Millimeter wave MIMO sparse channel estimation method based on dynamic path verification mechanism

Abstract

The application discloses a kind of millimeter wave MIMO sparse channel estimation method based on dynamic path verification mechanism, belong to wireless communication technical field, solve the technical problem of low estimation accuracy in prior art cannot be adaptively adjusted according to real channel complexity.This application provides millimeter wave MIMO sparse channel estimation method based on dynamic path verification mechanism based on the dynamic path verification estimation of measured data, i.e.based on real millimeter wave channel measurement data training and testing, dynamic path verification mechanism is introduced in each iteration;Combined with the sparsity adaptive mechanism, according to the training statistical information dynamic control iteration termination, based on the output estimation vector and reconstructs channel in the verification of measured data;Particularly, a kind of sparse channel estimation method combining dynamic path verification mechanism and sparsity adaptive strategy is suitable for low-complexity robust estimation under the environment of measured channel, especially suitable for low SNR or deep fading system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless communication, in particular to a millimeter wave MIMO sparse channel estimation method based on a dynamic path verification mechanism. BACKGROUND

[0002] Millimeter wave communication is considered as one of the key supporting technologies for 5G and future 6G due to its wideband spectrum resources. Millimeter wave MIMO systems usually employ large-scale array and hybrid beamforming architecture. Due to the serious path loss and shadowing effect of high-frequency signals during propagation, millimeter wave channels usually exhibit angle-domain or delay-domain sparsity. In recent years, many sparse channel estimation methods have been proposed, such as orthogonal matching pursuit (OMP), compressed sampling matching pursuit (CoSaMP), weighted OMP (WOMP), sparse Bayesian learning (SBL), etc. However, these methods generally have the following problems: (1) Most methods use a fixed sparsity parameter, which cannot be adaptively adjusted according to the complexity of the real channel; (2) Lack of path verification mechanism, easy to introduce false paths, affecting the estimation accuracy; (3) Although SBL and other methods have better performance, the computational complexity is high, which is not conducive to real-time system implementation; (4) Most researches are based on ideal geometric channel models, lack of adaptation and verification for real measured channels.

[0003] Therefore, there is an urgent need for a sparse channel estimation scheme that can adapt to complex actual channel structure and has robustness and low complexity, a channel estimation method for millimeter wave multiple-input multiple-output (mmWave MIMO) system. SUMMARY

[0004] The present application provides a millimeter wave MIMO sparse channel estimation method based on a dynamic path verification mechanism, which is trained and tested based on real millimeter wave channel measurement data, close to engineering practice; a dynamic path verification mechanism is introduced in each iteration to improve the path selection accuracy; combined with a sparsity adaptive mechanism, the iteration termination is dynamically controlled according to the training statistical information; a channel estimation algorithm with high precision, low complexity and strong generalization ability is realized.

[0005] To achieve the above purpose, the present application is implemented by the following technical scheme: the millimeter wave MIMO sparse channel estimation method based on the dynamic path verification mechanism provided by the present application is trained and tested based on real millimeter wave channel measurement data, and a dynamic path verification mechanism is introduced in each iteration; combined with a sparsity adaptive mechanism, the iteration termination is dynamically controlled according to the training statistical information, and the specific steps include the following:

[0006] S1. Constructing channel observation model: based on the OFDM frame structure of hybrid analog-digital structure, the channel observation model of the millimeter wave MIMO system is established; the channel is sparsely modeled in the angle domain, and a measurement matrix is constructed;

[0007] S2. Dynamic path verification weighted OMP algorithm: the input includes the received measurement vector, the measurement matrix, the dictionary matrix, and the sparsity upper limit, the path verification candidate number and the threshold parameter; the algorithm steps include initialization, calculation of the correlation of the current residual to all candidate paths, path candidate screening, dynamic path verification, path selection and support set updating, termination condition judgment;

[0008] Among them, the dynamic path verification mechanism is specifically: in each iteration, multiple candidate paths are selected for verification, and least square estimation and residual calculation are performed on each candidate path, and the optimal path is selected to update the support set;

[0009] The adaptive termination mechanism is: setting the residual drop rate threshold and the absolute residual threshold as the double judgment conditions, and adaptively controlling the number of iterations according to the channel quality condition;

[0010] S3. The estimated vector is output and the channel is reconstructed based on the measured data in the verification.

[0011] Preferably, the channel observation model in step S1 is specifically as follows:

[0012] The millimeter wave MIMO system adopts a hybrid analog-digital structure, and based on the OFDM frame structure, the received signal of the i-th subcarrier is modeled as follows:

[0013] (1)

[0014] Among them, is the frequency domain channel matrix on the i-th subcarrier; , , respectively represent the transmit and receive hybrid beamforming matrix in the i-th time slot of the training frame, and the superscript represents the conjugate transpose; represents the pilot vector of the i-th training frame, and the dimension is , wherein is the number of data streams; represents a Gaussian white noise;

[0015] The channel is sparsely modeled in the angle domain and represented as:

[0016] (2)

[0017] In the above formula, ​​​, Represent the dictionary matrices of the transmitting and receiving arrays respectively, and their column vectors are the response vectors in the discrete angle domain direction;

[0018] For the The subcarrier angle domain sparse channel matrix is ​​vectorized and the measurement matrix is ​​constructed. ,have to

[0019] (3)

[0020] The measurement matrix Representation matrix The complex conjugate of represents the transpose of the matrix, represents the Kronecker product; For the The receiving measurement vector corresponding to the subcarriers is For the The angle domain sparse channel vector under subcarriers.

[0021] Preferably, in step S2, the dynamic path verification weighted OMP algorithm is as follows:

[0022] enter:

[0023] Receive measurement vector , measurement matrix , dictionary matrix ;

[0024] Sparsity upper limit , path verification candidate number ;

[0025] Threshold parameter 、 ;

[0026] Output:

[0027] Sparse channel estimation vector , channel estimation matrix ;

[0028] The specific algorithm steps include the following:

[0029] S21. Initialization: Initial residual , support set is an empty set, the number of iterations ;

[0030] S22. Calculate the correlation:

[0031] Calculate the current residual r (t) Correlation of all candidate paths:

[0032]

[0033] in, is the measurement matrix, represents the residual vector of the previous iteration, is the projection coefficient of all paths and the residual of the previous iteration;

[0034] S23. Path candidate screening:

[0035] according to Sort the paths by their magnitude and select the one with the largest magnitude. The paths constitute the candidate set .

[0036] Preferably, a dynamic path verification mechanism is used to select multiple candidate paths for verification in each iteration. The specific steps are as follows:

[0037] a) Calculate the current residual r (t) Correlation with all candidate paths;

[0038] b) Select the top K paths with the greatest correlation as the candidate set ;

[0039] c) For the candidate set Perform least squares estimation and residual calculation on each candidate path;

[0040] d) Select the path with the smallest residual as the optimal path and update the support set.

[0041] Preferably, the dynamic path verification mechanism includes the following specific algorithms:

[0042] Pair Collection Each path Perform the following steps:

[0043] a) Set up a temporary support set:

[0044] b) Solve the least squares problem and estimate the sparse coefficients:

[0045]

[0046] c) Calculate the residual:

[0047] ;

[0048] Path selection and support set update:

[0049] Select the optimal path:

[0050] and update the support set ; current residual .

[0051] Preferably, the termination condition in the adaptive termination mechanism is determined as follows:

[0052] If any of the following conditions is met, the iteration is terminated:

[0053] a) the residual reduction rate is too low:

[0054]

[0055] b) the absolute value of the current residual is too small:

[0056] wherein, is the residual reduction rate threshold, which is set to 0.01, is the convergence threshold, which is set to .

[0057] Preferably, in the verification based on the measured data, the estimated vector is output and the channel is reconstructed:

[0058] a) constructing the support subset of the sparse dictionary , obtaining the channel vector:

[0059] b) reconstructing the vector into a two-dimensional channel matrix: ;

[0060] wherein, denotes the sparse dictionary matrix, denotes the sparse coefficient vector.

[0061] The present application provides a millimeter wave MIMO sparse channel estimation method based on a dynamic path verification mechanism. It has the following advantages:

[0062] (1) The present application adopts training and testing based on real millimeter wave channel measurement data, which is close to engineering practice. In each iteration, a dynamic path verification mechanism is introduced to improve the path selection accuracy. Combined with the sparse degree adaptive mechanism, the iteration termination is dynamically controlled according to the training statistical information. A high-precision, low-complexity channel estimation algorithm with strong generalization ability is realized. It has the following advantages:

[0063] The dynamic path verification mechanism is introduced: Compared with the traditional OMP algorithm which only selects the maximum correlation path in each round, the proposed scheme performs residual verification in multiple candidate paths, effectively avoiding the error problem caused by early misselection of paths, and significantly improving the estimation accuracy and robustness.

[0064] The adaptive termination mechanism is reasonable: through the residual error reduction rate threshold and the absolute residual error double judgment mechanism, the number of iterations can be adaptively controlled according to different channel quality conditions (such as low SNR or deep fading), overfitting and unnecessary calculation overhead are avoided, and the accuracy and efficiency are considered.

[0065] The implementation complexity is controllable: although the multi-path verification mechanism is introduced, the overall calculation complexity is still significantly lower than that of the global search algorithm because the residual error estimation of a limited number of paths is performed in each round.

[0066] The robustness is better than that of the traditional OMP: the simulation experiment and the evaluation results of the public measured data set (North Carolina State University AIfor 5G Challenge channel data set) show that the algorithm has lower NMSE and more stable convergence performance than the baseline algorithm (such as weighted OMP) under the measured channel conditions. BRIEF DESCRIPTION OF DRAWINGS

[0067] Figure 1 is the measured data channel acquisition process diagram based on embodiment 2 of the present application;

[0068] Figure 2 is the NMSE performance comparison curve diagram of the present application and different algorithms under different SNR conditions;

[0069] Figure 3 is the NMSE performance comparison curve diagram of the present application and different algorithms under different training frame numbers. DETAILED DESCRIPTION

[0070] The technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments.

[0071] In order to further illustrate the actual operability of the present application, two specific embodiments are provided. Embodiment 1

[0072] As shown in Figure 1 , the millimeter wave MIMO sparse channel estimation method based on the dynamic path verification mechanism of the present application includes the following contents:

[0073] S1: Channel observation model

[0074] The millimeter wave MIMO system adopts a hybrid analog-digital structure, and the received signal of the first subcarrier is modeled as follows based on the OFDM frame structure:

[0075] (1)

[0076] in For the Frequency domain channel matrix on subcarriers; , Represents the training frame The transmit and receive hybrid beamforming matrix in the time slots, superscript It means to find the conjugate transpose; Indicates the The pilot vector of training frames has the dimension ,in is the number of data streams; represents Gaussian white noise.

[0077] The channel is expressed as follows under sparse modeling in the angle domain:

[0078] (2)

[0079] In the above formula, , Represent the dictionary matrices of the transmitting and receiving arrays respectively, and their column vectors are the response vectors in the discrete angle domain direction. For the The subcarrier angle domain sparse channel matrix.

[0080] Vectorize it and construct the measurement matrix ,have to

[0081] (3)

[0082] The measurement matrix Representation matrix The complex conjugate of represents the transpose of the matrix, represents the Kronecker product. For the The receiving measurement vector corresponding to the subcarriers is For the The angle domain sparse channel vector under subcarriers.

[0083] S2: Dynamic path verification weighted OMP algorithm

[0084] enter:

[0085] Receive measurement vector , measurement matrix , dictionary matrix

[0086] Sparsity upper limit , path verification candidate number

[0087] Threshold parameter 、

[0088] Output:

[0089] Sparse channel estimation vector , Channel estimation matrix

[0090] The specific algorithm steps are as follows:

[0091] 1. Initialization: initial residual , support set is an empty set, and the number of iterations ;

[0092] 2. Calculate the correlation:

[0093] Calculate the correlation of the current residual r (t) for all candidate paths:

[0094]

[0095] wherein is the measurement matrix, denotes the residual vector of the last iteration, is the projection coefficient of all paths and the residual of the last iteration.

[0096] 3. Path candidate screening:

[0097] Sort the paths according to the magnitude of , and select the first paths with the largest magnitude to form the candidate set .

[0098] 4. Dynamic path verification:

[0099] For each path in the candidate set , the following steps are performed:

[0100] a) Set the temporary support set:

[0101] b) Solve the least squares problem to estimate the sparse coefficient:

[0102]

[0103] c) Calculate the residual:

[0104]

[0105] 5. Path selection and support set update:

[0106] Select the optimal path:

[0107] and update the support set ; current residual :

[0108] 6. termination condition judgment:

[0109] If any of the following conditions is met, terminate the iteration:

[0110] a) the residual reduction rate is too low:

[0111]

[0112] b) the absolute value of the current residual is too small:

[0113] wherein, is the residual reduction rate threshold, set to 0.01, is the convergence threshold, set to .

[0114] 7. output the estimated vector and reconstruct the channel:

[0115] a) construct the support subset of the sparse dictionary , and obtain the channel vector:

[0116] b) reconstruct the vector into a two-dimensional channel matrix: .

[0117] wherein, denotes the sparse dictionary matrix, denotes the sparse coefficient vector. Embodiment 2

[0118] The dynamic path verification estimation based on measured data of the present application. This embodiment uses the real scene channel measurement samples provided in the public data set to construct a typical single user millimeter wave MIMO channel estimation scene, and uses the method proposed in the present application to complete the channel estimation.

[0119] System parameter setting:

[0120] 1. Use the real scene channel measurement samples in the AI for 5G Challenge channel dataset of North Carolina State University, each channel has multiple paths, including angle, gain, time delay, etc.

[0121] 2. Number of transmit antennas: ; number of receive antennas , both are uniform linear arrays;

[0122] 3. Number of radio frequency links: , carrier frequency is 28 GHz, subcarrier number ;

[0123] 4. Analog beamforming matrix , Adopting a directional beam based on discrete Fourier transform;

[0124] 5. The number of training frames used for each estimation is 20;

[0125] 6. Upper limit of sparsity , path verification candidate number .

[0126] Estimation process:

[0127] 1. Adopting channel data in the measured scene , construct observation vector ;

[0128] 2. Using the method of the application, perform dynamic path iteration for each subcarrier ;

[0129] Initialize the residual and support set.

[0130] Calculate the correlation of the current residual with all candidate paths.

[0131] Select the top K paths with the largest correlation as the candidate set.

[0132] Perform least squares estimation and residual calculation for each candidate path.

[0133] Select the path with the smallest residual to update the support set.

[0134] Judge the termination condition, if not satisfied, continue iteration.

[0135] 3. Finally obtain channel estimation ;

[0136] 4. Calculate the normalized mean square error index to verify the algorithm performance.

[0137] Verification results:

[0138] The method of the application is superior to the traditional weighted OMP and its improved algorithm under the condition of low signal-to-noise ratio and limited training frames. Simulation results show that the dynamic path verification mechanism can effectively eliminate the misselected paths and improve the estimation accuracy. Embodiment 3

[0139] The millimeter wave MIMO sparse channel estimation method based on the dynamic path verification mechanism of the application is compared with the measured data results of the existing millimeter wave MIMO sparse channel simulation estimation method, as Figure 2 and Figure 3The simulation performance comparison results are shown.

[0140] The Matlab simulation is adopted, the system and channel parameters are consistent with those in Embodiment 2, and the normalized mean square error is defined as: .

[0141] The algorithm of the application is compared with the following traditional sparse channel estimation algorithms: weighted OMP, improved weighted OMP, improved weighted CoSaMP, and multi-path weighted OMP.

[0142] Result analysis: Figure 2 The NMSE comparison curves of the above-mentioned five algorithms under the condition of low signal-to-noise ratio of-15 dB to-5 dB are shown, and the results are as follows: (1) the algorithm of the application performs best in the whole range. The algorithm proposed by the application shows the lowest NMSE value, verifying the robustness and anti-interference ability of the algorithm in a high-noise environment. (2) Compared with the weighted OMP with static sparsity control and the multi-path extended version, the application introduces a dynamic path verification mechanism and a sparsity adaptive update strategy, which effectively reduces the risk of misselected paths under low SNR conditions, thereby achieving better performance. (3) Although the improved weighted CoSaMP combines the path backtracking feature of the CoSaMP framework, it has certain performance in some medium SNR scenarios, but it is sensitive to noise in the low SNR area, and the overall NMSE performance is not as good as the verification result of the algorithm of the application. (4) The NMSE of the algorithm of the application decreases steadily as the SNR gradually increases, indicating that the estimation process has good numerical stability and generalization ability. It is proved that the overall trend of the method of the application is stable and has good convergence.

[0143] This embodiment further analyzes the influence of the number of training frames (denoted as M) on the normalized mean square error performance of each algorithm. As shown in Figure 3 The estimation accuracy change trend of different algorithms when the training frame number increases from 20 to 100 under the same SNR condition is shown. The results are as follows: the overall performance of the application is significantly improved as the training frame number increases: as the training frame number increases from a small value (such as M=20) to a larger value, the NMSE of all algorithms shows a downward trend. This is consistent with the theory of compressed sensing, that is, more observation samples can improve the recoverability of sparse channels. The algorithm of the application performs best under all training frame numbers. Figure 3 The important influence of the number of training frames on the performance of sparse channel estimation is verified, and the adaptability, high efficiency and leading estimation accuracy of the algorithm proposed by the application under different training sample conditions are particularly shown. The algorithm does not depend on fixed sparsity setting, and can automatically adjust the estimation process according to the observation data, which has higher practical application value.

[0144] In conclusion, the following conclusions can be drawn: under the condition of low signal-to-noise ratio (-15dB to -5dB), the NMSE performance of the method of the present application is superior to other comparative algorithms, and better robustness and anti-interference ability are shown. With the increase of the number of training frames (from 20 to 100), the method of the present application performs best under all numbers of training frames, showing good adaptability and estimation accuracy.

[0145] The above examples fully demonstrate the application effect of the method of the present application in the actual millimeter wave MIMO system, and verify the superior performance of the method of the present application in the complex channel environment.

[0146] Finally, although the present specification is described in terms of embodiments, not every embodiment contains only one independent technical solution, and the description manner of the specification is only for the sake of clarity, and the skilled person in the art should consider the specification as a whole, and the technical solutions in each embodiment can be properly combined to form other embodiments that can be understood by the skilled person.

[0147] The above is only the preferred specific implementation of the present application, but the protection scope of the present application is not limited to this, and any skilled person in the art can make equivalent replacement or change according to the technical solutions and the inventive concept of the present application within the technical range disclosed by the present application, which should be covered in the protection scope of the present application.

Claims

1. A method for millimeter wave MIMO sparse channel estimation based on dynamic path validation mechanism, characterized in that, Training and testing are based on real millimeter wave channel measurement data, and a dynamic path verification mechanism is introduced in each iteration; combined with a sparsity adaptive mechanism, the iteration termination is dynamically controlled according to the training statistical information, and the specific steps include the following: S1. Construct a channel observation model: based on the OFDM frame structure of the mixed analog-digital structure, the channel observation model of the millimeter wave MIMO system is established; the channel is modeled in the angle domain, and the measurement matrix is constructed; S2. Dynamic path verification weighted OMP algorithm: the input of the algorithm includes the received measurement vector, the measurement matrix, the dictionary matrix, and the sparsity upper limit, the path verification candidate number and the threshold parameter; the algorithm steps include initialization, calculation of the correlation of the current residual to all candidate paths, path candidate screening, dynamic path verification, path selection and support set update, termination condition judgment; Among them, the dynamic path verification mechanism is as follows: in each iteration, multiple candidate paths are selected for verification, and least square estimation and residual calculation are performed on each candidate path, and the optimal path is selected to update the support set; the specific steps are as follows: a) compute the current residual r (t) relevance to all candidate paths; b) select the top K paths with the largest relevance as the candidate set ; c) for each candidate path in the candidate set performing least squares estimation and residual calculation for each candidate path in the candidate set for each candidate path in the candidate set performing the following steps: 1) Set temporary support set: ; wherein, denotes the temporary support set, denotes the support set of the previous iteration; 2) Solve the least square problem to estimate the sparse coefficient: ； wherein, represents each candidate path sparse coefficient vector, represents a measurement matrix, is the received measurement vector corresponding to the th subcarrier. 3) Calculate the residual: ； wherein, denotes a residual vector of a candidate path c, is a measurement matrix, denotes a residual vector of the last iteration, denotes each candidate path sparse coefficient vector; d) Select the path with the smallest residual as the optimal path and update the support set; path selection and support set update include: Selecting the optimal path: ; and updating the support set ; current residual ; The adaptive termination mechanism is as follows: set the residual drop rate threshold and the absolute residual threshold as the double judgment conditions, and control the iteration number adaptively according to the channel quality condition, which includes: The termination condition judgment is as follows: If any of the following conditions is met, the iteration is terminated: a) The residual drop rate is too low: ； b) current residual absolute value too small: ; wherein, is a residual drop rate threshold, set to 0.01, is a convergence threshold, set to ; S3. The estimated vector is output and the channel is reconstructed based on the measured data in the verification.

2. The method of claim 1, wherein, The channel observation model in step S1 is as follows: The millimeter wave MIMO system adopts a hybrid analog-digital structure, and based on an OFDM frame structure, a received signal of a first subcarrier is modeled as follows: y = Hx + n (1)； wherein is the frequency domain channel matrix on the th subcarrier; , respectively represent the transmit and receive hybrid beamforming matrices in the th time slot of the training frame, the superscript denotes the conjugate transpose; denotes the pilot vector of the th training frame, with dimension , where is the number of data streams; denotes the Gaussian white noise; The frequency domain channel matrix is expressed as: (2)； In the above formulae, , respectively denote the dictionary matrices of the transmit and receive arrays, whose column vectors are the response vectors in the discrete angular domain directions. For the first angular domain sparse channel matrix, vectorize it and construct a measurement matrix , we have (3)； The measurement matrix Representation matrix The complex conjugate of represents the transpose of the matrix, represents the Kronecker product; For the The receiving measurement vector corresponding to the subcarriers is For the The angle domain sparse channel vector under subcarriers.

3. The method of claim 1, wherein, In step S2, the dynamic path verification weighted OMP algorithm is as follows: Input: Receiving a measurement vector , a measurement matrix , a dictionary matrix ; sparsity upper limit , number of path verification candidates ; Threshold parameter , ; Output: Sparse channel estimation vector , channel estimation matrix ; The specific algorithm steps include the following: S21. Initialization: initial residual , support set is empty set, iteration number ; S22. Calculate the correlation: Compute the current residual r (t) Relevance of all candidate paths: ； wherein, is the measurement matrix, denotes the residual vector of the last iteration, is the projection coefficient of all paths and the residual of the last iteration; S23. Path candidate screening: According to the magnitude of the path, the first path with the largest magnitude is selected to form a candidate set .

Images