary learning problem P3 is about A, B and are non-convex, respectively optimizing the dictionaries and dictionaries A and B, the specific steps are as follows:

[0042] S51: initialization dictionary k = 1, the maximum number of iterations M, calculation;

[0043] S52: calculate the kth iteration of the dictionary

[0044] Fixing the dictionary A, the dictionary B, the dictionary learning problem P3 is transformed into an optimization problem about the dictionary :

[0045]

[0046] Sparse representation is obtained by using the orthogonal matching pursuit algorithm According to get Where reshape(·) represents the matrix rearrangement function;

[0047] S53: update to Calculate the kth iteration of the dictionary A (k) ;

[0048] Fixing the sparse representation Dictionary B, the dictionary learning problem P3 is transformed into an optimization problem about the dictionary A:

[0049]

[0050] The problem is solved by using the gradient projection method, A (k) :

[0051]

[0052] Where normalize represents a matrix whose columns are all unit Euclidean norm.

[0053] S54: update A (k) to A, calculate the kth iteration of the dictionary B (k) ;

[0054] Fixing the sparse representation Dictionary A, the dictionary learning problem P3 is transformed into an optimization problem about the dictionary B:

[0055]

[0056] Gradient projection method is used to solve the problem, B (k) is:

[0057]

[0058] S55: determine whether k >= M, if yes, execute step S57, otherwise execute step S56;

[0059] S56: update B (k) , execute k' = k + 1, and update k' to k, and return to step S52;

[0060] S57: obtain A (M) , B (M) , obtain sparse representation by using orthogonal matching pursuit algorithm according to A (M) , B (M) , and h hynrid .

[0061] Preferably, in the step S52, the obtaining sparse representation by using orthogonal matching pursuit algorithm comprises:

[0062]

[0063] y temp = vectorize(Y)

[0064]

[0065] Wherein, vectorize(·) represents vectorization operation, OMP represents orthogonal matching pursuit algorithm, and s represents sparse factor.

[0066] The application also provides a mixed intelligent reflecting surface auxiliary based on all-in-one system for realizing the above method, comprising:

[0067] A scene establishing module is used to establish a mixed field channel scene, the scene comprising a base station, at least one near field scatterer, at least one far field scatterer and K users;

[0068] A model constructing module is used to construct a mixed field channel model according to the mixed channel scene;

[0069] A problem solving module is used to establish a mixed field channel recovery problem and solve it according to the mixed field channel model.

[0070] Compared with the prior art, the technical scheme of the application has the beneficial effects that:

[0071] The application provides a super large scale MIMO mixed field channel estimation method based on dictionary learning, a mixed field channel scene is established, the scene comprises a base station, at least one near field scatterer, at least one far field scatterer and K users; a mixed field channel model is constructed according to the mixed channel scene; and a mixed field channel recovery problem is established according to the mixed field channel model and is solved. The application can well capture the channel characteristics of the mixed field channel and realize accurate estimation. BRIEF DESCRIPTION OF DRAWINGS

[0072] Figure 1 A flowchart of the super large scale MIMO mixed field channel estimation method based on dictionary learning described in embodiment 1 is shown in the figure.

[0073] Figure 2 A system block diagram of the super large scale MIMO mixed field channel estimation method based on dictionary learning described in embodiment 2 is shown in the figure.

[0074] Figure 3 A comparison chart of channel estimation performance of different dictionaries in various signal-to-noise ratios described in embodiment 2 is shown in the figure.

[0075] Figure 4 A comparison chart of channel estimation performance of different dictionaries in various distances of users or scattering clusters described in embodiment 2 is shown in the figure.

[0076] Figure 5 A comparison chart of channel estimation performance of different dictionaries in various pilot lengths described in embodiment 2 is shown in the figure.

[0077] Figure 6 A structure schematic diagram of the super large scale MIMO mixed field channel estimation system based on dictionary learning described in embodiment 3 is shown in the figure. DETAILED DESCRIPTION

[0078] The drawings are only used for illustrative description and cannot be understood as limitation to the patent;

[0079] In order to better illustrate the embodiments, some components in the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product;

[0080] It is understandable that some well-known structures in the drawings and their descriptions may be omitted for those skilled in the art.

[0081] The technical solutions of the application will be further described below in combination with the drawings and embodiments.

[0082] Embodiment 1

[0083] The embodiment provides a super large scale MIMO mixed field channel estimation method based on dictionary learning, as shown in the figure, comprising the following steps: Figure 1

[0084] ​S1: a mixed field channel scene is established, the scene includes one base station, at least one near field scatterer, at least one far field scatterer and K users;

[0085] S2: a mixed field channel model is constructed according to the mixed channel scene;

[0086] S3: a mixed field channel recovery problem is established according to the mixed field channel model and is solved.

[0087] In the specific implementation process, first, a mixed field channel scene is established, then a mixed field channel model is constructed according to the mixed channel scene, finally, a mixed field channel recovery problem is established according to the channel model and the channel characteristics of the mixed field channel are obtained by solving.

[0088] Embodiment 2

[0089] The application provides a super large scale MIMO mixed field channel estimation method based on dictionary learning, comprising the following steps:

[0090] S1: a mixed field channel scene is established, the scene includes one base station, at least one near field scatterer, at least one far field scatterer and K users;

[0091] As shown in the figure, it is a system block diagram of the super large scale MIMO mixed field channel estimation method based on dictionary learning. Figure 2

[0092] S2: a mixed field channel model is constructed according to the mixed channel scene;

[0093] The signal model and channel model of the uplink mixed field of the super large scale MIMO communication system are considered. The base station is equipped with N antennas. Due to the different positions of the scatterers in the mixed field channel scene, not only the near field path component exists, but also the far field path component exists. The antenna spacing is Where λ c is the carrier wavelength. The base station is equipped with a mixed wave number forming structure of N RF strip radio frequency chains, N RF <<N, wherein N represents the number of antennas equipped by the base station, p=[p1,p1,…,p s ,…,p S ], represents the transmission pilot sequence of the user in S time periods, In the s time period, the received signal of the Kth user can be expressed as:

[0094] y s =D s hp s +D s n s ​

[0095] wherein, denotes the configurable analog combining matrix of the base station in the s-th time period, denotes the uplink channel of the user to the base station.

[0096] In the actual communication scenario, the near-field region and the far-field region are determined by the Rayleigh distance Z = 2R 2 / λ c , where R represents the aperture of the antenna array. When the distance between the scatterer and the base station exceeds the Rayleigh distance, the far-field path component is introduced; when the distance between the scatterer and the base station is less than the Rayleigh distance, the near-field path component is introduced. Due to the limited number of scatterers, the far-field channel is sparse in the angular domain, and the near-field channel is sparse in the polar domain. By utilizing the sparsity of the channel, the traditional channel state information-based method can obtain the two channels respectively. The channel in the mixed field communication system can be expressed as:

[0097] h = h f + h n

[0098] The far-field channel path component h f can be expressed as:

[0099]

[0100] wherein, L f denotes the number of far-field channel path components. α l and θ l respectively represent the gain and angle of the l-th path component. Due to the consumption of the far-field plane wave front, the steering vector of the l-th path component a(θ l ) can be expressed as:

[0101]

[0102] wherein, d and λ c respectively represent the antenna spacing and the carrier wavelength. Wherein, φ l ∈(0, π) is the actual physical angle. Then, the non-sparse far-field channel h f is converted into the sparse angular domain representation as:

[0103]

[0104] wherein, denotes the discrete Fourier transform matrix. D f contains N orthogonal steering vectors uniformly sampled from the entire angular space, D f = [a(θ0), …, a(θ N-1 )], wherein n = 0, 1, …, N-1. for the corresponding sparse angular domain channel.

[0105] Similarly, h n can be expressed as:

[0106]

[0107] where L n denotes the number of path components of the near-field channel. β l denotes the gain of the l-th path component. Due to the assumption of near-field spherical wavefront, the steering vector of the l-th path component b(θ l ,r l ) can be expressed as:

[0108]

[0109] where θ l and r l denote the angle and distance of the l-th near-field scatterer to the center of the antenna array. and θ l denote the distance of the l-th near-field scatterer to the n-th antenna of the base station, n = 1, 2, …, N. Similar to the far-field channel, the non-sparse near-field channel h n can be converted to a sparse representation in the polar domain:

[0110]

[0111] where h is the sparse near-field polar domain channel, D n is the near-field polar domain dictionary, and is expressed as:

[0112]

[0113] where the sampling angles θ n are uniformly sampled within , s n = 1, 2, …, S N denotes the number of sampling distances of the sampling angle at θ n , and In order to ensure that the coherence between any two columns in D n is low enough, the sampling is non-uniform for the distance .

[0114] According to the mixed channel scenario, a mixed field channel model is constructed:

[0115] y = Ph hybrid + n

[0116] where y represents the received signal of the user, where N RF represents the number of base station radio frequency chains, N RF << N, S represents the number of time periods, (·) T represents the transpose, represents the overall observation matrix, p S represents the transmission pilot sequence of the user in the S time slot, represents the configuration simulation combination matrix of the base station in the S time slot, where N represents the number of antennas equipped by the base station, h hybrid represents the mixed field channel, represents the noise, h hybrid The calculation formula is:

[0117]

[0118] where L f and L n are the number of far field and near field path components, respectively, L = L f + L n is the total number of path components, and represent the gain and angle of the l f th far field path component, respectively, d and λ c represent the antenna spacing and carrier wavelength, respectively, d = λ c / 2, is the actual physical angle, represents the steering vector of the l f th far field path component, and is specifically:

[0119]

[0120] where (·) H represents the conjugate transpose;

[0121] represents the gain of the l n th path component, and represent the angle and distance of the l n th near field scatterer from the center of the antenna array, represents the distance of the l n th near field scatterer to the n th antenna of the base station,

[0122] S3: Establishing a mixed field channel recovery problem according to the mixed field channel model and solving it.

[0123] The mixed-field channel recovery problem is established and solved, including:

[0124] (1) Establishing the mixed-field channel recovery problem

[0125] The mixed-field channel h hybrid is converted into a sparse representation form:

[0126]

[0127] wherein, is the sparse representation of h hybrid , is to be estimated, and D is the corresponding mixed dictionary;

[0128] The mixed-field channel sparse recovery problem is represented as:

[0129]

[0130] wherein, represents the L0 norm, and represents the number of non-zero items of a vector .

[0131] (2) Converting the mixed-field channel recovery problem into a dictionary learning problem P1

[0132]

[0133] wherein,

[0134] (3) Converting the dictionary learning problem P1 into a dictionary learning problem P2

[0135] According to the separable structure characteristics of , we obtain wherein represents the Kronecker product, and the dictionary learning problem P1 is further converted into:

[0136]

[0137] wherein Y = [y1, …, y K ], y K represents the received signal of the Kth user, and represents the sparse representation of the mixed-field channel of the Kth user.

[0138] (4) Converting the dictionary learning problem P2 into a dictionary learning problem P3

[0139]

[0140] wherein,

[0141] (5) solving the dictionary learning problem P3.

[0142] The dictionary learning problem P3 is about A, B and are all non-convex, respectively optimizing the dictionaries and the dictionaries A and B, the specific steps are as follows:

[0143] S51: initialization dictionary k = 1, the maximum number of iterations M, calculation;

[0144] S52: calculate the kth iteration of the dictionary

[0145] Fixing the dictionary A, the dictionary B, the dictionary learning problem P3 is transformed into an optimization problem about the dictionary :

[0146]

[0147] Obtain sparse representation by using orthogonal matching pursuit algorithm Specifically:

[0148]

[0149] y temp = vectorize(Y)

[0150]

[0151] Where, vectorize(·) represents vectorization operation, OMP represents orthogonal matching pursuit algorithm, and s represents sparse factor.

[0152] According to get Where reshape(·) represents matrix rearrangement function;

[0153] S53: update to Calculate the kth iteration of the dictionary A (k) ;

[0154] Fixing the sparse representation The dictionary learning problem P3 is transformed into an optimization problem about the dictionary A:

[0155]

[0156] Solve this problem by using gradient projection method, A (k) is:

[0157]

[0158] Here, normalize means that all columns are matrices with unit Euclidean norm.

[0159] S54: A (k) Re-update A and calculate the k-th iteration dictionary B (k) ;

[0160] Fixed sparse representation Dictionary A, dictionary learning problem P3 is transformed into an optimization problem about dictionary B:

[0161]

[0162] The gradient projection method is used to solve this problem. (k) for:

[0163]

[0164] S55: Determine whether k≥M, if yes, execute step S57, otherwise execute step S56;

[0165] S56: B (k) Re-update B, execute k′=k+1, and re-update k′ to k, and return to step S52;

[0166] S57: Get A (M) , B (M) , according to A (M) , B (m) , using the orthogonal matching pursuit algorithm to obtain sparse representation as the mixed field channel h hybrid .

[0167] In order to demonstrate the performance of the method, this embodiment performs algorithm simulation and compares it with other algorithms. Figure 3 As shown in the figure, the channel estimation performance comparison of different dictionaries under various signal-to-noise ratio conditions is shown. It can be seen that as the signal-to-noise ratio increases, the smaller the MMSE of the three dictionary channel estimations, the better the channel estimation performance. In addition, compared with the far-field OMP algorithm and the near-field OMP algorithm, the channel estimation performance of the method proposed in this paper is better. Figure 4 As shown in Figure 1, the channel estimation performance comparison of different dictionaries at various distances between users or scattering clusters is shown. It can be seen that when the distance is 20m-120m, the channel estimation performance of the method proposed in this paper is better. Figure 5 As shown in Figure 2, the channel estimation performance comparison of different dictionaries under different pilot lengths is shown. It can be seen that as the pilot length increases, the smaller the MMSE of the three dictionary channel estimations, the better the channel estimation performance. The channel estimation performance of the method proposed in this paper is better than that of the far-field OMP algorithm and the near-field OMP algorithm.

[0168] Example 3

[0169] The embodiment provides a mixed intelligent reflecting surface auxiliary-based integrated sensing and communication system, which comprises the following as shown in the figure: Figure 6

[0170] A scene establishing module is configured to establish a mixed field channel scene, wherein the scene comprises one base station, at least one near-field scatterer, at least one far-field scatterer and K users;

[0171] A model constructing module is configured to construct a mixed field channel model according to the mixed channel scene;

[0172] A problem solving module is configured to establish a mixed field channel recovery problem and solve it according to the mixed field channel model.

[0173] The same or similar reference signs correspond to the same or similar components;

[0174] The terms describing the positional relationship in the drawings are only used for example illustration, and should not be understood as a limitation to the patent;

[0175] Obviously, the above embodiment of the present application is only an example for clearly illustrating the present application, and is not a limitation to the embodiments of the present application. Any modification, equivalent replacement and improvement made on the basis of the above description for the ordinary skilled in the art should be included in the protection scope of the claims of the present application.​

Claims

1. A method for channel estimation in ultra-large-scale MIMO mixed-field based on dictionary learning, characterized in that: The following steps are involved: Establishing a mixed-field channel scenario, the scenario including a base station, at least one near-field scatterer, at least one far-field scatterer, and K users; Constructing a mixed field channel model according to the mixed field channel scenario; Establishing and solving a mixed field channel restoration problem based on the mixed field channel model; Constructing a hybrid field channel model according to the hybrid channel scenario includes: y=Ph hybrid +n Where y represents the user's received signal, where N RF Indicates the number of base station radio frequency chains, N RF <<N, where N represents the number of antennas equipped by the base station, S represents the number of time periods, (·) T represents transpose, represents the overall observation matrix, p S represents the user's transmission pilot sequence in the Sth time slot, represents the configuration simulation combination matrix of the S-th time slot base station, h hybrid represents the mixed field channel, represents noise, h hybrid The calculation formula is: Among them, l f and l n are the number of far-field and near-field path components, L = l f +L n is the number of all path components, and Respectively represent the first f The gain and angle of the far-field path components, d and λ c Represents antenna spacing and carrier wavelength, s = λ c / 2, is the actual physical angle, Indicates the first f The steering vectors of the far-field path components are: in,(·) H represents the conjugate transpose; Indicates the L n The gain of the path component, and Indicates the first n The angle and distance of the secondary near-field scattering from the center of the antenna array, Indicates the first n The distance from the near-field scatterer to the nth antenna of the base station, Establishing and solving the mixed field channel recovery problem based on the mixed field channel model includes: Establishing the hybrid field channel recovery problem; Convert the mixed field channel recovery problem into a dictionary learning problem P1; Transform the dictionary learning problem P1 into the dictionary learning problem P2; Transform the dictionary learning problem P2 into the dictionary learning problem P3; Solve the dictionary learning problem P3; The problem of establishing a hybrid field channel recovery includes: The mixed field channel h hynrid Convert to sparse representation: in, h hybrid The sparse representation of is the value to be estimated, and D is The corresponding hybrid dictionary; The sparse recovery problem of mixed field channels can be expressed as: in, Represents the L0 norm, which means finding the vector The number of non-zero entries of ; The conversion of the mixed field channel recovery problem into a dictionary learning problem P1 includes: in, The conversion of the dictionary learning problem P1 into the dictionary learning problem P2 includes: according to The separable structural characteristics of in represents the Kronecker product, and then the dictionary learning problem P1 is transformed into: Where Y=[y1,…,y K ], y K represents the received signal of the Kth user, A sparse representation of the mixed field channel representing the Kth user; The conversion of the dictionary learning problem P2 into the dictionary learning problem P3 includes: in, Solving the dictionary learning problem P3 includes: Dictionary learning problem P3 about A, B and Are non-convex, optimize the dictionary separately And dictionaries A and B, the specific steps are as follows: S51: Initialization dictionary k=1, maximum number of iterations M, calculation; S52: Calculate the dictionary for the kth iteration Fixed dictionary A, dictionary B, dictionary learning problem P3 is transformed into about dictionary The optimization problem is: Obtaining sparse representation using orthogonal matching pursuit algorithm according to get where reshape(·) represents the matrix rearrangement function; S53: Re-update Calculate the k-th iteration dictionary A (k) ; Fixed sparse representation Dictionary B, dictionary learning problem P3 is transformed into an optimization problem about dictionary A: The gradient projection method is used to solve this problem. (k) for: Among them, normalize means that all columns are matrices with unit Euclidean norm; S54: A (k) Re-update A and calculate the k-th iteration dictionary B (k) ; Fixed sparse representation Dictionary A, dictionary learning problem P3 is transformed into an optimization problem about dictionary B: The gradient projection method is used to solve this problem. (k) for: S55: Determine whether k≥M, if yes, execute step S57, otherwise execute step S56; S56: B (k) Re-update B, execute k′=k+1, and re-update k′ to k, and return to step S52; S57: Get A (M) , B (M) , according to A (M) , B (M) , using the orthogonal matching pursuit algorithm to obtain sparse representation as the mixed field channel h hybrid ; In step S52, obtaining a sparse representation by using an orthogonal matching pursuit algorithm includes: y temp =vectorize(Y) Where vectorize(·) represents the vectorization operation, OMP represents the orthogonal matching pursuit algorithm, and s represents the sparsity factor.

2. A synaesthesia integrated system based on hybrid intelligent reflective surface assistance, used to implement the method described in claim 1, characterized in that: include: A scenario establishment module, configured to establish a mixed-field channel scenario, the scenario comprising a base station, at least one near-field scatterer, at least one far-field scatterer, and K users; A model building module, configured to build a mixed field channel model according to the mixed channel scenario; The problem solving module is used to establish and solve the mixed field channel restoration problem according to the mixed field channel model.

Images