Two-stage super-resolution parameter channel estimation method and device based on RIS

Abstract

The application discloses a two-stage super-resolution parameter channel estimation method and device based on RIS, which cascades the channel between a base station and RIS and the channel between RIS and user equipment to obtain a cascaded channel; an expression of the cascaded channel is deformed into the product of an angle of arrival matrix, a cascaded channel path gain and a cascaded angle of departure matrix conjugate transpose, so as to convert a channel estimation problem into a sparse matrix recovery problem; based on a received signal matrix, the physical position invariability of a reconfigurable intelligent surface and a base station and the same angle of arrival of different user equipment are utilized to estimate the number of angles of arrival and paths; based on the number of angles of arrival and paths, an angle of arrival matrix and a grid are generated; the sparse matrix recovery problem is converted into an optimization problem to estimate the cascaded angle of departure matrix and the cascaded channel path gain of different users; and based on the angle of arrival matrix, the cascaded angle of departure matrix and the cascaded channel path gain, the estimated cascaded channel matrix is calculated.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and in particular to a two-stage super-resolution parameter channel estimation method and apparatus for reconfigurable intelligent surface (RIS) communication systems. Background Technology

[0002] The statements in this section merely refer to the background art related to this invention and do not necessarily constitute prior art.

[0003] To further reduce power consumption and cost while improving communication capacity, researchers are constantly seeking suitable methods for future 6G research. RIS (Radio Reduction System) has received widespread attention from academia and industry as a potential 6G technology. In 5G, millimeter wave (mmWave) is one of the main methods to significantly improve communication capacity and solve the problem of spectrum shortage; however, its short wavelength leads to poor coverage. Deploying RIS can increase signal coverage and compensate for the shortcomings of mmWave. However, to achieve effective and reliable wireless communication in a RIS-mmWave system, accurate Channel State Information (CSI) is required using channel estimation techniques, which is a significant challenge.

[0004] In communication systems, the least squares (LS) method is commonly used to complete channel estimation. However, this method has poor resistance to noise and does not utilize the characteristics of RIS. Therefore, the literature [1] (Jensen T, Carvalho E. An optimal channel estimation scheme for intelligent reflecting surfaces based on a minimum variance unbiased estimator[C]. Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing, Barcelona, ​​Spain, 2020: 5000–5004.) proposed the minimum variance unbiased (MVU) LS estimation method and designed a RIS phase shift matrix based on discrete Fourier transform, which improved the accuracy of channel estimation, but had a large training overhead. Reference [2] (Hu C, Dai L, Han S, et al. Two-timescale channel estimation for reconfigurable intelligent surface aided wireless communications[J].IEEE Transactions on Communications, 2021, 69(11): 7736–7747.) proposes a dual-timescale method, which utilizes the quasi-static characteristics of the base station (BS) and RIS to perform channel estimation on a large timescale. Performing small-scale channel estimation between the user equipment (UE) and RIS can significantly reduce pilot overhead, but the accuracy is slightly worse than the method proposed in reference [1], and the multi-step process can easily cause error propagation problems. mmWave channels are sparsity, and compressed sensing (CS) algorithms are commonly used to recover the channel. This characteristic can be used to reduce training overhead. Commonly used CS algorithms include the orthogonal matching pursuit (OMP) algorithm and the simultaneous orthogonal matching pursuit (SOMP) algorithm.Reference [3] (Chen J, Liang Y, Cheng H, et al. Chancellestimation for reconfigurable intelligent surface aided multi-user mmWaveMIMO systems[J].IEEE Transactions on Wireless Communications, 2023, 22(10): 6853–6869.) proposes a subspace multi-user joint channel estimation method (SMJCE) by utilizing the row and column sparsity unique to RIS channels. This method further reduces training overhead compared to OMP and SOMP. However, using a fixed grid will reduce the accuracy of channel estimation.

[0005] The above methods have improved the accuracy of channel estimation and reduced training overhead to varying degrees. However, how to achieve high-accuracy channel estimation with low training overhead while maintaining complexity remains one of the problems that need to be solved. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a two-stage super-resolution parameter channel estimation method and apparatus based on RIS; it improves channel estimation accuracy, reduces training overhead, and ensures computational complexity by utilizing the sparsity of mmWave channels.

[0007] On the one hand, a two-stage super-resolution parameter channel estimation method based on RIS is provided, including:

[0008] Construct a millimeter-wave multiple input multiple output (MIMO) system assisted by a reconfigurable smart surface (RIS); construct a channel model for the millimeter-wave MIMO system; the user equipment sends orthogonal pilot signals through the reconfigurable smart surface (RIS) to the base station, obtains the received signals from the base station, eliminates the pilot sequence of the received signals, and obtains the received signal matrix;

[0009] The channel between the base station and the reconfigurable smart surface (RIS) is concatenated with the channel between the RIS and the user equipment to obtain the concatenated channel. The expression of the concatenated channel is transformed into the product of the angle of arrival matrix, the path gain of the concatenated channel, and the conjugate transpose of the concatenated departure angle matrix. At the same time, the channel estimation problem is transformed into a sparse matrix recovery problem.

[0010] Based on the received signal matrix, the physical location invariance of the reconfigurable smart surface and the base station, as well as the fact that different user equipment have the same angle of arrival, is used to estimate the angle of arrival and the number of paths; based on the angle of arrival and the number of paths, the angle of arrival matrix and grid are generated.

[0011] The sparse matrix recovery problem is transformed into an optimization problem, and the cascaded departure angle matrix and cascaded channel path gain for different users are estimated using the grid.

[0012] Based on the angle of arrival matrix, the concatenated departure angle matrix, and the concatenated channel path gain, the estimated concatenated channel matrix is ​​calculated; finally, based on the estimated concatenated channel matrix, channel state information is obtained.

[0013] On the other hand, a two-stage super-resolution parameter channel estimation system based on RIS is provided, including:

[0014] The module is configured to: construct a millimeter-wave MIMO system assisted by a reconfigurable smart surface RIS; construct a channel model for the millimeter-wave MIMO system; send orthogonal pilot signals from the user equipment to the base station via the reconfigurable smart surface RIS; obtain the received signals from the base station; eliminate the pilot sequence of the received signals; and obtain the received signal matrix.

[0015] The cascading module is configured to: cascade the channel between the base station and the reconfigurable smart surface RIS with the channel between the reconfigurable smart surface RIS and the user equipment to obtain a cascaded channel; transform the expression of the cascaded channel into the product of the angle of arrival matrix, the path gain of the cascaded channel, and the conjugate transpose of the cascaded departure angle matrix, and at the same time transform the channel estimation problem into a sparse matrix recovery problem;

[0016] The first estimation module is configured to: estimate the angle of arrival and the number of paths based on the received signal matrix, utilizing the invariance of the physical location of the reconfigurable smart surface and the base station, and the fact that different user equipment have the same angle of arrival; and generate the angle of arrival matrix and grid based on the angle of arrival and the number of paths.

[0017] The second estimation module is configured to: transform the sparse matrix recovery problem into an optimization problem, and estimate the cascaded departure angle matrix and cascaded channel path gain for different users using the grid.

[0018] The output module is configured to: calculate the estimated concatenated channel matrix based on the angle of arrival matrix, the concatenated departure matrix, and the concatenated channel path gain; and finally, obtain channel state information based on the estimated concatenated channel matrix.

[0019] The above technical solution has the following advantages or beneficial effects:

[0020] This method guarantees good channel estimation accuracy and low training overhead without significantly increasing complexity, greatly improving channel estimation performance. This invention employs a two-stage channel estimation method. In angle estimation, a fixed grid is not used; instead, an optimization method is employed to find the accurate angle, improving channel estimation accuracy. In the first stage, an error judgment method is used to prevent error propagation. The high-precision grid generated from the angle of arrival estimated in the first stage is used as input for the second stage. Even if there is a small error in the first stage, the second stage can reduce the overall channel error and improve channel estimation accuracy. Attached Figure Description

[0021] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0022] Figure 1 This is a flowchart of the method in Example 1;

[0023] Figure 2 This is the RIS-mmWave communication system of Example 1;

[0024] Figure 3 Angle and path number estimation for Example 1;

[0025] Figure 4 The NMSE curve for Example 1 is shown as a function of training overhead.

[0026] Figure 5 The NMSE curve with respect to SNR in Example 1 (M=64);

[0027] Figure 6 The NMSE curve with respect to SNR in Example 1 (M=16);

[0028] Figure 7 The runtime curve for Example 1 is shown as a function of rounds. Detailed Implementation

[0029] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0030] Reconfigurable Smart Surfaces (RIS) refer to a programmable electromagnetic surface structure developed from metamaterials technology. It consists of a large number of electromagnetic units arranged and combined. By controlling these units, electromagnetic waves can be manipulated in a programmable manner, achieving changes in amplitude, phase, polarization, and frequency, thereby reconfiguring the wireless communication environment. It offers advantages such as low power consumption, low cost, programmability, and ease of deployment.

[0031] Example 1

[0032] This embodiment provides a two-stage super-resolution parameter channel estimation method based on RIS, including:

[0033] S101: Construct a millimeter-wave MIMO system assisted by a reconfigurable smart surface RIS; construct a channel model for the millimeter-wave MIMO system; the user equipment sends orthogonal pilot signals through the reconfigurable smart surface RIS to the base station, obtains the received signal from the base station, eliminates the pilot sequence of the received signal, and obtains the received signal matrix;

[0034] S102: Concatenate the channel between the base station and the reconfigurable smart surface RIS with the channel between the reconfigurable smart surface RIS and the user equipment to obtain the concatenated channel; transform the expression of the concatenated channel into the product of the angle of arrival matrix, the path gain of the concatenated channel, and the conjugate transpose of the concatenated departure angle matrix, and at the same time transform the channel estimation problem into a sparse matrix recovery problem.

[0035] S103: Based on the received signal matrix, taking advantage of the physical position invariance of the reconfigurable smart surface and the base station, and assuming that different user equipment have the same angle of arrival, estimate the angle of arrival and the number of paths; based on the angle of arrival and the number of paths, generate the angle of arrival matrix and grid.

[0036] S104: Transform the sparse matrix recovery problem into an optimization problem, and use the grid to estimate the cascaded departure angle matrix and cascaded channel path gain for different users;

[0037] S105: Based on the angle of arrival matrix, the concatenated departure angle matrix, and the concatenated channel path gain, calculate the estimated concatenated channel matrix; finally, based on the estimated concatenated channel matrix, obtain the channel state information.

[0038] The channel estimation method provided by this invention mainly includes two stages: the arrival angle matrix and BS-RIS path number estimation stage, and the cascaded departure angle matrix and cascaded channel path gain estimation stage.

[0039] Furthermore, the S101: RIS-assisted millimeter-wave MIMO system includes: a plurality of user equipment, a fixed-location reconfigurable smart surface RIS, and a fixed-location base station;

[0040] The system includes two channels: one between the base station and the reconfigurable smart surface RIS, and the other between the reconfigurable smart surface RIS and the user equipment.

[0041] The base station is equipped with multiple antennas; each user equipment is equipped with a single antenna; and the user equipment and the base station cannot directly transmit signals due to obstacles.

[0042] For example, an uplink RIS-mmWaveMIMO system model is established using the Saleh-Valenzuela channel model. The user end sends orthogonal pilot signals through the reconfigurable smart surface RIS to reach the base station. The specific process is as follows: In this system, there is a multi-antenna base station at a fixed location, a RIS at a fixed location, and multiple single-antenna user equipments; there are RIS-UE channels and BS-RIS channels in the system, and the two channels are combined into a reflection cascaded channel; the direct channel between the base station and the user equipment is blocked by obstacles and cannot directly transmit signals; the user sends an orthogonal pilot signal at a certain moment, which is reflected by the RIS and reaches the base station receiver. The received signal of the base station is obtained based on the cascaded channel matrix.

[0043] like Figure 2 As shown, an uplink RIS-mmWave MIMO system model is established using the Saleh-Valenzuela channel model. The UE transmits pilot signals through the RIS to the base station BS. It is assumed that the direct link between the UE and the base station BS is blocked. The base station BS and RIS are equipped with M antennas and N reflection elements respectively, and there are K single-antenna users. The system has two channels: RIS-UE and BS-RIS.

[0044] Furthermore, S101: Constructing the channel model of the millimeter-wave MIMO system specifically includes:

[0045] S101-1: The base station antenna uses a uniform linear array (ULA), and the RIS's reflector element also uses a uniform linear array. The channel is established using the Saleh-Valenzuela model, and the channel matrix between the RIS and the base station is defined. As shown in equation (1):

[0046]

[0047] Where L1 is the number of paths between base station BS and RIS. For the complex gain of the l1st path, Let l1 be the arrival angle of the path. Let (·) be the departure angle of the l1th path. H It is the conjugate transpose;

[0048] S101-2: RIS and the kth user equipment U k Channel model As shown in equation (2):

[0049]

[0050] Where L2 is BS and Uk The number of paths between them and These are the complex gain and angle of arrival for the l2nd path, respectively.

[0051] S101-3: Uniform linear array steering vector available express, As shown in equation (3):

[0052] a P (z)=[1,e -j2πz ,…,e -j2π(P-1)z ] T (3)

[0053] in,(·) T As a transpose, only the horizontal direction angle θ is considered in the ULA array, where P is the number of antennas or elements, and z is the direction cosine corresponding to angle θ, as shown in equation (4):

[0054]

[0055] Where, λ c Let λ be the carrier wavelength and d be the element spacing. Assume d ≤ λ. c / 2.

[0056] S101-4: The direct channel between the base station and the user equipment can be blocked by obstacles, preventing direct signal transmission. The signal is reflected by the RIS (Reflection Matrix), which adjusts the amplitude and phase of the incident signal. As shown in equation (6):

[0057]

[0058] Where, β n ∈[0,1] and α n ∈[0,2π] represents the amplitude and phase shift of the nth (n∈[1,2,…,N]) passive reflection unit in the RIS.

[0059] It should be understood that there is a one-to-one relationship between the spatial frequency and the physical angle on the ULA side. Therefore, for the sake of simplicity, the independent variable of the array steering vector is referred to as the angle.

[0060] Alternatively, S101-3 is replaced with:

[0061] The RIS reflector elements employ a uniform planar array (UPA). For a UPA of P = P1 × P2, P is the total number of antennas or elements, P1 is the number of antennas or elements in the horizontal direction, and P2 is the number of antennas or elements in the vertical direction. The UPA array takes into account the horizontal angle ξ and the elevation angle η, and its array steering vector... As shown in equation (5):

[0062]

[0063] in, For Kronecker product,

[0064] Since UPA is the Kronecker product of two ULAs, only the azimuth and elevation angles of the cascaded departure angle need to be estimated in S105, while the rest of the steps remain unchanged. Therefore, this invention can be easily extended to the case where RIS is UPA.

[0065] Further, in step S101: the user equipment sends an orthogonal pilot signal to the base station via a reconfigurable smart surface (RIS) to obtain the received signal from the base station, specifically including:

[0066] The entire channel estimation process consists of B subframes, and the user equipment U k Transmit an orthogonal pilot symbol sequence of length T in the b-th subframe (b∈B). After reflection through two channels and the RIS, the received signal at the BS end As shown in equation (7):

[0067]

[0068] in, noise power σ 2 Additive White Gaussian Noise (AWGN), I M Let G be an M-dimensional matrix of all ones, and let G = Hdiag(h k ) is a cascaded channel.

[0069] Further, S101: eliminating the pilot sequence of the received signal to obtain the received signal matrix specifically includes:

[0070] Orthogonal pilot signals have The characteristics of P t The transmit power is given. Utilizing the orthogonal pilot characteristics, multiply equation (7) by x on the right. k,b To eliminate the pilot sequence, as shown in equation (8):

[0071] y' k,b =Gφ+n' k,b (8)

[0072] in, and

[0073] By superimposing the received signals of b subframes, Y can be obtained from equation (8). k As shown in equation (9):

[0074] Y k =GΦ+N k (9)

[0075] in, For the received signal matrix, Here is the reflection coefficient matrix. This is the noise matrix.

[0076] Further, before concatenating the channel between the base station and the reconfigurable smart surface RIS with the channel between the reconfigurable smart surface RIS and the user equipment in step S102, the method further includes: converting the channel from a spatial domain representation to an angular domain representation; the conversion of the channel from a spatial domain representation to an angular domain representation specifically includes:

[0077] The channel H between BS and RIS is shown in equation (10):

[0078]

[0079] in, For the arrival angle matrix, For the gain of the BS-RIS channel, The RIS end departure angle matrix;

[0080] RIS and the kth user equipment U k Channel model h k As shown in equation (11):

[0081]

[0082] in, For the RIS end arrival angle matrix, This is the gain of the RIS-UE channel.

[0083] Further, S102: Concatenate the channel between the base station and the reconfigurable smart surface RIS with the channel between the reconfigurable smart surface RIS and the user equipment to obtain a concatenated channel, specifically including:

[0084] H and h k Expressed as a cascaded channel G, as shown in equation (12):

[0085]

[0086] in, For cascaded departure angle matrices, For cascaded path gain, The difference in departure angle is L = L1 × L2, which is the number of paths in the cascaded channel, and ⊙ is the Khatri-Rao product. Angle of arrival. For ease of expression, A will be used hereafter. C and A A To represent A respectively C (θ C ) and A A (θ A ).

[0087] Further, S102: transforming the expression of the cascaded channel into the product of the angle of arrival matrix, the path gain of the cascaded channel, and the conjugate transpose of the cascaded departure angle matrix, specifically includes:

[0088] The obtained G is transformed for subsequent calculations, as shown in equation (13):

[0089]

[0090] Where Λ'=diag[Γ] is the modified cascaded channel path gain, and Γ=vec(Λ)=[γ1,γ2,…,γ L [This refers to the path gain of the cascaded channel.] For the deformed angle of arrival matrix, It is a row vector consisting entirely of 1s.

[0091] Furthermore, step S102: simultaneously transforms the channel estimation problem into a sparse matrix recovery problem, specifically including:

[0092] The channel estimation problem is transformed into a sparse matrix recovery problem, as shown in equation (14):

[0093]

[0094] in, norm is The number of non-zero elements in the neutron. Let be the estimated concatenated channel matrix, and ε be the threshold for the estimation error, ||·|| F It is the Frobenius norm.

[0095] It should be understood that the channel is transformed from a spatial domain to an angular domain representation; using the properties of matrix operations Kronecker product and Khatri-Rao product, the cascaded channel is represented as the product of the base station's angle of arrival matrix, the cascaded channel path gain, and the conjugate transpose of the cascaded departure angle matrix; for ease of subsequent optimization, the cascaded channel path gain is transformed into a diagonal matrix form, because the number of paths in the cascaded channel corresponds to the product of the number of paths in the BS-RIS and RIS-UE, so the corresponding angle of arrival matrix needs to be expanded by a factor of the number of paths in the RIS-UE; the channel estimation problem is transformed into the estimation problem of the parameter angle of arrival matrix, the cascaded channel path gain, and the cascaded departure angle matrix; the sparse channel estimation problem is expressed as a constrained l0-norm optimization problem.

[0096] Further, S103: Based on the received signal matrix, utilizing the physical position invariance of the reconfigurable smart surface and the base station, and assuming different user equipment have the same angle of arrival, the number of angles of arrival and paths is estimated. Based on the number of angles of arrival and paths, an angle of arrival matrix and a grid are generated, specifically including:

[0097] S103-1: Calculate the covariance matrix of the received signal. And on Perform eigenvalue decomposition, from The feature vector V is obtained;

[0098] Determine the noise subspace U from V N And define the polynomial f(θ1) as shown in equation (15):

[0099]

[0100] Find the root θ1 of equation (15);

[0101] S103-2: Using the roots θ1 of the calculated polynomial, and utilizing d c =||θ1|-1| calculates the distance d from the unit circle. c Set threshold υ t Find all d c <υ t The number of roots is equal to the number of paths. Select The roots are used as the desired angle θ. A .

[0102] S103-3: Due to the invariance of the physical location of BS-RIS, for different users θ A The process is the same: randomly select the received signals of two user equipments to complete S103-1 and S103-2, and then perform error judgment.

[0103] If the two user equipment estimates θ AIf the difference is less than the threshold, then the average is obtained.

[0104] If the error exceeds the threshold, a third user is randomly selected to complete S103-1, and the set of data with an error exceeding the set threshold is removed. The average of the remaining two sets of data is then calculated.

[0105] Angle of arrival estimated after output error judgment

[0106] S103-4: Obtained using S103-2 and S103-3 and Calculate the angle of arrival matrix And Each column is copied L2 times and expanded to And Each column is copied (L2+2) times and expanded into a grid.

[0107] It should be understood that because the physical locations of BS and RIS are fixed, different users will have the same angle of arrival. The Root-MUSIC algorithm is used to estimate the angle of arrival, and a threshold is set to estimate the number of paths between BS and RIS. Error judgment is performed to prevent error propagation. Using the estimated angle of arrival and the number of paths, an angle of arrival matrix and a high-precision grid are generated. The specific process is as follows: Utilizing the invariance of the physical locations of BS and RIS, it can be determined that the angle of arrival of BS will remain constant; the Root-MUSIC algorithm is used to estimate the angle of arrival of the user's BS-RIS and calculate the distance from the unit circle, setting a threshold to estimate the number of paths; to prevent angle estimation errors from causing error propagation in subsequent estimations, two users are randomly selected for estimation, and the error is calculated. If it is not within the set threshold, a third user is randomly selected for estimation, erroneous estimates are removed, and the average of the remaining correct estimates is taken; if it is within the set threshold, the average is taken; the set threshold is used to detect the number of paths, ensuring the accuracy of the detected number of paths; the array response vector is calculated using the estimated angle of arrival and the number of paths, generating the angle of arrival matrix, which is repeatedly expanded into a high-precision grid.

[0108] like Figure 1 The first stage, as shown, aims to obtain the estimated angle of arrival using the Root-MUSIC algorithm. And set a threshold to determine the number of paths. Different users have the same To prevent errors from propagating from the two-stage estimation to the next stage, an error judgment method is employed. An angle-of-arrival matrix and a high-precision grid are generated using the estimated angles and path numbers.

[0109] Further, S104: transforming the sparse matrix recovery problem into an optimization problem, using the grid to estimate the cascaded departure angle matrix and cascaded channel path gain for different users, specifically including:

[0110] S104-1: Equation (14) is non-convex. We replace equation (14) with a logarithmic sum function, so the sparse recovery problem is as shown in equation (16):

[0111]

[0112] Here, δ is a positive parameter that guarantees that the value of the logarithmic function is not 0;

[0113] S104-2: Transform equation (16) in S104-1 into a Lagrangian function, resulting in an unconstrained optimization problem, as shown in equation (17):

[0114]

[0115] The purpose of the regularization weight parameter λ is to control the sparsity of the solution and the data fitting error.

[0116] S104-3: By replacing equation (17) in S104-2 with an iterative substitution function, the optimal solution is obtained through t iterations, as shown in equation (18):

[0117]

[0118] Among them, D (t) As shown in equation (19):

[0119]

[0120] in, This is the gain estimate for the t-th iteration.

[0121] S104-4: Finding the optimal value through the gain function Find the optimal angle function To find the optimal value of equation (17), the gain function is shown in equation (20):

[0122]

[0123] Among them, C b =A A diag(A C φ). The angle function is shown in equation (21):

[0124]

[0125] S104-5: Using Singular Value Decomposition (SVD) preprocessing, find the closest approximation to the true θ. C Y can be obtained from equations (9) and (13). k As shown in equation (22):

[0126] Y k =A' A diag(Γ)(Φ H A C ) H +N k (twenty two)

[0127] After SVD, θ C,l The corresponding l right singular vectors are shown in equation (23):

[0128] v i ≈Φ H a N (θ C,l ) / ||Φ H a N (θ C,l )||2 (23)

[0129] Where, ||·|2 is the 2-norm. i As

[0130] S104-6: Obtained using S103 And obtained from S104-5 Substituting into equation (20) yields the cascaded channel path gain.

[0131] S104-7: The expression for the residual r of the t-th iteration is shown in equation (24):

[0132]

[0133] S104-8: r obtained from S104-7 (t) The weight parameter λ is updated as shown in equation (25):

[0134] λ=min(f / r (t) ,λ max (25)

[0135] Where f is a constant scaling factor, λ max These are the weight parameters.

[0136] S104-9: Obtained using S103 S104-5 obtained And obtained from S104-6 Construct a constructor using equation (21)

[0137] S104-10: Update using gradient descent algorithm As shown in equation (26):

[0138]

[0139] Where μ is the step size; by adjusting the step size, we can prevent getting trapped in local optimization.

[0140] S104-11: Updated using S104-10 Japanese style (20) update Set the clipping threshold γ t When γ l <γ t Cut off the path;

[0141] S104-12: Determine if the value is less than the termination threshold. Set the termination threshold. when When, cycle from S104-5 to S104-11; when When the loop terminates, output the value. and

[0142] To understand, the optimization method for estimating the cascaded channel path gain and cascaded departure angle matrix involves the following steps: replacing the l0 norm with a logarithmic sum function; transforming the constrained optimization problem into an unconstrained one by adding a regularization parameter to convert it into a Lagrangian function; replacing the logarithmic sum function with an iterative surrogate function; solving the optimization problem by updating the iterative weight parameters using an iterative reweighting algorithm; finding the optimal solution for the cascaded channel path gain and the corresponding optimal solution with respect to the angle function; finding the optimal cascaded departure angle using a gradient descent algorithm and adjusting the step size; performing preprocessing using singular value decomposition to find an initial value close to the true cascaded departure angle; repeating the above optimization steps until a threshold is reached, thus estimating the cascaded channel path gain and cascaded departure angle matrix, which are then multiplied by the arrival angle matrix to obtain the estimated cascaded channel matrix.

[0143] Step S104 corresponds to Figure 1 The second phase aims to obtain through S103 Estimate the value of different users using SVD, iterative reweighting algorithm and gradient descent algorithm and

[0144] Further, S105: Based on the angle of arrival matrix, the concatenated departure angle matrix, and the concatenated channel path gain, the estimated concatenated channel matrix is ​​calculated, specifically including:

[0145] Utilize and Calculate the estimated cascaded channel matrix As shown in equation (27):

[0146]

[0147] Further, S105: Based on the estimated concatenated channel matrix, obtain channel state information, specifically including: channel state information refers to the concatenated channel matrix.

[0148] The parameter settings for the RIS-mmWave communication system are shown in Table 1.

[0149] Table 1 Parameter Settings

[0150]

[0151] The normalized mean square error (NMSE) was compared under different signal-to-noise ratios (SNR) and different training overheads. The NMSE is shown in Equation (27):

[0152]

[0153] in, To determine the expected value, the complexity of channel estimation is compared by calculating the sum of the running times for each round. Results are obtained using 100 Monte Carlo experiments.

[0154] To demonstrate that the present invention achieves higher channel estimation accuracy and lower training overhead without significantly increasing complexity, NMSE was compared under different SNRs and different training overheads. The running time of the present invention was also compared with that of the highest-accuracy channel estimation method to verify the channel estimation performance of the present invention.

[0155] Figure 3 This illustrates the angle of arrival and path number for the channel estimation in the first stage of this invention, with the estimated angle value at a point on the unit circle. Figure 3 It can be seen that there are 2 points closest to the unit circle, so the estimated number of paths is 2.

[0156] Under different training overheads, this invention is compared with three different compressed sensing channel estimation methods: OMP, SOMP, and SMJCE. Figure 4 The NMSE under different training overheads is shown for 16 and 64 BS antennas, with SNR set to 10dB. Figure 4As can be seen, the NMSE decreases with increasing training overhead, and this invention has a smaller NMSE under the same training overhead. It can also be observed that the larger the number of BS antennas, the smaller the NMSE.

[0157] Under different SNRs, this invention is compared with three different compressed sensing channel estimation methods: OMP, SOMP, and SMJCE, and three common channel estimation methods: LS, MVU, and dual time-scale channel estimation. Figure 5 and Figure 6 The NMSE is shown for different SNR values ​​with 64 and 16 BS antennas, with a training overhead of 64. Figure 5 and Figure 6 It is evident that as the SNR increases, the NMSE decreases. Therefore, this invention exhibits a smaller NMSE at the same SNR, demonstrating better estimation performance. Furthermore, it can be observed that the larger the number of BS antennas, the smaller the NMSE.

[0158] Figure 7 This is a curve showing the runtime as a function of rounds, with the time for each round being summed. The complexity of the most accurate method, SMJCE, and this invention were compared by summing their runtimes. Figure 7 It can be seen that the present invention has a shorter running time and can maintain a lower complexity.

[0159] Based on the sparsity of wireless channels and the characteristics of RIS (Real-Resolution Interval), this invention proposes a two-stage super-resolution parameter channel estimation method based on RIS. This method improves channel estimation accuracy and reduces training overhead by first estimating a precise angle and then using angle optimization to estimate another precise angle. Applying this method can improve the channel estimation accuracy and reduce training overhead in wireless communication systems.

[0160] This invention discloses a two-stage super-resolution parameter channel estimation method based on RIS, comprising: First, representing the cascaded channel as the product of the angle of arrival matrix, the cascaded channel path gain, and the conjugate transpose of the cascaded departure angle matrix. Then, transforming the cascaded channel path gain into a diagonal matrix, and correspondingly expanding the angle of arrival matrix. Finally, estimating the parameters using the two-stage super-resolution parameter channel estimation method based on RIS. In the first stage, leveraging the physical location invariance of RIS and BS, different users have the same angle of arrival, and using the Root-MUSIC algorithm and error judgment method to accurately estimate the angle of arrival and the number of paths. The estimated angle of arrival matrix and the number of paths are used to generate the angle of arrival matrix and a high-precision grid. In the second stage, the sparse recovery problem is transformed into an optimization problem, using an iterative reweighting algorithm, a gradient descent optimization algorithm, and the high-precision grid from the first stage to estimate the cascaded departure angle matrix and cascaded channel path gain for different users. To reduce complexity, singular value decomposition is used for preprocessing. Furthermore, dynamic adjustment of the step size and path pruning methods are employed to further improve accuracy. This invention achieves high estimation accuracy and low training overhead with acceptable computational complexity, greatly improving channel estimation performance.

[0161] Example 2 This example provides a two-stage super-resolution parameter channel estimation system based on RIS, including: a construction module configured to: construct a millimeter-wave MIMO system assisted by a reconfigurable smart surface RIS; construct a channel model of the millimeter-wave MIMO system; the user equipment sends orthogonal pilot signals through the reconfigurable smart surface RIS to the base station, obtains the received signal of the base station, eliminates the pilot sequence of the received signal, and obtains the received signal matrix;

[0162] The cascading module is configured to: cascade the channel between the base station and the reconfigurable smart surface RIS with the channel between the reconfigurable smart surface RIS and the user equipment to obtain a cascaded channel; transform the expression of the cascaded channel into the product of the angle of arrival matrix, the path gain of the cascaded channel, and the conjugate transpose of the cascaded departure angle matrix, and at the same time transform the channel estimation problem into a sparse matrix recovery problem;

[0163] The first estimation module is configured to: estimate the angle of arrival and the number of paths based on the received signal matrix, taking advantage of the invariance of the physical location of the reconfigurable smart surface and the base station, and the fact that different user equipments have the same angle of arrival; and generate the angle of arrival matrix and grid based on the angle of arrival and the number of paths.

[0164] The second estimation module is configured to: transform the sparse matrix recovery problem into an optimization problem, and estimate the cascaded departure angle matrix and cascaded channel path gain for different users using the grid.

[0165] The output module is configured to: calculate the estimated concatenated channel matrix based on the angle of arrival matrix, the concatenated departure matrix, and the concatenated channel path gain; and finally, obtain channel state information based on the estimated concatenated channel matrix.

[0166] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A two-stage super-resolution parameter channel estimation method based on RIS, characterized by: include: Constructing a reconfigurable smart surface RIS-assisted millimeter-wave multiple-input multiple-output MIMO system; Constructing a channel model for a millimeter-wave MIMO system; The user equipment sends orthogonal pilot signals through the reconfigurable smart surface RIS to the base station, obtains the received signal from the base station, eliminates the pilot sequence of the received signal, and obtains the received signal matrix. The channel between the base station and the reconfigurable smart surface (RIS) is concatenated with the channel between the RIS and the user equipment to obtain the concatenated channel. The expression of the concatenated channel is transformed into the product of the angle of arrival matrix, the path gain of the concatenated channel, and the conjugate transpose of the concatenated departure angle matrix. At the same time, the channel estimation problem is transformed into a sparse matrix recovery problem. Based on the received signal matrix, the physical location invariance of the reconfigurable smart surface and the base station, as well as the fact that different user equipment have the same angle of arrival, is used to estimate the angle of arrival and the number of paths; based on the angle of arrival and the number of paths, the angle of arrival matrix and grid are generated. The sparse matrix recovery problem is transformed into an optimization problem, and the cascaded departure angle matrix and cascaded channel path gain for different users are estimated using the grid. Based on the angle of arrival matrix, the concatenated departure angle matrix, and the concatenated channel path gain, the estimated concatenated channel matrix is ​​calculated; finally, based on the estimated concatenated channel matrix, channel state information is obtained. Based on the received signal matrix, the physical location invariance of the reconfigurable smart surface and the base station, as well as the fact that different user equipment have the same angle of arrival, is used to estimate the angle of arrival and the number of paths. Based on the number of angles of arrival and paths, generate an angle of arrival matrix and a grid, specifically including: Step (301): Calculate the covariance matrix of the received signal. and to Perform eigenvalue decomposition, from Obtain the feature vector ;Depend on Determine the noise subspace and define polynomial As shown in equation (15): (15) Find the root of equation (15). ; Step (302): Use the roots of the calculated polynomial ,use Calculate the distance from the unit circle. Set threshold Find all The number of roots is equal to the number of paths. Select The root is used as the desired angle of arrival. ; Step (303): Due to the invariance of the physical location of BS-RIS, for different users The steps are the same: randomly select the received signals of two user equipments to complete steps (301) and (302), and perform error judgment. If the two user equipment estimates If the difference is less than the threshold, then the average is obtained. ; If the error is greater than the threshold, a third user is randomly selected to complete steps (301) and (302), and the set of data with an error greater than the set threshold is removed. The average of the remaining two sets of data is then calculated. ; Angle of arrival estimated after output error judgment ; Step (304): Utilize the obtained and Calculate the angle of arrival matrix and will Copy each column The second expansion is and will Copy each column The next step is to expand it into a grid.

2. The two-stage super-resolution parameter channel estimation method based on RIS as described in claim 1, characterized in that, Constructing a channel model for a millimeter-wave MIMO system specifically includes: The base station antenna uses a uniform linear array, and the RIS's reflector element also uses a uniform linear array. The channel is established using the Saleh-Valenzuela model, and the channel matrix between the RIS and the base station is calculated. As shown in equation (1): (1) in, This represents the number of paths between BS and RIS. For the first Path complex gain, For the first The path reaches the angle. For the first The path leaves the corner. For the conjugate transpose, M is the number of BS antennas; RIS and the k individual user devices Channel model As shown in equation (2): (2) in, For BS and The number of paths between them and The first Path complex gain and angle of arrival; Uniform linear array steering vector available express, As shown in equation (3): (3) in, As a transpose, only the horizontal direction angle is considered in the ULA array. , The number of antennas or units. for The direction cosine corresponding to the angle is shown in equation (4): (4) in, For carrier wavelength, Assuming the element spacing is... ; The direct channel between the base station and the user equipment can be blocked by obstacles, preventing direct signal transmission; the signal is reflected by the RIS (Reflection Matrix), which adjusts the amplitude and phase of the incident signal. As shown in equation (6): (6) in, and These are the first in RIS n The amplitude and phase shift of each passive reflector unit, .

3. The two-stage super-resolution parameter channel estimation method based on RIS as described in claim 2, characterized in that, User equipment sends orthogonal pilot signals through a reconfigurable smart surface (RIS) to the base station, and obtains the received signals from the base station, specifically including: The entire channel estimation process includes B Subframes, User Equipment In the The length of each subframe transmission is T orthogonal pilot symbol sequence , After reflection through two channels and the RIS, the received signal at the BS end As shown in equation (7): (7) in, noise power Additive white Gaussian noise, For dimension M A matrix of all 1s It is a cascaded channel; The pilot sequence of the received signal is eliminated to obtain the received signal matrix, which specifically includes: Orthogonal pilot signals have , The characteristics of, among which For the transmission power; utilizing the orthogonal pilot characteristics, multiply the right side of equation (7) by... To eliminate the pilot sequence, as shown in equation (8): (8) in, and ; Will b The received signals of each subframe are superimposed, and can be obtained from equation (8). As shown in equation (9): (9) in, For the received signal matrix, Here is the reflection coefficient matrix. This is the noise matrix.

4. The two-stage super-resolution parameter channel estimation method based on RIS as described in claim 3, characterized in that, Before concatenating the channel between the base station and the reconfigurable smart surface (RIS) with the channel between the RIS and the user equipment, the method further includes: converting the channel from a spatial domain representation to an angular domain representation; the conversion of the channel from a spatial domain representation to an angular domain representation specifically includes: Channel between BS-RIS As shown in equation (10): (10) in, For the arrival angle matrix, For the gain of the BS-RIS channel, The RIS end departure angle matrix; RIS and the k individual user devices Channel model As shown in equation (11): (11) in, For the RIS end arrival angle matrix, For the gain of the RIS-UE channel; The channel between the base station and the reconfigurable smart surface (RIS) is concatenated with the channel between the reconfigurable smart surface (RIS) and the user equipment to obtain a concatenated channel, which specifically includes: Will and Expressed as a concatenated channel As shown in equation (12): (12) in, For cascaded departure angle matrices, For cascaded path gain, Due to the difference in departure angle, This represents the number of paths in the cascaded channel. For Khatri-Rao product, The angle of arrival.

5. The two-stage super-resolution parameter channel estimation method based on RIS as described in claim 4, characterized in that, The expression for a cascaded channel is transformed into the product of the angle of arrival matrix, the path gain of the cascaded channel, and the conjugate transpose of the cascaded departure angle matrix, specifically including: The result The shape is modified for subsequent calculations, as shown in equation (13): (13) in, For the modified cascaded channel path gain, For cascaded channel path gain, For the deformed angle of arrival matrix, It is a row vector consisting entirely of 1s.

6. The two-stage super-resolution parameter channel estimation method based on RIS as described in claim 5, characterized in that, Simultaneously, the channel estimation problem is transformed into a sparse matrix recovery problem, specifically including: The channel estimation problem is transformed into a sparse matrix recovery problem, as shown in equation (14): (14) in, norm is The number of non-zero elements in the neutron. For the estimated cascaded channel matrix, To estimate the threshold of error, It is the Frobenius norm.

7. The two-stage super-resolution parameter channel estimation method based on RIS as described in claim 6, characterized in that, The sparse matrix recovery problem is transformed into an optimization problem. Using the aforementioned grid, the cascaded departure angle matrix and cascaded channel path gain for different users are estimated, specifically including: Step (401): Equation (14) is non-convex. We replace equation (14) with a logarithmic sum function, so the sparse recovery problem is as shown in equation (16): (16) in, It is a positive parameter that guarantees that the value within the logarithmic function is not 0; Step (402): Transform equation (16) in step (401) into a Lagrangian function, resulting in an unconstrained optimization problem, as shown in equation (17): (17) Among them, the regularization weight parameter The goal is to control the sparsity of the solutions and the data fitting error; Step (403): Replace equation (17) in step (402) with an iterative substitution function, and through t After several iterations, the optimal solution is obtained, as shown in equation (18): (18) in, As shown in equation (19): (19) in, For the first t The gain estimate for the next iteration; Step (404): Find the optimal value through the gain function Find the optimal angle function To find the optimal value of equation (17), the gain function is shown in equation (20): (20) in, The angle function is shown in equation (21): (21) Step (405): Use singular value decomposition preprocessing to find the closest approximation to reality. From equations (9) and (13), we can obtain As shown in equation (22): (22) After singular value decomposition correspond There are three right singular vectors, as shown in equation (23): (23) in, It is a 2-norm; As ; Step (406): Utilize And obtained from step (405) Substituting into equation (20) yields the cascaded channel path gain. ; Step (407): The t The residual of the next iteration r The expression for is shown in equation (24): (24) Step (408): Using the results obtained in step (407) Update weight parameters As shown in equation (25): (25) in, f A constant scaling factor. These are weight parameters; Step (409): Utilize The result obtained in step (405) And obtained from step (406) Construct a constructor using equation (21) ; Step (410): Update using gradient descent algorithm As shown in equation (26): (26) in, The step size is used to prevent getting trapped in local optimization. Step (411): Updated using step (410) Japanese style (20) update Set the clipping threshold ,when Cut off the path; Step (412): Determine if the value is less than the termination threshold, and set the termination threshold. ;when When, cycle through step (405) to step (411); when When the loop terminates, output the value. and .

8. The two-stage super-resolution parameter channel estimation method based on RIS as described in claim 7, characterized in that, Based on the angle-of-arrival matrix, the concatenated departure angle matrix, and the concatenated channel path gain, the estimated concatenated channel matrix is ​​calculated, specifically including: Utilize , and Calculate the estimated cascaded channel matrix As shown in equation (27): (27)。 9. A two-stage super-resolution parameter channel estimation system based on RIS, characterized in that... include: The building block is configured to: build a reconfigurable smart surface RIS-assisted millimeter-wave MIMO system; Constructing a channel model for a millimeter-wave MIMO system; The user equipment sends orthogonal pilot signals through the reconfigurable smart surface RIS to the base station, obtains the received signal from the base station, eliminates the pilot sequence of the received signal, and obtains the received signal matrix. The cascading module is configured to: cascade the channel between the base station and the reconfigurable smart surface RIS with the channel between the reconfigurable smart surface RIS and the user equipment to obtain a cascaded channel; transform the expression of the cascaded channel into the product of the angle of arrival matrix, the path gain of the cascaded channel, and the conjugate transpose of the cascaded departure angle matrix, and at the same time transform the channel estimation problem into a sparse matrix recovery problem; The first estimation module is configured to: estimate the angle of arrival and the number of paths based on the received signal matrix, utilizing the invariance of the physical location of the reconfigurable smart surface and the base station, and the fact that different user equipment have the same angle of arrival; and generate the angle of arrival matrix and grid based on the angle of arrival and the number of paths. The second estimation module is configured to: transform the sparse matrix recovery problem into an optimization problem, and estimate the cascaded departure angle matrix and cascaded channel path gain for different users using the grid. The output module is configured to: calculate the estimated concatenated channel matrix based on the angle of arrival matrix, the concatenated departure matrix, and the concatenated channel path gain; and finally, obtain channel state information based on the estimated concatenated channel matrix. Based on the received signal matrix, the physical location invariance of the reconfigurable smart surface and the base station, as well as the fact that different user equipment have the same angle of arrival, is used to estimate the angle of arrival and the number of paths. Based on the number of angles of arrival and paths, generate an angle of arrival matrix and a grid, specifically including: Step (301): Calculate the covariance matrix of the received signal. and to Perform eigenvalue decomposition, from Obtain the feature vector ;Depend on Determine the noise subspace and define polynomial As shown in equation (15): (15) Find the root of equation (15). ; Step (302): Use the roots of the calculated polynomial ,use Calculate the distance from the unit circle Set threshold Find all The number of roots is equal to the number of paths. Select The root is used as the desired angle of arrival. ; Step (303): Due to the invariance of the physical location of BS-RIS, for different users The steps are the same: randomly select the received signals of two user equipments to complete steps (301) and (302), and perform error judgment. If the two user equipment estimates If the difference is less than the threshold, then the average is obtained. ; If the error is greater than the threshold, a third user is randomly selected to complete steps (301) and (302), and the set of data with an error greater than the set threshold is removed. The average of the remaining two sets of data is then calculated. ; Angle of arrival estimated after output error judgment ; Step (304): Utilize the obtained and Calculate the angle of arrival matrix and will Copy each column The second expansion is and will Copy each column The next step is to expand it into a grid.

Images