Adaptive channel estimation method based on parameter perturbation

Abstract

The present invention discloses an adaptive channel estimation method based on parameter perturbation, which mainly solves the problem of low accuracy of channel parameter estimation due to off-grid effect in the prior art in the case of unknown channel sparsity and in the large-scale MIMO system based on millimeter waves. The implementation scheme is as follows: 1) Initialize the step size and select atoms of the step size number in the dictionary matrix; 2) Determine and update the support set elements using the parameter perturbation method; 3) Calculate the channel related parameters according to the mapping relationship between the updated support set elements and the received signal; 4) Calculate the channel matrix according to the channel related parameters to complete the channel estimation. The present invention can effectively suppress the off-grid effect, and by accurately capturing the angle disturbance that may occur in the grid in the angle domain, determine the support set elements with the minimum orthogonal residual, and can achieve optimal sparse recovery estimation in the case of unknown channel sparsity, and can be used for super-resolution channel estimation in the large-scale MIMO system based on millimeter waves.

Description

Technical Field

[0001] The present invention belongs to the field of communication technology, and further relates to an adaptive channel estimation method, which can be used in a millimeter wave large-scale MIMO system with unknown channel sparsity. By accurately capturing the angular disturbance that may occur in a grid in an angle domain, the off-grid effect is effectively suppressed, optimal sparse recovery estimation is achieved, and super-resolution channel parameter estimation is completed. Background Art

[0002] In the B5G and 6G era, a large number of emerging fields have emerged, such as 3D, autonomous driving, and panoramic virtualization. Although these fields have improved the accuracy of applicable data, the grid construction method still has certain defects in parameter resolution due to the limitation of adjacent grid points. In addition, since the grid is discrete and mapped in the angle domain, the success rate of mapping must be considered. This is because the channel environment is changeable, so different angle of arrival-angle of departure (AoA-AoD) groups correspond to different paths, and the grid points are limited. When the number of multipaths is greater than the total number of grid points, there must be at least two paths mapped to the same grid point, which greatly affects the final result of the channel parameters, that is, the off-grid effect.

[0003] Beijing Institute of Technology disclosed in its patent document with application number 201810778028.6 "A wideband channel estimation method based on super-resolution angle and delay estimation". Its implementation steps are as follows: (1) For the channel estimation process of the millimeter wave massive MIMO-OFDM system under the hybrid analog-digital precoding architecture, the training signal of the transceiver is designed to obtain a low-dimensional equivalent channel with array response shift invariance; (2) Using the designed training symbols, the low-dimensional equivalent channel matrix corresponding to all subcarriers at the receiving end is jointly processed to obtain the equivalent channel vector, and the signal parameter estimation using rotational invariance technology (ESPRIT) in the spatial spectrum estimation theory is used to estimate the super-resolution estimation values ​​of the arrival angle, departure angle and multipath delay corresponding to the sparse multipath components in the wideband millimeter wave channel; (3) Using the estimated angle and delay information, combined with the equivalent channel vector and the designed training symbols, the original frequency domain channel matrix corresponding to the subcarrier is reconstructed. Since this method needs to determine the angle quantization bits to design the hybrid precoder and hybrid combiner, and different angle quantization bits correspond to different hybrid precoder and hybrid combiner forms, it cannot adaptively perform super-resolution channel parameter estimation in millimeter-wave large-scale MIMO systems.

[0004] Anjinappa CK et al. proposed a SOMP algorithm based on parameter perturbation in their paper "Off-grid aware channel and covariance estimation in mmwave networks[J].IEEE Transactions on Communications, 2020, 68(6):3908-3921." The implementation steps of this method are as follows: (1) Find the initial grid point, determine the initial grid point, and limit the search space of the perturbation; (2) Find the perturbation through the perturbation solver, starting from the initial grid point, and jointly process the AoA-AoD parameters so that they move in the direction of minimizing the sum of the residual norms in their respective grid areas; (3) Reconstruct the channel matrix. Since the iterative process of this method is improved on the basis of the SOMP algorithm, it is necessary to input the channel sparsity at the beginning of the iteration to determine the number of iterations, which makes it impossible to apply when the channel sparsity is unknown. Summary of the invention

[0005] The purpose of the present invention is to address the deficiencies of the above-mentioned prior art and propose an adaptive channel estimation method based on parameter perturbation to adaptively determine the number of iterations of the channel estimation algorithm and improve the performance of channel parameter estimation under unknown channel sparsity.

[0006] To achieve the above object, the technical solution of the present invention comprises the following steps:

[0007] (1) Initialization:

[0008] Assume the received signal is The support set is S0, and the initial orthogonal residual components are The 2-norm of The termination residual is ε; let the number of iterations k = 1, the step size is S, and the angle deviates from the upper bound φ max ;

[0009] (2)Judgment Is it established:

[0010] If so, find For the S grid points with the highest correlation, number j, execute step (3);

[0011] Otherwise, record the maximum number of iterations L and jump to step (7);

[0012] (3) Determine the support set elements:

[0013] (3a) Select the dictionary matrix column vector element Φ corresponding to the S grid point numbers t Ψ, update the support set S k =S0∪(Φt Ψ) j , determine the received signal In the current support set S k The mapping component on Determine the initial orthogonal residuals of the received signal: in represents the pseudo-inverse of a matrix;

[0014] (3b) From the mapping component Select the S items with the largest absolute value and determine the column vector Λ of the dictionary matrix s , calculate the corrected orthogonal residual of the received signal:

[0015] (3c) Determine whether e′≥e holds:

[0016] If yes, update the step size S=S+1 and return to step (2);

[0017] Otherwise, replace the column vector elements of the S-term dictionary matrix in the support set selected in step (3a) with the S-term with the largest absolute value selected in step (3b) to determine the support set S k The elements in update the mapping component and weight coefficient β k , execute step (4);

[0018] (4) Calculate the angle φ of the element deviation in the support set i , update the support set elements:

[0019] (4a) When the angle deviates from the upper bound φ max Under the constraint of i ;

[0020] (4b) For the support element s determined in the kth iteration i , based on the deviation angle φ of each element i , update the support set elements:

[0021] (5) Calculate channel related parameters:

[0022] (5a) Take the updated support set element s i ′ and receiving signal The real part of the product of is used as the updated weight coefficient β k ′;

[0023] (5b) Calculate the channel arrival angle and departure angle

[0024] (5b1) In the dictionary matrix Φt Find the element s in the support set that matches the updated one i ′ One-to-one corresponding grid point number, get the initial arrival angle and initial departure angle

[0025] (5b2) According to the weight coefficient β k ′ and the updated support set element s i ′, calculate the offset arrival angle and the offset departure angle

[0026] (5b3) Based on the results of (5b1) and (5b2), the corrected arrival angle is obtained and the corrected departure angle

[0027] (6) Enter the next iteration process:

[0028] (6a) Update the orthogonal residual components Update the initial orthogonal residual e;

[0029] (6b) Let the number of iterations k = k + 1, and return to step (2);

[0030] (7) Calculate the channel matrix:

[0031]

[0032] in It refers to the array response of the user end based on the corrected arrival angle, It refers to the array response at the base station based on the corrected departure angle.

[0033] Compared with the prior art, the present invention has the following advantages:

[0034] First, by removing the limitations of the angle quantization bits in designing hybrid precoders and hybrid combiners, compared with traditional orthogonal matching pursuit (OMP) and other methods, the present invention does not require the sparsity of the known channel to determine the number of iterations. The iterative process can be terminated when the orthogonal residual reaches the preset accuracy. It is suitable for super-resolution channel parameter estimation in millimeter-wave large-scale MIMO systems with unknown channel sparsity, and has a wider range of application scenarios.

[0035] Second, the present invention converts the sparse recovery problem of the received signal into the problem of minimizing orthogonal residual components, accurately captures the angular perturbations that may occur in the grid in the angle domain, effectively suppresses the off-grid effect, and achieves optimal sparse recovery estimation in the case of unknown channel sparsity, thereby improving the performance of super-resolution channel parameter estimation in millimeter-wave large-scale MIMO systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flow chart of the implementation of the present invention;

[0037] Figure 2 and Figure 3 It is a simulation result diagram of the present invention. Specific implementation methods

[0038] The embodiments and effects of the present invention are further described in detail below with reference to the accompanying drawings.

[0039] The adaptive channel estimation method based on parameter perturbation in this example is to adaptively determine the number of iterations of channel estimation by judging whether the orthogonal residual reaches the preset accuracy when the channel sparsity is unknown. In each iteration process, the dictionary matrix is ​​updated by calculating the deviation angle of the elements in the support set. Finally, the relevant channel parameters are estimated and the corresponding channel matrix is ​​output, thereby realizing super-resolution channel parameter estimation in millimeter-wave massive MIMO systems.

[0040] Reference Figure 1 , the implementation steps of this example are as follows:

[0041] Step 1, initialization.

[0042] When the received signal, dictionary matrix and step size are known, the received signal is assumed to be The support set is S0, and the initial orthogonal residual components are The 2-norm of The termination residual is ε; let the dictionary matrix column vector element be Φ t Ψ; Assume that the number of iterations k = 1, the maximum number of iterations is L, the step size is S, and the angle deviates from the upper bound φ max .

[0043] Step 2: Determine whether the orthogonal residual reaches the preset accuracy.

[0044] Will The 2-norm of Compare with the termination residual ε:

[0045] if The orthogonal residual does not reach the preset accuracy, and it is necessary to find The numbers of the S grid points with the highest correlation Execute step 3;

[0046] Otherwise, the orthogonal residual reaches the preset accuracy, the maximum number of iterations L=k is recorded, and the process goes to step 7.

[0047] Step 3: Determine the support set elements.

[0048] 3.1) Select the dictionary matrix column vector elements (Φ) corresponding to the S grid point numbers t Ψ) j , update the support set S k =S0∪(Φ t Ψ) j , determine the received signal In the current support set S k The mapping component on in Represents the pseudo-inverse of the matrix and determines the initial orthogonal residual of the received signal:

[0049] 3.2) From the calculated mapping components Select the S items with the largest absolute value from the dictionary matrix, and record the S column vector combinations in the dictionary matrix as Λ s , calculate the corrected orthogonal residual of the received signal:

[0050] 3.3) Compare the modified orthogonal residual e′ with the initial orthogonal residual e:

[0051] If e′≥e, update the step size S=S+1 and return to step 2;

[0052] Otherwise, replace the column vector elements of the S-term dictionary matrix in the support set selected in the first step with the S-term with the largest absolute value selected in the second step to determine the support set S k The elements in update the mapping component and weight coefficient Go to step 4.

[0053] Step 4: Calculate the angle φ of the element deviation in the support set i , update the support set elements.

[0054] 4.1) When the angle deviates from the upper bound φ max Under the constraint of i :

[0055] 4.1.1) Formula for the angle of deviation of the support concentrated elements: Carry out series expansion, that is in,

[0056] 4.1.2) Filter the parameter items of the above series expansion, calculate the parameter item set when the initial orthogonal residual e is the smallest, and obtain the perturbation angle φ of the element deviation in the support set i ;

[0057] 4.2) For the support element s determined in the kth iteration i, based on the deviation angle φ of each element i , update the support set elements, and get the updated support set elements s′ i :

[0058]

[0059] Step 5: Calculate relevant channel parameters.

[0060] 5.1) Take the updated support element s′ i and receive signals The real part of the product of is used as the updated weight coefficient

[0061] 5.2) Calculate the channel arrival angle and departure angle

[0062] 5.2.1) In the dictionary matrix Φ t Find the element s′ in the support set that matches the updated one in Ψ i One-to-one corresponding grid point numbers to obtain the initial arrival angle and initial departure angle

[0063] 5.2.2) Using β′ k and s′ i The product formula Get the array response from the user and the conjugate transposed array response at the base station

[0064] 5.2.3) According to the calculation results of 5.2.2), the array response of the user end is and the array response at the base station Do vector expansion:

[0065]

[0066]

[0067] Where M and N are the number of antennas configured at the base station and user end, respectively, and T represents transposition;

[0068] 5.2.4) According to the above vector expansion, the product of the corresponding terms of the vector expansion of the user-side array response and the base station-side array response is added to β′ k s′ to obtain the offset arrival angle and the offset departure angle

[0069] 5.2.5) Based on the calculation results of 5.2.1) and 5.2.4), obtain the corrected arrival angle and the corrected departure angle

[0070]

[0071]

[0072] Step 6: Enter the next iteration process.

[0073] 6.1) According to the support element S obtained in step 4 k Update the orthogonal residual components

[0074] 6.2) Set the number of iterations k = k + 1 and return to step 2.

[0075] Step 7: Calculate the channel matrix H.

[0076] When the orthogonal residual component reaches the preset accuracy, the iteration process is terminated and the channel matrix is ​​calculated:

[0077]

[0078]

[0079]

[0080] in, It refers to the array response of the user end based on the corrected arrival angle, and N is the number of antennas configured at the user end. It refers to the array response of the base station based on the corrected departure angle, and M is the number of antennas configured at the base station.

[0081] The effect of the present invention can be further illustrated by the following simulation experiment:

[0082] 1. Simulation experiment conditions:

[0083] The hardware platform of the simulation experiment of the present invention has a processor of Intel i5-6400 CPU with a main frequency of 2.7 GHz and a memory of 16 GB.

[0084] The software platforms of the simulation experiment of the present invention are: Windows 10 operating system and MATLAB R2019a.

[0085] The simulation experiment parameters of the present invention are shown in Table 1:

[0086] Table 1 Experimental parameter settings list

[0087] Settings Value Number of base station antennas 16,32 Number of user antennas 8,16 Number of multipaths 16:2:30 Number of grid points 16 Signal-to-noise ratio (dB) -10:5:30 Pilot length 400:800:1600 Stop error <![CDATA[10 -3 ]]> Step Length 1

[0088] 2. Simulation content and result analysis:

[0089] Simulation 1, based on the multipath channel with the parameter settings in Table 1, the method of the present invention and the existing least squares LS algorithm and orthogonal matching pursuit OMP algorithm are used to estimate the channel parameters when the pilot sequence length is 400 and 800, respectively, when the signal-to-noise ratio is between -10dB and 30dB, and the normalized mean square error between the estimated channel and the actual channel is calculated, and the normalized mean square error comparison curves estimated by the three methods are obtained, as shown in FIG. Figure 2 shown.

[0090] Simulation 2, based on the multipath channel with the parameter settings in Table 1, the method of the present invention and the existing least squares LS algorithm and orthogonal matching pursuit OMP algorithm are used to estimate the channel parameters when the number of base station antennas is 32 and the number of user antennas is 16, and when the number of base station antennas is 16 and the number of user antennas is 8, respectively, with the signal-to-noise ratio between -10dB and 30dB, and the normalized mean square error NMSE between the estimated channel and the actual channel is calculated, and the normalized mean square error comparison curves estimated by the three methods are obtained, as shown in FIG. Figure 3 shown.

[0091] The prior art least squares LS algorithm refers to the method proposed by French scientist Legendre in 1806 to find the best function matching of data by minimizing the sum of squares of errors, referred to as the least squares LS algorithm.

[0092] The prior art orthogonal matching pursuit OMP algorithm refers to the signal recovery method proposed by Pati.YC et al. in "Orthogonal matching pursuit: Recursive function approximation with applications to wavelet decomposition [C] / / Proceedings of 27th Asilomar conference on signals, systems and computers. IEEE, 1993: 40-44", referred to as the orthogonal matching pursuit OMP algorithm.

[0093] The normalized mean square error NMSE between the estimated channel and the actual channel is calculated as follows:

[0094]

[0095] in represents the estimated channel, ||H|| FIt represents the frobenius norm of the matrix H, which is the sum of the squares of the absolute values ​​of the elements of the matrix H.

[0096] Depend on Figure 2 It can be seen that when the pilot length is 400 and 800 respectively, the normalized mean square error between the channel estimated by the present invention and the actual channel when the signal-to-noise ratio is between -10dB and 30dB is significantly lower than the normalized mean square error between the channel estimated by the least squares LS algorithm and the orthogonal matching pursuit OMP algorithm when the signal-to-noise ratio is between -10dB and 30dB, indicating that the present invention has a better effect on channel parameter estimation.

[0097] Depend on Figure 3 It can be seen that the normalized mean square error between the channel estimated by the present invention and the actual channel when the signal-to-noise ratio is between -10dB and 30dB is significantly lower than the normalized mean square error between the channel estimated by the least squares LS algorithm and the orthogonal matching pursuit OMP algorithm when the signal-to-noise ratio is between -10dB and 30dB, respectively, indicating that the present invention has a better effect on channel parameter estimation.

[0098] Depend on Figure 2 It can be seen that in the scenario of pilot sequences of the same length, as the signal-to-noise ratio increases, the performance of the three algorithms gradually improves, and the performance of the method of the present invention is significantly better than the least squares LS algorithm and the orthogonal matching pursuit OMP algorithm. When the pilot sequence length changes from 400 to 800, the performance of the orthogonal matching pursuit OMP algorithm remains almost unchanged, and the performance of the least squares LS algorithm increases slightly, but is still weaker than the performance of the method of the present invention when the pilot length is 400 and 800. When the pilot length increases, the performance of the method of the present invention also increases.

[0099] Depend on Figure 3 It can be seen that when the antenna configurations at the base station and user end are the same, the normalized mean square error of the method of the present invention is significantly lower than that of the least squares LS algorithm and the orthogonal matching pursuit OMP algorithm after the signal-to-noise ratio is greater than 5dB. When the antenna configurations at the base station and antenna ends change, the performance of the method of the present invention increases significantly when the number of antennas configured at the base station and user ends increases, and it is still better than the least squares LS algorithm and the orthogonal matching pursuit OMP algorithm.

[0100] The above simulation experiments show that: the present invention uses an adaptive channel estimation method based on parameter perturbation to effectively suppress the off-grid effect by accurately capturing the angle disturbance that may occur in the grid in the angle domain, and solves the adverse effect of the off-grid effect on channel parameter estimation in the millimeter-wave large-scale MIMO system under the condition of unknown channel sparsity, realizes the optimal sparse recovery estimation, and completes the super-resolution channel parameter estimation. Compared with the traditional sparse recovery algorithm, the present invention does not require the sparsity of the known channel, has a wider range of applicable situations, and has a better effect on channel parameter estimation.

Claims

1. An adaptive channel estimation method based on parameter perturbation, characterized in that: The steps include: (1) Initialization: Assume the received signal is The support set is S0, and the initial orthogonal residual components are The 2-norm of The termination residual is ε; let the number of iterations k = 1, the step size is S, and the angle deviates from the upper bound φ max ; (2)Judgment Is it established: If so, find For the S grid points with the highest correlation, number j, execute step (3); Otherwise, record the maximum number of iterations L and jump to step (7); (3) Determine the support set elements: (3a) Select the dictionary matrix column vector element Φ corresponding to the S grid point numbers t Ψ, update the support set S k =S0∪(Φ t Ψ) j , determine the received signal In the current support set S k The mapping component on Determine the initial orthogonal residuals of the received signal: in represents the pseudo-inverse of a matrix; (3b) From the mapping component Select the S items with the largest absolute value and determine the column vector Λ of the dictionary matrix s , calculate the corrected orthogonal residual of the received signal: (3c) Determine whether e′≥e holds: If yes, update the step size S=S+1 and return to step (2); Otherwise, replace the column vector elements of the S-term dictionary matrix in the support set selected in step (3a) with the S-term with the largest absolute value selected in step (3b) to determine the support set S k The elements in update the mapping component and weight coefficient β k , execute step (4); (4) Calculate the angle φ of the element deviation in the support set i , update the support set elements: (4a) When the angle deviates from the upper bound φ max Under the constraint of i ; (4b) For the support element s determined in the kth iteration i , based on the deviation angle φ of each element i , update the support set elements: (5) Calculate channel related parameters: (5a) Take the updated support set element s i ′ and receiving signal The real part of the product of is used as the updated weight coefficient β k ′; (5b) Calculate the channel arrival angle and departure angle (5b1) In the dictionary matrix Φ t Find the element s in the support set that matches the updated one i ′ One-to-one corresponding grid point number, get the initial arrival angle and initial departure angle (5b2) According to the weight coefficient β k ′ and the updated support set element s i ′, calculate the offset arrival angle and the offset departure angle (5b3) Based on the results of (5b1) and (5b2), the corrected arrival angle is obtained and the corrected departure angle (6) Enter the next iteration process: (6a) Update the orthogonal residual components Update the initial orthogonal residual e; (6b) Let the number of iterations k = k + 1, and return to step (2); (7) Calculate the channel matrix: in It refers to the array response of the user end based on the corrected arrival angle, It refers to the array response at the base station based on the corrected departure angle.

2. The method according to claim 1, characterized in that In step (4a), the parameter perturbation method is used to calculate the angle φ of each element in the support set. i , is based on the orthogonal residual components, mapping components and termination residuals, using the angle formula for solving the deviation of the elements in the support set: It is implemented as follows: (4a1) The inverse tangent relation in the angle formula is expanded as follows: in, (4a2) Filter the parameter terms of the series expansion of step (4a1), calculate the parameter term set when the initial orthogonal residual e is the smallest, and obtain the perturbation angle φ of the element deviation in the support set i .

3. The method according to claim 1, characterized in that Step (5a) obtains the updated weight coefficient β k ′, which is expressed as follows: 。 4. The method according to claim 1, characterized in that: In step (5b2), according to the weight coefficient β k ′ and the updated support set element s i ′, calculate the offset arrival angle and the offset departure angle The implementation is as follows: (5b21) Using β k ′ and s i The product formula of ′ Calculate the array response at the user end and the conjugate transposed array response at the base station (5b22) According to the result of (5b21), the array response of the user end is and the array response at the base station Do vector expansion: Where M and N are the number of antennas configured at the base station and user end, respectively, and T represents transposition; (5b23) According to the vector expansion in (5b22), the product of the corresponding terms of the vector expansion of the user-side array response and the base station-side array response is added to β′ k s′ to obtain the offset arrival angle and the offset departure angle 5. The method according to claim 1, characterized in that The array response of the user terminal involved in the formula in step (7) based on the corrected arrival angle is It is expressed as follows: Wherein, N is the number of antennas configured at the user end, and T represents transposition.

6. The method according to claim 1, characterized in that The array response of the base station based on the corrected departure angle involved in the formula of step (7) is It is expressed as follows: Wherein, M is the number of antennas configured at the base station, and T represents transposition.

Images