Channel estimation method and system for millimeter wave multiple-input multiple-output system

Abstract

The invention provides a channel estimation method and system for a millimeter wave multiple-input-multiple-output system, and the method comprises the steps: decomposing a channel measurement matrix determined based on a reference signal through employing a singular value decomposition algorithm, and obtaining an initial first factor matrix and an initial second factor matrix; inputting an initial set comprising the initial first estimation matrix, the initial first factor matrix and the initial second factor matrix into a pre-trained iterative neural network to perform multiple rounds of updating operation so as to obtain a channel estimation matrix; wherein each round of updating operation comprises the following steps: obtaining a first factor matrix and a second factor matrix output by the current round by utilizing a projection gradient descent algorithm and a back projection convolution layer; and obtaining a first estimation matrix output by the current round based on the first estimation matrix input by the current round and the first factor matrix and the second factor matrix output by the current round. According to the invention, the accuracy and efficiency of channel estimation can be improved.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and in particular to a channel estimation method and system for a millimeter wave multi-input multi-output system. Background Art

[0002] In mmWave massive Multiple Input Multiple Output (MIMO) systems, efficient beamforming depends on accurate channel state information (CSI). However, as the number of antennas increases, the pilot signal (i.e., reference signal) and computational overhead of traditional channel estimation algorithms also increase, affecting the efficiency of MIMO systems. Although compressed sensing-based channel estimation algorithms can utilize the sparse characteristics of mmWave channels to reduce training costs, they often ignore the low-rank structure of the channel. In order to make full use of the low-rank and sparse characteristics of the channel, recent studies have regarded channel estimation as a low-rank matrix completion problem. However, existing channel estimation methods based on low-rank matrix completion still face challenges in terms of computational complexity and estimation accuracy.

[0003] Therefore, there is an urgent need for a channel estimation method suitable for millimeter-wave multiple-input multiple-output systems, which can complete the estimated channel matrix through a low-rank matrix, while reducing the computational complexity and improving the accuracy of channel estimation. Summary of the invention

[0004] In view of this, an embodiment of the present invention provides a channel estimation method and system for a millimeter wave multi-input multi-output system, which can reduce computational complexity and improve channel estimation accuracy.

[0005] One aspect of the present invention provides a channel estimation method for a millimeter wave multiple-input multiple-output system, the method comprising the following steps: Determine a channel measurement matrix based on a reference signal and a set channel sensing position, and decompose the channel measurement matrix using a singular value decomposition algorithm to obtain an initial first factor matrix and an initial second factor matrix; Inputting an initial input set including a channel measurement matrix used as an initial first estimation matrix, an initial first factor matrix, and an initial second factor matrix into a pre-trained iterative neural network for multiple rounds of update operations, and finally obtaining an estimation matrix of the channel; wherein the iterative neural network uses the output of the current round of update operations as the input of the next round of update operations; Each update operation includes: Using the projected gradient descent algorithm, the first intermediate projection matrix and the second intermediate projection matrix of the current round are obtained based on the first factor matrix, the second factor matrix and the first estimation matrix inputted in the current round, and the first intermediate projection matrix and the second intermediate projection matrix of the current round are respectively converted into the first factor matrix and the second factor matrix outputted in the current round through the back-projection convolution layer; and the first estimation matrix outputted in the current round is obtained based on the first estimation matrix inputted in the current round, the first factor matrix outputted in the current round and the second factor matrix outputted in the current round.

[0006] In some embodiments of the present invention, the initial input set further includes a channel measurement matrix used as an initial second estimation matrix; and In the case where the initial input set includes an initial second estimation matrix, each round of updating operation further includes: inputting the first estimation matrix outputted in the current round into the denoising convolution layer, and outputting the second estimation matrix outputted in the current round.

[0007] In some embodiments of the present invention, obtaining the first estimation matrix outputted by the current round based on the first estimation matrix inputted by the current round, the first factor matrix outputted by the current round, and the second factor matrix outputted by the current round includes: In the case where the first estimation matrix outputted from the final round is used as the estimation matrix of the channel, the first estimation matrix outputted from the current round is obtained based on the first estimation matrix inputted from the current round, the first factor matrix and the second factor matrix outputted from the current round, the channel measurement matrix and the set first step length; When the second estimation matrix output from the final round is used as the estimation matrix of the channel, the first estimation matrix output from the current round is obtained based on the first estimation matrix input from the current round, the first factor matrix and the second factor matrix output from the current round, the channel measurement matrix, the set first step length and the second estimation matrix input from the current round.

[0008] In some embodiments of the present invention, a first intermediate projection matrix and a second intermediate projection matrix of the current round are obtained based on a first factor matrix, a second factor matrix and a first estimation matrix input in the current round by using a projected gradient descent algorithm, including: for the first factor matrix, based on the first factor matrix, the second factor matrix, the first estimation matrix and a set second step size input in the current round, a gradient is calculated to obtain a gradient descent value; the gradient descent value is input into a projected convolution layer, and the first intermediate projection matrix is ​​obtained as an output; For the second factor matrix, the gradient is calculated based on the first factor matrix output by the current round, the second factor matrix and the first estimation matrix input by the current round, and the third set step size to obtain the gradient descent value, or the gradient is calculated based on the first factor matrix, the second factor matrix and the first estimation matrix input by the current round, and the third set step size to obtain the gradient descent value; the gradient descent value is input into the projection convolution layer, and the second intermediate projection matrix is ​​output.

[0009] In some embodiments of the present invention, for the first factor matrix, the calculation process of the gradient descent value includes the following steps: calculating the product of the first factor matrix input in the current round and the conjugate transpose of the second factor matrix; calculating the difference between the product and the first estimation matrix input in the current round, multiplying the obtained difference, the set second step size and the second factor matrix input in the current round to obtain the gradient; subtracting the gradient from the first factor matrix input in the current round to obtain the gradient descent value; For the second factor matrix, the calculation process of the gradient descent value includes the following steps: calculating the product of the first factor matrix output of the current round and the conjugate transpose of the second factor matrix input of the current round, or calculating the product of the first factor matrix input of the current round and the conjugate transpose of the second factor matrix input of the current round; calculating the difference between the product and the first estimation matrix input of the current round, multiplying the obtained difference, the set third step size and the first factor matrix output of the current round to obtain the gradient, and subtracting the gradient from the second factor matrix input of the current round to obtain the gradient descent value.

[0010] In some embodiments of the present invention, after calculating the gradient, the method further includes: using the product of the scaling operator and the calculated gradient as the optimized gradient to calculate the gradient descent value; wherein, for the first factor matrix, the scaling operator is the inverse matrix of the product of the second factor matrix input in the current round and its conjugate transpose; for the second factor matrix, the scaling operator is the inverse matrix of the product of the first factor matrix output in the current round and its conjugate transpose, or the scaling operator is the inverse matrix of the product of the first factor matrix input in the current round and its conjugate transpose.

[0011] In some embodiments of the present invention, before obtaining the first factor matrix and the second factor matrix of the current round output through the back-projection convolution layer, the update operation also includes: performing a soft threshold operation on each element in the first intermediate projection matrix and the second intermediate projection matrix obtained by using the projection gradient descent algorithm to obtain an updated first intermediate projection matrix and a second intermediate projection matrix.

[0012] In some embodiments of the present invention, when the first estimation matrix outputted in the final round is used as the estimation matrix of the channel, the calculation formula of the first estimation matrix outputted in the current round is as follows: When the second estimation matrix outputted in the final round is used as the estimation matrix of the channel, the calculation formula of the first estimation matrix outputted in the current round is as follows: Among them, Z k+1 Represents the first estimation matrix of the current round output, Z k represents the first estimation matrix of the current round input, Indicated by Z k The partial matrix composed of the element values ​​of the channel sensing position set in , η Z Indicates the first step length, Y Ω represents the channel measurement matrix, M k Represents the second estimation matrix of the current round input, L k+1 and R k+1 They represent the first factor matrix and the second factor matrix of the current round output respectively, and β1 and β2 both represent weights.

[0013] In some embodiments of the present invention, the loss function of the iterative neural network in the training phase is the mean square error between the true channel matrix and the channel estimation matrix output by the iterative neural network; The projection convolution layer and the back-projection convolution layer are symmetrical convolution layers separated by the ReLU activation function.

[0014] Another aspect of the present invention provides a channel estimation system for a millimeter wave multi-input multi-output system, including a processor and a memory, wherein the memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the system implements the steps of the method described in any of the above embodiments.

[0015] The channel estimation method and system for the millimeter wave multi-input multi-output system proposed in the present invention can realize low-rank matrix factorization (Low-Rank Matrix Factorization) through the projected gradient descent method and the back-projection convolution layer, so as to realize the iteration of the first factor matrix and the second factor matrix in each round of update operation, and use the partial matrix of the current round input and output to iterate the first estimation matrix, thereby obtaining the channel estimation matrix. Compared with the existing scheme, the channel estimation method based on low-rank matrix completion proposed in this application makes full use of the low-rank and sparse characteristics of the channel, which can effectively reduce the computational complexity in the estimation process and enhance the estimation accuracy.

[0016] Additional advantages, purposes, and features of the present invention will be described in part in the following description, and will become apparent to those skilled in the art after studying the following, or may be learned from the practice of the present invention. The purposes and other advantages of the present invention may be achieved and obtained by the structures specifically indicated in the specification and the accompanying drawings.

[0017] Those skilled in the art will appreciate that the objectives and advantages that can be achieved with the present invention are not limited to the above specific description, and the above and other objectives that can be achieved by the present invention will be more clearly understood from the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present invention, constitute a part of the present application, and do not constitute a limitation of the present invention. In the drawings: Figure 1 FIG. 4 is a flow chart of a channel estimation method for a millimeter wave MIMO system according to an embodiment of the present invention.

[0019] Figure 2 Schematic diagram of the design of an iterative neural network in one embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments and the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0021] It should also be noted that, in order to avoid obscuring the present invention due to unnecessary details, only structures and / or processing steps closely related to the solutions according to the present invention are shown in the accompanying drawings, while other details that are not closely related to the present invention are omitted.

[0022] It should be emphasized that the term “include / comprises” when used herein refers to the presence of features, elements, steps or components, but does not exclude the presence or addition of one or more other features, elements, steps or components.

[0023] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the accompanying drawings, the same reference numerals represent the same or similar components, or the same or similar steps.

[0024] Existing channel estimation methods still have challenges in terms of computational complexity and estimation accuracy. In view of this, the present application proposes to use an iterative neural network that combines a low-rank factorization network (Low-Rank Factorization) and a deep unfolding network (DeepUnfolding) to realize channel estimation of millimeter wave large-scale MIMO systems. The channel estimation method proposed in this application regards the channel estimation problem as a low-rank matrix completion problem to make full use of the sparsity and low-rank structure of the millimeter wave channel. Specifically, the present application can decompose the channel measurement matrix determined based on the reference signal into two low-rank factor matrices, and optimize them by the projected gradient descent method; and the convolution layer is used as a nonlinear sparse transformation module in the iterative neural network to realize the projection and back projection of the matrix to improve the channel estimation accuracy. Furthermore, the iterative neural network of the present application may also include a denoising module to improve the channel estimation accuracy by reducing the influence of Gaussian noise during signal transmission. The channel estimation method for the millimeter wave MIMO system proposed in this application not only reduces the computational complexity, but also enhances the accuracy of the estimation, and can be applied to a variety of wireless communication scenarios.

[0025] Since this application mainly relies on iterative neural networks to achieve channel estimation, this application uses the following description to illustrate the design ideas of iterative neural networks. Specifically: Assume that the downlink channel matrix to be estimated in this application is After transmitting T training symbols If the receiver remains stationary during the period, then at each time t (i.e., when the t-th training symbol is transmitted), the combined signal y(t) received by the receiver on one side of the channel to be estimated can be expressed as: y(t)=w(t) H M real q(t)s(t)+w(t) H n(t); (1-1) Where q(t) and w(t) represent the beamforming codebooks from the predefined and combined codebook The randomly selected beamforming vector and combining vector, w(t) H represents the conjugate transpose of w(t), and n(t) represents a zero mean and a variance of σ 2 Additive White Gaussian Noise (AWGN) is an additive white Gaussian noise. In addition, to simplify the calculation process, s(t)=1 can usually be set.

[0026] Assumptions (express There are N T elements), (express There are N R elements), and the beamforming matrix and the combination matrix To use the collection and The full rank matrix constructed by all vectors in . At this time, if formula (1-1) is expressed in the form of a matrix, when the training length of the reference signal mentioned above is T, the signal γ received by the receiver can be defined as: in, The matrix representing additive white Gaussian noise, W H represents the conjugate transpose of the combination matrix W, W H M real Q represents a low-rank matrix obtained by combining the beamforming matrix, the combination matrix and the channel true matrix.

[0027] Furthermore, since the number of input channel signals T is usually smaller than the channel real matrix M real The number of elements in N R ×NT The signal transmitted in the channel may not pass through all the positions in the channel, and considering that the signal may be attenuated, phase shifted or noisy during transmission, the sampling operator can be used. To define the partial observation value matrix (i.e., channel measurement matrix) Y obtained based on the reference signal transmitted in the channel to be estimated Ω . Sampling operator It can be defined as: Wherein, Ω represents the set of positions of the to-be-estimated channels through which the reference signal passes, that is, the set channel sensing positions (Ω is an invariant set), and Ω contains T elements (denoted as |Ω|=T). i，j represents the element value at position (i, j) in the matrix Y (determined based on the reference signal of the input and output of the channel to be estimated). That is, the element value at each position in Ω is a known value determined based on the reference signal.

[0028] By introducing the sampling operator, the channel estimation problem can be transformed into a low-rank matrix completion problem. Moreover, considering that the subsequent matrix completion involves using the channel measurement matrix Y Ω The signal estimation matrix Y is obtained to distinguish the channel measurement matrix and the Steps for sampling the matrix Y (since the channel sensing position Ω is pre-set, the result of sampling the matrix Y is also Y Ω ), hereinafter referred to as Y Ω Represents the channel measurement matrix obtained according to the signal, Indicates sampling of the completed matrix Y.

[0029] After the sampling operator is introduced into formula (1-2), it can be expressed as At the same time, in order to facilitate the realization of low-rank matrix completion, the following simplifications can be made: Assume that Q and W can be unit matrices, that is, and at this time

[0030] By formula It can be seen that the channel estimation process based on low-rank matrix completion can be filled with the channel measurement matrix Y Ω The signal estimation matrix Y is obtained, and then the channel estimation matrix M is constructed by removing the influence of additive white Gaussian noise. recon , so that the solved M recon Considered as M real Therefore, in the low-rank matrix completion process, the channel can be reconstructed by solving the following constrained optimization problem (which can be regarded as the loss function of the channel estimation process): Where r is the channel true matrix Mreal The estimated value of the rank of the matrix can be determined by channel tracking technology or based on existing channel statistics in practical applications. The present invention does not specifically limit the method for determining the estimated rank r. Since the F-norm is often used to evaluate the error, stability and convergence of the matrix, this application uses the F-norm to evaluate the convergence of the low-rank matrix completion. express The square of the F-norm.

[0031] As an example, to meet the above operation conditions, the estimated rank r mentioned in this application is less than N R 、N T and T, and N is not specifically limited R 、N T and T. In addition, Y Ω , Y, M recon and M real The rank and size of the matrix (N R ×N T ) remain consistent.

[0032] The difficulty in solving formula (1-3) lies in reconstructing the channel matrix M recon , and low-rank decomposition is a technique that decomposes a high-dimensional matrix into multiple low-rank matrices. It is often used to reduce the number of parameters and calculations in low-rank matrix completion, thereby reducing the complexity of the model. Among them, the singular value decomposition (SVD) algorithm is a commonly used matrix completion method that can represent any complex matrix by multiplying three smaller and simpler sub-matrices. Therefore, according to the channel true matrix M of the channel to be estimated real and the channel estimation matrix M recon With low-rank characteristics, M recon (or M real ) is decomposed into a left singular vector matrix The transpose of a singular value diagonal matrix ∑ and a right singular vector matrix The formula is: M recon =U∑V H (or ). Since ∑ is a diagonal matrix, it can be simplified to M recon =LR H , where L = U∑ 1 / 2 ,R=V∑ 1 / 2 ,and and On the contrary, the channel estimation channel M can be constructed using the low-rank factor matrices L and R recon .

[0033] However, in the process of applying the singular value decomposition algorithm, in order to evaluate the feasibility of low-rank matrix completion, the following constraints need to be defined: If formula (1-4) is met, it is considered that the matrices U and V satisfy the μ-incoherent condition.

[0034] Among them, ||U|| 2,∞ represents the maximum l2-norm of each row of the matrix U, ||U|| F represents the F-norm of the matrix U, and μ represents the measurement parameter of the irrelevant condition.

[0035] Usually, the Projected Gradient Descent (PGD) algorithm can be used in the matrix completion process to ensure that the matrices U and V satisfy the μ-incoherence condition, so L and R can be searched on the following constraint set: Among them, the projection operator and The channel sparsity in the angular domain can be used to design L ＊ and R ＊ Denote the real channel matrix M real The first factor matrix and the second factor matrix obtained by SVD, L ＊ =U ＊ ∑ 1 / 2 , R ＊ =V ＊ ∑ 1 / 2 , and Respectively indicate that there are a L and a R A sparse matrix with non-zero elements, express The maximum l1-norm of each row, express The maximum l1-norm of each row.

[0036] Furthermore, considering the constraints on L and R in formula (1-5), the constrained optimization problem can be restated as follows based on formula (1-3) and formula (1-5): Solving formula (1-6) can obtain the low-rank factor matrices L and R that meet the above constrained optimization problem, thereby obtaining the channel estimation matrix M recon To facilitate the solution, formula (1-6) can be divided into the projected gradient descent process and projection recovery (back-projection) process

[0037] ① Use the projected gradient descent algorithm to solve and obtain the projection matrix corresponding to the optimization matrix L′ and R′. Using the known L and R that do not necessarily satisfy the small μ incoherence condition, the projected gradient descent algorithm can be used to obtain the optimized L′ and R′ that satisfy the small μ incoherence condition (gradient descent is used to optimize L and R, and projection is used to ensure that the R of L is searched on the constraint set during the optimization process). The formula for solving using projected gradient descent can be expressed as: Among them, A′ is L′ in the convex set On the projection, B′ is the projection of R′ on the convex set The projection on and are all intermediate variables in the gradient descent process, η L and η R are the gradient step sizes corresponding to matrices L and R respectively.

[0038] As an example, in order to accelerate the convergence process of the intermediate variables A′ and B′, the scaling operator in the scaled gradient descent can also be introduced in the gradient descent process. H R) -1 and (L H L) -1 are the scaling operators corresponding to matrices L and R respectively, so the process of solving using the projected gradient descent method can be expressed by the following formula: ② After obtaining A′ and R′ on the convex set using the projected gradient descent method, the back-projection operation can be used to convert A′ into L′ in the original space, and convert B′ into R′ in the original space. That is, the back-projection operation realizes the conversion of A′ and B′ from the convex set to the original space into L′ and R′, which can be expressed by the following formula: Furthermore, before performing the back-projection operation, A′ and B′ can be optimized by a soft threshold operation, that is, the soft threshold operation is applied element by element to each element value A′ in the matrices A′ and B′. i，j and B′ i，j Among them, A′ i，j represents the element value at position (i, j) in matrix A′, B′ i，j represents the element value at position (i, j) in matrix B′, i∈{1, 2, ..., N R}, j∈{1, 2, ..., N T}. Therefore, the conversion formula of L and R′ can be expressed as: Among them, θL and θ R L′ and R′ represent the corresponding soft thresholds, soft(A′, θ L )=∑soft(A′ i，j ,θ L ), and soft(B′,θ R )=∑soft(B′ i，j ,θ R ).

[0039] as well as

[0040] That is, the above formula (1-6) can use low-rank matrix factorization (LRMF) to implement the channel estimation algorithm. However, in practical applications, it is challenging to obtain the sparsity of the real factor matrix, which limits the potential application of LRMF; moreover, using the discrete Fourier transform (DFT) matrix for sparse transformation cannot obtain the exact transformation matrix of the channel, which may lead to energy leakage, especially when the actual channel parameters are not on a discrete grid.

[0041] Furthermore, since formula (1-6) includes the multiplication operation of the factor matrices L and R to be solved, the amount of calculation in the solution operation process is relatively large. Therefore, in order to reduce the amount of calculation and the complexity of calculation during optimization, the present application introduces an auxiliary matrix Z, which is used to replace the product of the conjugate transpose of L and R in formula (1-6), that is, adding the constraint condition Z = LR H And, since the channel estimation matrix M is obtained recon Its low-rank property can be used to decompose L and R, and further add the constraint Z = M recon If the influence of additive white Gaussian noise is ignored, the solved Z can be regarded as the channel estimation matrix; if the Gaussian noise contained in Z is removed, the channel estimation matrix can be obtained.

[0042] Without considering the influence of noise N, formula (1-6) can be expressed as: Considering the influence of noise N, formula (1-6) can be expressed as: in, represents the constraint item that ensures the input signal is consistent with the received signal, Indicates M k The prior constraints of Representing constraints and Zk =M k The penalty item, represents the channel prior sparse constraint term.

[0043] Inspired by the concept of deep expansion, the iterative update method can be used to solve formula (1-6). In addition, in order to optimize the iterative performance, this application adopts a learnable nonlinear transformation function and Instead of using D R and D L The projection process is realized and the learnable threshold θ is used * (θ L and θ R ) implements soft threshold iteration to optimize A′ and B′. Therefore, formula (1-7) can be expressed as: Formula (1-8) can be expressed as: in, Indicates the projection optimization operation on L. Indicates the projection optimization operation on R. Indicates that a sparse transformation operation is performed on M through a nonlinear transformation function

[0044] As an example, and It can be a learnable nonlinear transformation function implemented by a convolutional neural network (CNN), which can adaptively learn the optimal nonlinear sparse transformation from data. This application does not specifically limit the parameters of the convolutional neural network (for example, the number of convolution kernels, the size of the convolution kernel, and the number of layers, etc.), as long as the nonlinear sparse transformation can be implemented.

[0045] From the above reasoning process, it can be seen that using the neural network structure designed in this application, the iteration process of each parameter is as follows: ①L k+1 and R k+1 Iterative updates According to formula (1-9) and formula (1-10), after removing the terms in the loss function (constrained optimization problem) that are irrelevant to the update of the low-rank factor matrices L and R, the process of solving L and R through iterative updates can be obtained by solving the following sub-problem of formula (2-1): Where k represents the number of iterations, k = 0, 1, 2, ..., k max , L k and R kRepresents the factor matrix of the output of the kth round of iteration. Formula (2-1) can be used to express the L and R that are minimized in each round of iterative update process.

[0046] As an example, the two formulas in formula (2-1) are solved for L simultaneously. k+1 and R k+1 , if we first update L k+1 Update to get R k+1 The subproblems of solving L and R by iterative updating can be expressed as follows: Using L k or L k+1 The solved R k+1 is relatively close, so this application can solve L at the same time k+1 and R k+1 , or update L in a certain order k+1 and R k+1 .

[0047] Furthermore, when performing iterative neural network design, the projected gradient descent method and back-projection operation can be used to solve L k+1 and R k+1 This application uses nonlinear transformation Function representation L k The corresponding gradient descent value is projected to obtain A k+1 (or Function representation R k The corresponding gradient descent value is projected to obtain B k+1 ), and nonlinear transformations Function represents A k+1 Perform back projection to obtain L k+1 (or Function represents B k+1 Perform back projection to get R k+1 ), that is, using Functions and Functions replace D R and D L Perform projection optimization operations using Functions and Function instead of projection operator and Perform a back-projection operation.

[0048] Taking the solution order of formula (2-1) as an example, after the introduction of the deep unfolding network, the projection gradient descent and back-projection operations are also expressed as an iterative process. The iterative process of the k-th projection gradient descent algorithm can be expressed by the following formula (2-2): Among them, A k+1 YesL k+1 In convex set The projection on B k+1 YesR k+1 In convex set The projection on and are all intermediate variables in the gradient update process. L and η R L k and R k The gradient update step size.

[0049] As an example, you can also design and L k and R k , then according to formula (2-2), the iterative process of the k-th projection gradient descent algorithm can be expressed by the following formula (2-3): Using the projected gradient descent method to obtain A on the convex set k+1 and B k+1 Afterwards, the back-projection operation and soft threshold iteration algorithm can be used to transform A k+1 L converted to the original space k+1 , and B k+1 Convert to original space R k+1 , expressed using the following formula (2-4): As an example, consider D R and D L and The transposition relationship between Figure 2 As shown, the network structure is symmetrically designed Functions and function.

[0050] ② Although the constraint Z=LR in this application H , but the solution process may be affected by residual errors, so this application considers solving Z in the iterative update process based on the above formula (1-7) or (1-8) k+1 .

[0051] Without considering the noise, Z k+1 The update formula of can be obtained by solving the following sub-problems: When noise is taken into account, Z k+1The update formula of is obtained by solving the following sub-problems: From the above formula, we can see that when updating L k+1 and R k+1 Afterwards, Z can be further updated k+1 . For the above formula Applying the first-order Taylor expansion, assuming it satisfies l1-Lipschitz continuity, then After sorting, we can get Z k+1 The update formula is as follows: or Among them, C1 is a constant, β1 is the updated L k+1 and R k+1 The weight of β2 is given by Z k The noise reduction of M k The weight of . ξ and γ are the coefficients of the penalty term, η Z Z k The corresponding gradient descent step size, l1 is a constant.

[0052] (3) According to formula (1-9) and formula (1-10), after removing irrelevant terms, the iterative update process of M can be obtained by solving the following sub-problems: The solution process is similar to the LASSO problem and can be solved by convolutional neural network. k+1 The update formula can be expressed as: M k+1 =D Θ (Z k+1 ); (2-10) That is, by changing Z k+1 Input D Θ The convolutional neural network constructed by the function can output the denoised estimation matrix M k+1 .

[0053] This application proposes a channel estimation method for a millimeter wave MIMO system. The channel measurement matrix Y obtained based on the reference signal can be obtained by using the designed iterative neural network. Ω Get the channel estimation matrix Z kmax or Mk max .like Figure 1 As shown, the method proposed in this application may specifically include steps S110 to S120.

[0054] Step S110: Determine a channel measurement matrix based on a reference signal and a set channel sensing position, and decompose the channel measurement matrix using a singular value decomposition algorithm to obtain an initial first factor matrix and an initial second factor matrix.

[0055] The real channel matrix of the channel to be estimated is a blank matrix with a known matrix size. Since the channel sensing position set in the channel can sense the input signal and transmit the signal to the receiving end, the values ​​of some elements in the channel matrix can be determined by sending and receiving signals, and the channel measurement matrix Y is obtained. Ω Specifically, the signal transmitter on one side of the channel to be estimated can input the reference signal into the channel to be estimated, and the signal receiver on the other side of the channel to be estimated can receive the corresponding signal, and the reference signal can be transmitted through the channel sensing position Ω set in the channel to be estimated. Based on the transmitted reference signal and the received reference signal, the element value of the channel sensing position set in the channel matrix of the channel to be estimated can be determined to obtain the channel measurement matrix Y Ω (That is, the channel measurement matrix is ​​a channel partial estimation matrix determined based on the reference signal.) In the present application, the method for determining the channel measurement matrix may adopt the existing technology, and the present invention does not specifically limit it.

[0056] Furthermore, although the channel matrix itself has low rank and sparse characteristics, the channel measurement matrix Y Ω is the matrix of some observations, which may not have low rank characteristics, so it can be obtained from Y Ω The low-rank matrix of rank r is cut out from the matrix to perform singular value decomposition, that is, Y Ω Approximate decomposition into a low-rank factor matrix Therefore, the SVD algorithm can be used to decompose Y Ω The initial first factor matrix L0 and the initial second factor matrix R0 (or the initial first factor matrix R0 and the initial second factor matrix L0) are obtained. For ease of description, the first factor matrix is ​​taken as L k , the second factor matrix is ​​R k Take this as an example for description.

[0057] Step S120: Input the initial input set including the initial first estimation matrix Z0, the initial first factor matrix L0 and the initial second factor matrix R0 into the pre-trained iterative neural network for multiple rounds of update operations to obtain the channel estimation matrix M recon .

[0058] As an example, if Gaussian noise in the signal transmission process is considered, the second estimation matrix M output in the final round is kmax As the channel estimation matrix M recon, then the initial input set includes the initial first estimation matrix Z0, the initial first factor matrix L0, the initial second factor matrix R0 and the initial second estimation matrix M0; if the Gaussian noise in the signal transmission process is not considered, the first estimation matrix Z of the final round output kmax As the channel estimation matrix M recon , then the initial input set includes the initial first estimation matrix Z0, the initial first factor matrix L0 and the initial second factor matrix R0. In addition, in order to make full use of the information carried in the transmission signal, the present application uses the channel measurement matrix Y Ω As initial values ​​of the first estimation matrix and the second estimation matrix, that is, the channel measurement matrix is ​​used as the initial first estimation matrix Z0 and the initial second estimation matrix M0.

[0059] like Figure 2 As shown, the initial input set is input into the pre-trained iterative neural network for multiple rounds of update operations to obtain the channel estimation matrix. Moreover, the iterative neural network uses the output of the current round of update operations as the input of the next round of update operations, that is, the input parameter matrix (L, R and Z) and output parameter (also matrix L, R and Z) of each round of update operations are the same.

[0060] In some embodiments of the invention, each round of update operation in the iterative neural network includes: Step S210: Using the projected gradient descent algorithm, based on the first factor matrix L input in the current round k , the second factor matrix R k and the first estimation matrix Z k Get the first intermediate projection matrix A of the current round k+1 and the second intermediate projection matrix B k+1 , and the first intermediate projection matrix A of the current round is transformed through the back-projection convolution layer k+1 and the second intermediate projection matrix B k+1 Convert to the first factor matrix L of the current round output respectively k+1 and the second factor matrix R k+1 .

[0061] More specifically, step S210 can be used according to the first factor matrix L input in the current round k , the second factor matrix R k and the first estimation matrix Z k , update the first factor matrix L of the current round output k+1 and the second factor matrix R k+1 However, the projected gradient update steps for the first factor matrix and the second factor matrix can be the same or different. This is because, as can be seen from the above formula (2-1), the first factor matrix L k and the second factor matrix R kThe update order of is not necessarily synchronized. The first factor matrix can be updated first, and then the second factor matrix can be updated using the updated first factor matrix (or, the second factor matrix can be updated first and then the first factor matrix can be updated); the first factor matrix L can also be updated synchronously. k and the second factor matrix R k The present application does not specifically limit the order of updating the first factor matrix and the second factor matrix. k Get L k+1 , then based on L k+1 Update the second factor matrix R k Get R k+1 Take this as an example for description.

[0062] In some embodiments of the present invention, the updating process for the first factor matrix is ​​as follows: using the projected gradient descent algorithm, based on the first factor matrix L input in the current round k , the second factor matrix R k and the first estimation matrix Z k Get the first intermediate projection matrix A of the current round k+1 , and the first intermediate projection matrix A of the current round is transformed through the back-projection convolution layer k+1 Convert to the first factor matrix L of the current round output k+1 The update process for the second factor matrix is ​​as follows: Based on the first factor matrix L output in the current round k+1 , the second factor matrix R of the current round input k and the first estimation matrix Z k Get the second intermediate projection matrix B of the current round k+1 , and the second intermediate projection matrix B of the current round is transformed through the back-projection convolution layer k+1 Convert to the second factor matrix R of the current round output k+1 .

[0063] For the first factor matrix, the first intermediate projection matrix A is obtained using the projected gradient descent algorithm k+1 The steps include: based on the first factor matrix L input in the current round k , the second factor matrix R k , the first estimation matrix Z k and the second step length η L Calculate the gradient to get the gradient descent value; input the calculated gradient descent value into the projection convolution layer, and output the first intermediate projection matrix A k+1 For the second factor matrix, the second intermediate projection matrix B is obtained using the projected gradient descent algorithm. k+1 The steps include: based on the first factor matrix L output by the current round k+1 , the second factor matrix R of the current round input kand the first estimation matrix Z k , and the third setting step length η R Calculate the gradient to get the gradient descent value (if L is updated synchronously k+1 and R k+1 , then for the second factor matrix, the second intermediate projection matrix B is obtained using the projected gradient descent algorithm k+1 The steps include: based on the first factor matrix L input in the current round k , the second factor matrix R of the current round input k and the first estimation matrix Z k , and the third setting step length η R Calculate the gradient to get the gradient descent value); input the obtained gradient descent value into the projection convolution layer, and output the second intermediate projection matrix B k+1 .

[0064] In some embodiments of the present invention, as shown in formula (2-2), for the first factor matrix, the calculation process of the gradient descent value includes the following steps: Calculate the first factor matrix L of the current round input k With the second factor matrix R k The conjugate transpose of ; calculate the product Subtract the difference Z of the first estimated matrix of the current round input k , the difference The second step length η is set L And the second factor matrix R of the current round input k Multiply to get the gradient; input the first factor matrix L of the current round k Subtract the gradient to get the gradient descent value. For the second factor matrix, the calculation process of the gradient descent value includes the following steps: Calculate the first factor matrix L output in the current round k+1 The second factor matrix R of the current round input k The conjugate transpose of The product of (if L is updated synchronously k+1 and R k+1 , then for the second factor matrix, calculate the first factor matrix L of the current round input k The second factor matrix R of the current round input k The conjugate transpose of ), calculate the product Subtract the first estimated matrix Z of the current round input k The difference between The third step length η is set R And the first factor matrix L output in the current round k+1 Multiply to get the gradient, and input the second factor matrix R of the current round kSubtract the gradient to get the gradient descent value.

[0065] As an example, as shown in formula (2-3), after the gradient is calculated using formula (2-2), the method further includes: using the product of the scaling operator and the calculated gradient as the optimized gradient to calculate the gradient descent value, thereby accelerating convergence; wherein, for the first factor matrix, the scaling operator is the second factor matrix R input in the current round k and its conjugate transpose The inverse matrix of the product of; for the second factor matrix, the scaling operator is the first factor matrix L output in the current round k+1 and its conjugate transpose The inverse matrix of the product of (if L is updated synchronously k+1 and R k+1 , then for the second factor matrix, the scaling operator is the first factor matrix L output in the current round k and its conjugate transpose The inverse matrix of the product of ).

[0066] Further, as shown in formula (2-4), the first factor matrix L of the current round output is obtained by the back-projection convolution layer k+1 and the second factor matrix R k+1 Previously, each update operation also included: using the projection gradient descent algorithm to obtain the first intermediate projection matrix A k+1 and the second intermediate projection matrix B k+1 Perform soft threshold operation on each element in to obtain the updated first intermediate projection matrix soft(A k+1 ,θ L ) and the second intermediate projection matrix soft(B k+1 ,θ R ). That is, the first intermediate projection matrix A is optimized by using the soft threshold iteration algorithm in the iteration process. k+1 and the second intermediate projection matrix B k+1 .

[0067] As an example, the projection convolution layer in this application can be used as a projection operator in the projection gradient descent algorithm, and the first intermediate projection matrix A is obtained according to the matrix input in the current round through the projection optimization operation. k+1 and the second intermediate projection matrix B k+1 , and because A k+1 and B k+1 are the first factor matrix L output in the current round respectively k+1 and the second factor matrix R k+1 The projection on the convex set can be performed by the back-projection convolution layer to perform the back-projection operation and convert the first intermediate projection matrix A k+1 and the second intermediate projection matrix B k+1Mapped to the original space, we get the first factor matrix L of the current round output k+1 and the second factor matrix R k+1 , so the projection convolution layer and the back-projection convolution layer are symmetrical convolution layers separated by the ReLU activation function.

[0068] In some embodiments of the present invention, in step S210, L is updated k+1 and R k+1 Afterwards, step S220 may be performed: based on the first estimation matrix Z input in the current round k , the first factor matrix L output in the current round k+1 And the second factor matrix R output of the current round k+1 , get the first estimated matrix Z of the current round output k+1 Specifically, the first estimated matrix Z output by the final round is kmax As the channel estimation matrix M recon In the case of the first estimation matrix Z based on the current round input k , the first factor matrix L output in the current round k+1 and the second factor matrix R k+1 , channel measurement matrix Y Ω and the first step length η Z , get the first estimated matrix Z of the current round output k+1 . In the second estimation matrix Mk outputted by the final round max As the channel estimation matrix M recon In the case of the first estimation matrix Z based on the current round input k , the first factor matrix L output in the current round k+1 and the second factor matrix R k+1 , channel measurement matrix Y Ω , set the first step length η Z and the second estimated matrix M of the current round input k , get the first estimated matrix Z of the current round output k+1 .

[0069] More specifically, the first estimated matrix Z output by the final round kmax As the channel estimation matrix M recon In the case of the first estimation matrix Z output in the current round k+1 The calculation formula is as follows: The second estimated matrix Mk output by the final round max As the channel estimation matrix M recon In the case of the first estimation matrix Z output in the current round k+1 The calculation formula is as follows: Among them, Z k+1 Represents the first estimation matrix of the current round output, Z k represents the first estimation matrix of the current round input, Indicated by Z k The partial matrix composed of the element values ​​of the channel sensing position set in , η Z Indicates the first step length, Y Ω represents the channel measurement matrix, M k Represents the second estimation matrix of the current round input, L k+1 and R k+1 They represent the first factor matrix and the second factor matrix of the current round output respectively, and β1 represents L k+1 and R k+1 The corresponding weight, β2 is M k The weight of .

[0070] In some embodiments of the present invention, when the initial input set includes an initial second estimation matrix, each round of updating operation further includes step S230: replacing the first estimation matrix Z output by the current round k+1 Input the denoising convolution layer and output the second estimated matrix M of the current round output k+1 , so that the second estimation matrix output by the final round is used as the estimation matrix of the channel.

[0071] As an example, since the input of the projection convolution layer, the back-projection convolution layer, and the denoising convolution layer is a two-dimensional matrix, and the output of the convolution layer is also a two-dimensional matrix, the dimension needs to be transformed before the input convolution layer and after the output convolution layer. For example, the matrix dimension is N R ×rL k and dimension is r×N T R k It can be reshaped into 1×N through gradient update R ×r and 1×N T In addition, the projection convolution layer, the back-projection convolution layer and the denoising convolution layer mentioned in the present application may all be composed of a multi-layer CNN structure. This is because different layers of the convolutional neural network can remove noise interference of different types and levels, so the hierarchical feature learning mechanism of CNN helps to recover the signal more accurately.

[0072] According to the above steps S210 to S230, the iterative neural network can be divided into three parts: an LRMF module, an estimation matrix update module, and an estimation matrix denoising module. In the LRMF module, a high-dimensional feature representation of the input data can be obtained through the projection convolution layer and the back-projection convolution layer, and the sparsity of the matrix can be constrained by a soft threshold operation; in the estimation matrix denoising module, a denoising convolution layer is used to remove noise interference of different types and levels to achieve accurate signal recovery; in the estimation matrix update module, based on the matrix output by the LRMF module, the weight β1 of the LRMF module, and the weight β2 of the estimation matrix denoising module, it can be updated to obtain a matrix that meets the constraint condition Z=LR H Z k+1 . In each update operation in the iterative neural network, the learnable parameters to be used include the gradient update step size (the first step size η Z , the second step length η L and the third step length η R ), soft threshold parameter θ L and θ R , the update parameters β1 and β2 of the first estimation matrix, and the neural network weights (Θ M , Θ L and θ R D Θ , and as well as and weight).

[0073] As an example, multiple rounds of update operations can be performed with the aid of a multi-layer neural network in an iterative neural network. Each layer of the neural network can perform one round of update operations, and the values ​​of the learnable parameters of each layer of the neural network can be different. In this case, all learnable parameters of the iterative neural network can be expressed as

[0074] In some embodiments of the present invention, during the training phase of the iterative neural network, each batch (E=1, 2, ..., E max ) are all randomly generated. Moreover, the loss function of the iterative neural network in the training phase is the channel true matrix The channel estimation matrix output by the iterative neural network The mean square error between (used to evaluate computational complexity and accuracy) can be expressed as follows: Where f(*) represents an iterative neural network, Represents the loss function.

[0075] The channel estimation method for the millimeter wave multi-MIMO system proposed in this application has the following advantages: ① Millimeter wave channels have low rank and sparse characteristics, and channel estimation can be regarded as low rank matrix completion, thereby improving the accuracy and efficiency of channel estimation.

[0076] ② The iterative neural network designed using convolutional neural networks can adaptively learn the optimal nonlinear sparse transformation, avoiding the inaccuracy caused by discrete Fourier transform matrices, etc.

[0077] ③ Through the operations of projection and back-projection, the factor matrix is ​​searched on the constraint set during the channel estimation optimization process, thereby improving the estimation accuracy.

[0078] ④The denoising module can effectively remove noise interference of different types and levels, helping to restore the signal more accurately.

[0079] Corresponding to the above method, the present invention also provides a channel estimation system for a millimeter wave multi-input multi-output system, the system comprising a computer device, the computer device comprising a processor and a memory, the memory storing a computer program / instructions, the processor being used to execute the computer program / instructions stored in the memory, and when the computer program / instructions are executed by the processor, the system implements the steps of the method described above.

[0080] It should be understood by those skilled in the art that the exemplary components, systems and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software or a combination of the two. Whether it is performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present invention are programs or code segments used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link via a data signal carried in a carrier.

[0081] It should be clear that the present invention is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present invention.

[0082] In the present invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with features of other embodiments or replace features of other embodiments.

[0083] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the embodiments of the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A channel estimation method for a millimeter wave multiple-input multiple-output system, characterized in that: The method comprises the following steps: Determine a channel measurement matrix based on a reference signal and a set channel sensing position, and decompose the channel measurement matrix using a singular value decomposition algorithm to obtain an initial first factor matrix and an initial second factor matrix; Inputting an initial input set including a channel measurement matrix used as an initial first estimation matrix, the initial first factor matrix and the initial second factor matrix into a pre-trained iterative neural network for multiple rounds of update operations, and finally obtaining an estimation matrix of the channel; wherein the iterative neural network uses the output of the current round of update operations as the input of the next round of update operations; Each update operation includes: Using a projected gradient descent algorithm, a first intermediate projection matrix and a second intermediate projection matrix of the current round are obtained based on the first factor matrix, the second factor matrix and the first estimation matrix input in the current round, and the first intermediate projection matrix and the second intermediate projection matrix of the current round are respectively converted into the first factor matrix and the second factor matrix output in the current round through a back-projection convolution layer; and Based on the first estimation matrix input in the current round, the first factor matrix output in the current round and the second factor matrix output in the current round, a first estimation matrix output in the current round is obtained.

2. The method according to claim 1, characterized in that The initial input set also includes a channel measurement matrix used as an initial second estimation matrix; as well as In the case where the initial input set includes an initial second estimation matrix, each round of updating operation further includes: inputting the first estimation matrix outputted in the current round into the denoising convolution layer, and outputting the second estimation matrix outputted in the current round.

3. The method according to claim 2, characterized in that The step of obtaining the first estimation matrix outputted in the current round based on the first estimation matrix inputted in the current round, the first factor matrix outputted in the current round, and the second factor matrix outputted in the current round comprises: In the case where the first estimation matrix outputted from the final round is used as the estimation matrix of the channel, the first estimation matrix outputted from the current round is obtained based on the first estimation matrix inputted from the current round, the first factor matrix and the second factor matrix outputted from the current round, the channel measurement matrix and the set first step length; When the second estimation matrix output from the final round is used as the estimation matrix of the channel, the first estimation matrix output from the current round is obtained based on the first estimation matrix input from the current round, the first factor matrix and the second factor matrix output from the current round, the channel measurement matrix, the set first step length and the second estimation matrix input from the current round.

4. The method according to claim 1, characterized in that: The method of using the projected gradient descent algorithm to obtain the first intermediate projection matrix and the second intermediate projection matrix of the current round based on the first factor matrix, the second factor matrix and the first estimation matrix input in the current round includes: For the first factor matrix, the gradient is calculated based on the first factor matrix, the second factor matrix, the first estimation matrix and the set second step size input in the current round, so as to obtain a gradient descent value; the gradient descent value is input into the projection convolution layer, and the first intermediate projection matrix is ​​obtained as an output; For the second factor matrix, the gradient is calculated based on the first factor matrix output by the current round, the second factor matrix and the first estimation matrix input by the current round, and the third set step size to obtain a gradient descent value, or the gradient is calculated based on the first factor matrix, the second factor matrix and the first estimation matrix input by the current round, and the third set step size to obtain a gradient descent value; the gradient descent value is input into the projection convolution layer, and the second intermediate projection matrix is ​​output.

5. The method according to claim 4, characterized in that For the first factor matrix, the calculation process of the gradient descent value includes the following steps: calculating the product of the first factor matrix input in the current round and the conjugate transpose of the second factor matrix; calculating the difference between the product and the first estimation matrix input in the current round, multiplying the obtained difference, the set second step size and the second factor matrix input in the current round to obtain the gradient; subtracting the gradient from the first factor matrix input in the current round to obtain the gradient descent value; For the second factor matrix, the calculation process of the gradient descent value includes the following steps: calculating the product of the first factor matrix output of the current round and the conjugate transpose of the second factor matrix input of the current round, or calculating the product of the first factor matrix input of the current round and the conjugate transpose of the second factor matrix input of the current round; calculating the difference between the product and the first estimation matrix input of the current round, multiplying the obtained difference, the set third step size and the first factor matrix output of the current round to obtain a gradient, and subtracting the gradient from the second factor matrix input of the current round to obtain a gradient descent value.

6. The method according to claim 4, characterized in that After the gradient is calculated, the method further includes: using the product of the scaling operator and the calculated gradient as the optimized gradient to calculate the gradient descent value; wherein, for the first factor matrix, the scaling operator is the inverse matrix of the product of the second factor matrix input in the current round and its conjugate transpose; for the second factor matrix, the scaling operator is the inverse matrix of the product of the first factor matrix output in the current round and its conjugate transpose, or the scaling operator is the inverse matrix of the product of the first factor matrix input in the current round and its conjugate transpose.

7. The method according to claim 1, characterized in that Before obtaining the first factor matrix and the second factor matrix of the current round output through the back-projection convolution layer, the updating operation also includes: performing a soft threshold operation on each element in the first intermediate projection matrix and the second intermediate projection matrix obtained by the projection gradient descent algorithm to obtain an updated first intermediate projection matrix and a second intermediate projection matrix.

8. The method according to claim 3, characterized in that In the case where the first estimation matrix outputted in the final round is used as the estimation matrix of the channel, the calculation formula of the first estimation matrix outputted in the current round is as follows: In the case where the second estimation matrix outputted in the final round is used as the estimation matrix of the channel, the calculation formula of the first estimation matrix outputted in the current round is as follows: Among them, Z k+1 Represents the first estimation matrix of the current round output, Z k represents the first estimation matrix of the current round input, Indicated by Z k The partial matrix composed of the element values ​​of the channel sensing position set in , η Z Indicates the first step length, Y Ω represents the channel measurement matrix, M k Represents the second estimation matrix of the current round input, L k+1 and R k+1 They represent the first factor matrix and the second factor matrix of the current round output respectively, and β1 and β2 both represent weights.

9. The method according to claim 4, characterized in that The loss function of the iterative neural network in the training phase is the mean square error between the real channel matrix and the channel estimation matrix output by the iterative neural network; The projection convolution layer and the back-projection convolution layer are convolution layers with symmetrical structures separated by ReLU activation functions.

10. A channel estimation system for a millimeter wave multiple-input multiple-output system, comprising a processor and a memory, characterized in that: The memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the system implements the steps of the method as claimed in any one of claims 1 to 9.

Images