Two-stage Beam Training Method for Ultra-large-scale Array Communication System

Abstract

The present invention discloses a two-stage beam training method for a very large-scale array communication system. In the first stage of beam training, each sub-array of a very large-scale multiple-input multiple-output system performs analog combining using a series of far-field steering vectors. In the second stage of beam training, for line-of-sight path channel estimation, first, a dedicated digital combiner is designed for each codeword in a preset codebook. Then, the digital processing unit uses the dedicated digital combiner to perform digital combining processing on the signals after analog combining in the first stage. Finally, the codeword in the preset codebook that can achieve the maximum combining power is output as the result of beam training. For multipath channel estimation, digital processing is performed on the signals after analog combining in the first stage of beam training to complete the reconstruction of the channel transmission matrix. On the premise of greatly reducing the training overhead, the present invention can approximate the performance of the existing hybrid field beam scanning scheme and complete high-precision channel estimation.

Description

Technical Field

[0001] The present invention belongs to the field of millimeter-wave wireless communication, and relates to a two-stage beam training method for a very large-scale array communication system. Background Art

[0002] Massive multiple-input multiple-output (MIMO) is a key technology for traditional fifth-generation wireless communication. By configuring a large number of antennas at the base station, the spectral efficiency can be significantly improved through beamforming technology. In future sixth-generation wireless communication, a very large-scale MIMO system with more antennas will be adopted to further improve the spectral efficiency.

[0003] Due to the large energy loss of radio frequency (RF) chains, it is unrealistic to configure a dedicated RF chain for each antenna in a very large-scale MIMO system. Therefore, hybrid precoding / merging is usually used in a very large-scale MIMO system, that is, a small number of RF chains are connected to a large number of antennas. According to the connection method between the RF chains and the antennas, hybrid precoding can generally be divided into two types: fully connected structure and partially connected structure. In the fully connected structure, each RF chain is individually connected to an antenna, and this scheme has low efficiency due to the high insertion loss of the circuit. In contrast, in the partially connected structure, each RF chain is only connected to a part of a separated sub-array, which is more practical in terms of circuit configuration and system performance.

[0004] Since a very large-scale MIMO system is configured with an extremely large number of antennas, the characteristics of this system are different from those of traditional massive MIMO systems. According to the distance between the radiation source and the very large-scale MIMO system, with the Rayleigh distance as the boundary, the radiation field can be divided into the near field and the far field. At a specific carrier frequency, the Rayleigh distance increases quadratically with the increase in the number of antennas. Therefore, the far-field assumption in traditional massive MIMO systems may not be applicable to the near-field model of a very large-scale MIMO system. The method for obtaining the channel state information adapted to the near-field channel of a very large-scale MIMO system has become a research hotspot.

[0005] A direct method for obtaining channel state information is hybrid field beam scanning, that is, exhausting all codewords in the hybrid codebook to obtain the best matching codeword. However, this scheme will consume a large amount of training overhead and reduce communication efficiency. Reference [1] proposed a P-SOMP algorithm to estimate the near-field channel. However, in the training process of this method, a directional beamforming with high beam gain is not formed, which will reduce the channel estimation performance in the case of low signal-to-noise ratio. (Reference [1]: M. Cui and L. Dai, “Channel estimation for extremely large-scale MIMO: Far-field or near-field?” IEEE Trans. Commun., vol. 70, no. 4, pp. 2663–2677, Jan. 2022.). Reference [2] proposed a codebook-based far-field exhaustive beam training method. Although this method can obtain high beamforming gain in the far field, it does not consider the near-field channel. (Reference [2]: A. Alkhateeb, G. Leus, and R. W. Heath, “Limited feedback hybrid precoding for multi-user millimeter wave systems,” IEEE Trans. Wireless Commun., vol. 14, no. 11, pp. 6481–6494, Nov. 2015.). Summary of the Invention

[0006] Object of the Invention: Aiming at the above problems, the present invention proposes a two-stage beam training method for an extremely large-scale array communication system. The method includes two stages: in the first stage of beam training, each sub-array of the extremely large-scale multiple-input multiple-output system uses a series of far-field steering vectors for analog combining. In the second stage of beam training, for the line-of-sight path channel estimation, first, a dedicated digital combiner is designed for each codeword in the preset codebook. Then, the digital processing unit uses the dedicated digital combiner to perform digital combining processing on the signals after analog combining in the first stage. Finally, the codeword in the preset codebook that can achieve the maximum combining power is output as the result of beam training. For the multipath channel estimation, the signals after analog combining in the first stage of beam training are digitally processed to complete the reconstruction of the channel transmission matrix. On the premise of greatly reducing the training overhead, the present invention can approximate the performance of the existing hybrid field beam scanning scheme and complete high-precision channel estimation.

[0007] Technical Solution: To achieve the object of the present invention, the technical solution adopted by the present invention is: a two-stage beam training method for an extremely large-scale array communication system, the method comprising the following steps:

[0008] (1) Set the basic parameters of the massive multiple-input multiple-output system;

[0009] (2) Construct the massive multiple-input multiple-output system between the base station and the user and the channel model in the system;

[0010] (3) Design a hybrid codebook for estimating the channel in step (2);

[0011] (4) Based on the hybrid codebook in step (3), for line-of-sight path channel estimation, to determine the steering vector that best fits the channel, perform two-stage beam training on the massive multiple-input multiple-output system;

[0012] (5) Based on the first-stage beam training in step (4), for multipath channel estimation, to reconstruct the channel, perform two-stage beam training on the massive multiple-input multiple-output system.

[0013] Preferably, in step (1), the method for setting the basic parameters of the massive multiple-input multiple-output system is as follows: Set in the uplink beam training scenario between a base station and a user, the antenna array at the base station side is a uniform linear array with a half-wavelength spacing, the number of antennas is N, and a partially connected hybrid combining structure is adopted, including analog combining and digital combining, and the number of radio frequency links is N RF , the antenna contains N RF non-overlapping sub-arrays; each sub-array has M = N / N RF antennas, and after analog combining, it is connected to a radio frequency link; all N RF radio frequency links are connected to a digital processing unit for digital combining; the user side uses a single antenna.

[0014] Preferably, in step (2), the method for constructing the massive multiple-input multiple-output system model is as follows:

[0015] (2.1) Construct the massive multiple-input multiple-output system model between the base station and the user

[0016] In the uplink beam training, the training symbol sent by the user side is x k , k = 1, 2,..., K, where K is the signal length, and the channel between the base station and the user is represented as h. Then the received signal after hybrid combining at the base station is expressed as:

[0017] y k = v k W k hx k + v k W k η

[0018] Among them, W k represents analog combining, and v kIndicates digital merging, η represents additive white Gaussian noise, and the noise satisfies represents a complex Gaussian distribution with mean μ and variance σ 2 ;

[0019] (2.2) Construct the channel model in the very large-scale multiple-input multiple-output system

[0020] Set up a multipath channel composed of a main path and multiple secondary paths between the user and the base station. The N antennas of the base station are placed along the y-axis of the Cartesian coordinate system, and the coordinates of the nth antenna are (0, δ n λ), where n = 1, 2, …, N, λ represents the wavelength, and the coordinates of the center of the tth subarray are (0, Δ t λ), where t = 1, 2, …, N RF , Δ t = [(2t - 1)M - N] / 4. The coordinates of the user are expressed as p1 = (r1cosθ1, r1sinθ1), where r1 represents the distance between the user and the origin of the coordinates, and θ1 ∈ [-π / 2, π / 2] represents the angle of the user relative to the positive half-axis of the x-axis. The coordinates of the scattering point in the lth path are expressed as p l = (r l cosθ l , r l sinθ l ), where l > 2, r l represents the distance between the user and the origin of the coordinates, and θ l ∈ [-π / 2, π / 2] represents the angle of the user relative to the positive half-axis of the x-axis. The distance between p l and the nth antenna is expressed as

[0021]

[0022] where, is the sine value of the angle, and Ω l ∈ [-1, 1]; The channel between the base station and the user is modeled as follows:

[0023]

[0024] where, L and g l represent the number of paths and the channel gain of the lth path respectively, and α(·) represents the channel steering vector, which is defined as:

[0025]

[0026] Generally, the Rayleigh distance is used to distinguish the near field and the far field. The Rayleigh distance is expressed as

[0027]

[0028] Among them, D = Nλ / 2 represents the antenna array aperture; when the distance between the radiation source and the base station exceeds Z, the wireless channel is defined as a far-field channel; otherwise, the wireless channel is defined as a near-field channel.

[0029] When the distance r l > 2D 2 / λ, the following approximation is used.

[0030]

[0031] Among them, the steering vector of the far-field channel is defined as β(N, Ω l ).

[0032] Preferably, in step (3), the method for designing the hybrid codebook for estimating the channel in step (2) is as follows:

[0033] (3.1) Let C h represent the hybrid codebook, C f represent the far-field codebook, and C n represent the near-field codebook;

[0034] (3.2) The nth codeword in the far-field codebook C f described in step (3.1) is expressed as

[0035]

[0036] (3.3) The codewords in the near-field codebook C n described in step (3.1) are designed through the following steps:

[0037] ① Divide the near field into N equal parts in the angular dimension and S unequal parts in the distance dimension;

[0038] ② The nth quantization angle is Θ n = (2n - 1 - N) / N;

[0039] ③ The sth quantization distance at the nth angle is

[0040] ④ The near-field codebook C n is expressed as

[0041]

[0042] Among them, [C n :,s = α(N, Θ n , d n,s );

[0043] ⑤ The hybrid codebook described in step (3.1) is denoted as

[0044] Preferably, in step (4), based on the hybrid codebook in step (3), for line-of-sight path channel estimation, a two-stage beam training is performed on the massive multiple-input multiple-output system, and the method is as follows:

[0045] (4.1) For each subarray, the common beam training DFT codebook is

[0046]

[0047] where, Φ m = (2m - 1 - M) / M, m = 1, 2, …, M;

[0048] (4.2) In the first stage of beam training, the user equipment sends training symbols to the base station for M time slots continuously, and the base station receives them sequentially. For the k-th beam training, the received signal without digital combining is denoted as

[0049]

[0050] where, the k-th analog combining is denoted as blkdig{·} represents the block diagonalization operation. Thus, the first stage of the two-stage beam training scheme is completed;

[0051] (4.3) The analog combining is designed in the first stage, and in the second stage, the digital combining v p , p = 1, 2, … NS + N, will be designed by testing the NS + N codewords covered by the hybrid codebook. Denote the p-th codeword in the hybrid codebook as

[0052]

[0053] where, represents taking the lower bound of the element value;

[0054] (4.4) For the p-th codeword, the sine value of the angle corresponding to the quantization position of this codeword relative to the center of the t-th subarray is

[0055]

[0056] where, Δ t = [(2t - 1)M - N] / 4;

[0057] (4.5) Set

[0058]

[0059] where

[0060] (4.6) Digital combination is expressed as

[0061]

[0062] wherein,

[0063] (4.7) The received signal in the first stage is digitally combined and then expressed as

[0064]

[0065] wherein,

[0066] (4.8) Compare all NS+N combined signals, select the quantization position corresponding to the signal with the maximum energy, and express it as

[0067]

[0068] (4.9) The finally selected codeword that best suits the channel is

[0069] Preferably, in step (5), based on the first-stage beam training in step (4), for multipath channel estimation, two-stage beam training is performed on the very large-scale multiple-input multiple-output system.

[0070] (5.1) On the basis of the first stage, express the signal after analog combination as

[0071]

[0072] Define

[0073]

[0074] wherein,

[0075]

[0076] (5.2) Initialize the residual as R 0 = Y, and the index set Υ 0 = φ, where φ represents the empty set;

[0077] (5.3) For the l-th path, where is a preset number of path traversals, usually The specific steps are as follows:

[0078] ① Calculate the correlation matrix Γ l = Ψ H R l-1 ;

[0079] ② Obtain the index p corresponding to the maximum modulus value in the correlation matrix * = argmax p [|Γ l p |, where p = 1, 2, …, NS + N, and |·| represents taking the modulus;

[0080] ③ Update the index set Υ l = γ l-1 ∪ p * , where ∪ represents the union operation;

[0081] ④ Update the orthogonal function where represents finding the pseudo-inverse;

[0082] ⑤ Update the residual

[0083] (5.4) Repeat step (5.4) until all paths are traversed and ended, and the channel matrix estimation value is

[0084] Advantageous effects: Compared with the prior art, the technical solution of the present invention has the following advantageous technical effects:

[0085] (1) This method can significantly reduce the training overhead compared with the existing hybrid field beam scanning scheme;

[0086] (2) This method can obtain comparable sum rate performance, approach the hybrid field beam scanning performance at full signal-to-noise ratio, be superior to the performance of the method in [1] at low signal-to-noise ratio, and be superior to the performance of the method in [2] in the near field.

[0087] (3) This method can obtain channel estimation results with comparable accuracy, which is superior to the estimation accuracy of the method in [1].

[0088] (4) The digital combination in this method can be calculated offline before beam training, greatly reducing the computational amount during the beam training process.

[0089] (5) This method can implement multiple digital combinations in parallel, greatly improving the operation efficiency. Brief description of the drawings

[0090] Figure 1 is a schematic diagram of the very large-scale multiple-input multiple-output system model used in the embodiments of the present invention;

[0091] Figure 2 is a schematic diagram of the channel model in the very large-scale multiple-input multiple-output system used in the embodiments of the present invention;

[0092] ​Figure 3 is the comparison of the spectral efficiency between the present invention and those in Document [1], Document [2] and the hybrid-field beam scanning method;

[0093] Figure 4 is the comparison of the beamforming gain between the present invention and those in Document [1], Document [2] and the hybrid-field beam scanning method;

[0094] Figure 5 is the comparison of the normalized mean square error between the channel estimated by the method in Document [1] and the actual channel of the present invention. Detailed implementation manners

[0095] The technical solutions of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0096] (1) The basic parameters of the very large-scale multiple-input multiple-output system used in the present invention are as follows:

[0097] In the uplink beam training scenario between a base station and a user, the antenna array at the base station side is a uniform linear array with a half-wavelength interval, the number of antennas is N, and a partial connection hybrid combining structure is adopted, including analog combining and digital combining. The number of radio frequency links is N RF , and the antenna includes N RF non-overlapping sub-arrays; each sub-array has M = N / N RF antennas, and after analog combining, it is connected to a radio frequency link; all N RF radio frequency links are connected to a digital processing unit for digital combining; the user side adopts a single antenna.

[0098] (2) The model description of the very large-scale multiple-input multiple-output system between the base station and the user of the present invention is as follows:

[0099] In the uplink beam training, the training symbol sent by the user side is x k , k = 1, 2,..., K, where K is the signal length, and the channel between the base station and the user is represented by h. Then the received signal after hybrid combining at the base station is expressed as:

[0100] y k = v k W k hx k + v k W k η

[0101] where W k represents analog combining, v k represents digital combining, η represents additive white Gaussian noise, and the noise satisfies represents a complex Gaussian distribution with a mean of μ and a variance of σ 2 ;

[0102] As shown Figure 2 in the figure, the channel model in the ultra - large - scale multiple - input multiple - output system of the present invention is described as follows:

[0103] A multipath channel composed of a main path and multiple secondary paths is set between the user and the base station. N antennas of the base station are placed along the y - axis of the Cartesian coordinate system, and the coordinate of the nth antenna is (0, δ n λ), where n = 1, 2, …, N, λ represents the wavelength, and the coordinate of the center of the tth sub - array is (0, Δ t λ), where t = 1, 2, …, N RF , Δ t = [(2t - 1)M - N] / 4. The coordinate of the user is expressed as p1 = (r1cosθ1, r1sinθ1), where r1 represents the distance between the user and the origin of the coordinate system, and θ1 ∈[-π / 2, π / 2] represents the angle of the user relative to the positive half - axis of the x - axis. The coordinate of the scattering point in the lth path is expressed as p l = (r l cosθ l , r l sinθ l ), where l>2, r l represents the distance between the user and the origin of the coordinate system, and θ l ∈[-π / 2, π / 2] represents the angle of the user relative to the positive half - axis of the x - axis. The distance between p l and the nth antenna is expressed as

[0104]

[0105] where, is the sine value of the angle, and Ω l ∈[-1, 1]; The channel between the base station and the user is modeled as follows:

[0106]

[0107] where, L and g l represent the number of paths and the channel gain of the lth path respectively, and α(·) represents the channel steering vector, which is defined as:

[0108]

[0109] Generally, the Rayleigh distance is used to distinguish the near - field and the far - field. The Rayleigh distance is expressed as

[0110]

[0111] Among them, D = Nλ / 2 represents the antenna array aperture; when the distance between the radiation source and the base station exceeds Z, the wireless channel is defined as a far-field channel; otherwise, the wireless channel is defined as a near-field channel.

[0112] When the distance r l > 2D 2 / λ, the following approximation is used.

[0113]

[0114] Among them, the steering vector of the far-field channel is defined as β(N, Ω l ).

[0115] (3) The design method of the hybrid codebook for estimating the channel in step (2) proposed by the present invention is described as follows:

[0116] (3.1) Let C h represent the hybrid codebook, C f represent the far-field codebook, and C n represent the near-field codebook.

[0117] (3.2) The nth codeword in the far-field codebook C f described in step (3.1) is expressed as

[0118]

[0119] (3.3) The codewords in the near-field codebook C n described in step (3.1) are designed through the following steps:

[0120] ① Divide the near field into N equal parts in the angular dimension and S unequal parts in the distance dimension.

[0121] ② The nth quantization angle is Θ n = (2n - 1 - N) / N.

[0122] ③ The sth quantization distance at the nth angle is

[0123] ④ The near-field codebook C n is expressed as

[0124]

[0125] Among them, [C n :,s = α(N, Θ n , d n,s ).

[0126] ⑤ The hybrid codebook described in step (3.1) is expressed as

[0127] (4) Based on the hybrid codebook in step (3), for line-of-sight path channel estimation, the two-stage beam training method for the very large-scale multiple-input multiple-output system proposed by the present invention is described as follows:

[0128] (4.1) For each subarray, the common beam training DFT codebook is

[0129]

[0130] where, Φ m = (2m - 1 - M) / M, m = 1, 2, …, M;

[0131] (4.2) In the first stage of beam training, the user terminal sends training symbols to the base station for M time slots continuously, and the base station receives them sequentially. For the k-th beam training, the received signal without digital combining is expressed as

[0132]

[0133] where, the k-th analog combining is expressed as blkdig{·} represents the block diagonalization operation. Thus, the first stage of the two-stage beam training scheme is completed;

[0134] (4.3) The first stage designs analog combining, and the second stage will design digital combining v p , p = 1, 2, … NS + N, and express the p-th codeword in the hybrid codebook as

[0135]

[0136] where, represents taking the lower bound of the element value;

[0137] (4.4) For the p-th codeword, the sine value of the angle corresponding to the quantization position of this codeword relative to the center of the t-th subarray is

[0138]

[0139] where, Δ t = [(2t - 1)M - N] / 4;

[0140] (4.5) Set

[0141]

[0142] where

[0143] (4.6) Digital combining is expressed as

[0144]

[0145] Among them,

[0146] (4.7) The received signal in the first stage is digitally combined and is expressed as

[0147]

[0148] Among them,

[0149] (4.8) Compare all NS + N combined signals, select the quantization position corresponding to the signal with the maximum energy, and express it as

[0150]

[0151] (4.9) The finally selected codeword that best suits the channel is

[0152] (5) Based on the first-stage beam training in step (4), for multipath channel estimation, perform two-stage beam training on the very large-scale multiple-input multiple-output system.

[0153] (5.1) On the basis of the first stage, express the signal after analog combination as

[0154]

[0155] Define

[0156]

[0157] Among them,

[0158]

[0159] (5.2) Initialize the residual as R 0 = Y, the index set γ 0 = φ, where φ represents the empty set;

[0160] (5.3) For the l-th path, where is a preset number of path traversals, usually The specific steps are as follows:

[0161] ① Calculate the correlation matrix Γ l = Ψ H R l-1 ;

[0162] ② Obtain the index p corresponding to the maximum modulus value in the correlation matrix * = argmax p |[Γ lp |, p = 1, 2, …, NS + N, where |·| represents taking the modulus;

[0163] ③ Update the index set γ l = γ l-1 ∪ p * , where ∪ represents the union operation;

[0164] ④ Update the orthogonal function where represents finding the pseudo-inverse;

[0165] ⑤ Update the residual

[0166] (5.4) Repeat step (5.4) until all paths are traversed and ended, obtaining the channel matrix estimate as

[0167] The present invention is further described below in conjunction with simulation conditions and results:

[0168] Consider a very large-scale multiple-input multiple-output system with N = 256 antennas. The entire antenna array is divided into N RF = 4 sub-arrays, each sub-array having M = 64 antennas, and the wavelength is set to λ = 0.003 m. The channel between the base station and the user consists of 1 line-of-sight path and 2 non-line-of-sight paths, and the total number of paths L = 3. The path gain beam gain conforms to the complex Gaussian distribution The channel angle Ω of the l-th path l conforms to a uniform distribution within. The distance quantization number of the hybrid codebook C h is S = 6. The pilot length in the method of [1] is set to 64.

[0169] (1) As Figure 3 shown, compare the spectral efficiency of the two-stage beam training method proposed by the present invention, the method of [1], the hybrid-field beam scanning scheme, and the method of [2]. The distance from the base station to the user or the scatter point conforms to a uniform distribution within [5 m, 10 m]. It can be seen from the figure that, compared with the other three methods, the hybrid-field beam scanning scheme can achieve higher performance because the hybrid-field beam scanning scheme exhausts the hybrid codebook C h ​All codewords in [reference], but the hybrid field beam scanning will use a much larger number of pilots than the other three schemes. The method in reference [1] has the worst performance under low signal-to-noise ratio conditions, such as -10 dB, because the received signal-to-noise ratio of the method in reference [1] is smaller than that of the other three methods, resulting in a significant reduction in the performance of this scheme under low signal-to-noise ratio conditions. The method in reference [2] has the worst performance under high signal-to-noise ratio conditions because the gain of the far-field steering vector decays in the near-field model. However, the two-stage beam training method proposed in the present invention can approach the performance of the hybrid field beam scanning scheme under low and high signal-to-noise ratio environments.

[0170] (2) As Figure 4 shown, we compared the beamforming gains of the two-stage beam training method proposed in the present invention, the method in reference [1], the hybrid field beam scanning scheme, and the method in reference [2]. The distance from the base station to the user or scatter point follows a uniform distribution within [5 m, r], where r varies from 10 m to 120 m. The signal-to-noise ratio is set to -5 dB. As can be seen from Figure 4 , the hybrid field beam scanning scheme has the highest beam gain at all distances. As the distance decreases, the far-field beam scanning scheme will experience severe beamforming gain attenuation because this scheme only considers the far-field channel. The other three schemes consider both the near-field and far-field channels, so they are robust in terms of distance. In addition, the two-stage beam training proposed in the present invention can approach the performance of the hybrid field beam scanning scheme, with only a slight loss in beamforming gain at close and far distances.

[0171] (3) As Figure 5 shown, we compared the normalized mean square error between the channel estimated by the two-stage beam training method proposed in the present invention and the method in reference [1] and the actual channel. Define the mean square error as where is the channel estimate value, H is the actual channel, and denotes the calculation of the two-norm. Set in the NLOS model, the number of paths L = 6. As can be seen from the figure, compared with the method in reference [1], the method in the present invention can obtain a higher-precision channel estimate because the present invention adopts a partially connected structure, and using multiple radio frequency chains can obtain higher-dimensional signals. By jointly digitally processing the signals on multiple radio frequency chains, higher estimation accuracy can be obtained at the same training cost.

[0172] (4) Compare the training overheads of different schemes. The training overheads of the hybrid field beam scanning scheme, the method in reference [2], the method in reference [1], and the two-stage beam training method proposed in the present invention are N(S + 1), N, N P and M respectively. For example, in Figure 3 and Figure 4Under the simulation parameter settings, the training pilots used by the hybrid-field beam scanning scheme, the method in [2], the method in [1], and the two-stage beam training method proposed in the present invention are 1792, 256, 64, and 64 respectively. Generally speaking, compared with the hybrid-field beam scanning scheme, the two-stage beam training method proposed in the present invention reduces the training overhead by 96.43%, while approaching the hybrid-field beam scanning scheme in terms of performance.

Claims

1. A two-stage beam training method for a very large scale array communication system, characterized in that The method includes the following steps: (1) Set the basic parameters of the very large-scale multiple-input multiple-output system; (2) Construct a very large-scale multiple-input multiple-output system between the base station and the user and the channel model in the system; (3) Design a hybrid codebook for estimating the channel in step (2); (4) Based on the hybrid codebook in step (3), for line-of-sight path channel estimation, to determine the steering vector that best suits the channel, perform two-stage beam training on the very large-scale multiple-input multiple-output system; (5) Based on the first-stage beam training in step (4), for multipath channel estimation, to reconstruct the channel, perform two-stage beam training on the very large-scale multiple-input multiple-output system; In step (1), the method for setting the basic parameters of the very large-scale multiple-input multiple-output system is as follows: In the uplink beam training scenario between a base station and a user, the antenna arrays at the base station side are all uniform linear arrays with a half-wavelength spacing, the number of antennas is N, and a partially-connected hybrid combining structure is adopted, including analog combining and digital combining. The number of radio frequency links is N RF , the antenna includes N RF non-overlapping sub-arrays; each sub-array has M = N / N RF antennas, and after analog combining, it is connected to a radio frequency link; all N RF radio frequency links are connected to a digital processing unit for digital combining; the user side adopts a single antenna; In step (2), the method for constructing the very large-scale multiple-input multiple-output system model is as follows: (2.1) Construct a very large-scale multiple-input multiple-output system model between the base station and the user In the uplink beam training, the training symbol sent by the user equipment is x k , k = 1, 2, ..., K, where K is the signal length, and the channel between the base station and the user is represented by h. Then, the received signal after hybrid combining at the base station is expressed as: y k = v k W k hx k + v k W k η Among them, W k represents analog combination, v k represents digital combination, η represents additive white Gaussian noise, and the noise satisfies represents a complex Gaussian distribution with a mean of μ and a variance of σ 2 ; (2.2) Construct the channel model in the very large-scale multiple-input multiple-output system A multipath channel composed of a main path and multiple sub-paths is set up between the user and the base station. N antennas of the base station are placed along the y-axis of the Cartesian coordinate system. The coordinate of the nth antenna is (0, δ n λ), where n = 1, 2,..., N, λ represents the wavelength. The coordinate of the center of the tth sub-array is (0, △tλ), where t = 1, 2,..., N RF , △ t = [(2t - 1)M - N] / 4. The coordinate of the user is expressed as p1 = (r1cosθ1, r1sinθ1), where r1 represents the distance between the user and the origin of the coordinate system, and θ1 ∈ [-π / 2, π / 2] represents the angle of the user relative to the positive half-axis of the x-axis. The coordinate of the scattering point in the lth path is expressed as p l = (r l cosθ l , r l sinθ l ), where l > 2, r l represents the distance between the user and the origin of the coordinate system, and θ l ∈ [-π / 2, π / 2] represents the angle of the user relative to the positive half-axis of the x-axis. The distance between p l and the nth antenna is expressed as wherein, is the sine value of the angle, and Ωl ∈ [-1, 1]; the channel between the base station and the user is modeled as follows: where L and g l represent the number of paths and the channel gain of the l-th path, respectively, and α(·) represents the channel steering vector, which is defined as: Usually, the Rayleigh distance is used to distinguish the near field and the far field, and the Rayleigh distance is expressed as where D = Nλ / 2 represents the antenna array aperture; when the distance between the radiation source and the base station exceeds Z, the wireless channel is defined as a far-field channel; otherwise, the wireless channel is defined as a near-field channel; When the distance r l > 2D 2 / λ, the following approximation is used, Among them, the steering vector of the far-field channel is defined as β(N, Ω l ); In step (3), the method for designing the hybrid codebook for estimating the channel in step (2) is as follows: (3.1) Let C h represent the hybrid codebook, C f represent the far - field codebook, C n represent the near - field codebook; (3.2) The n-th codeword in the far-field codebook C described in step (3.1) f is denoted as (3.3) The codewords in the near-field codebook C described in step (3.1) n are designed through the following steps: ① Divide the near field into N equal parts in the angular dimension and S unequal parts in the distance dimension; ② The nth quantization angle is Θ n =(2n - 1 - N) / N; ③The s-th quantization distance at the n-th angle is ④ Near-field codebook C n Denoted as where [C n ：，s = α(N, Θ n , d n，s );​ ⑤ The hybrid codebook described in step (3.1) is denoted as In step (4), based on the hybrid codebook in step (3), for line-of-sight path channel estimation, perform two-stage beam training on the very large-scale multiple-input multiple-output system, and the method is as follows: (4.1) For each subarray, the commonly used beam training DFT codebook is where, Φ m = (2m - 1 - M) / M, m = 1, 2, …, M; (4.2) In the first stage of beam training, the user terminal sends training symbols to the base station for M time slots, and the base station receives them sequentially. For the kth beam training, the received signal without digital combining is expressed as Among them, the k-th simulated combination is denoted as blkdig{·} represents the block diagonalization operation. So far, the first stage of the two-stage beam training scheme is completed. (4.3) In the first stage, analog combining is designed. In the second stage, digital combining v will be designed by testing NS+N codewords covered by the hybrid codebook p , where p = 1, 2, … NS+N, and the p-th codeword in the hybrid codebook is denoted as Among them, means taking the lower bound of the element value; (4.4) For the pth codeword, the sine value of the angle corresponding to the quantization position of this codeword relative to the center of the tth subarray is Among them, △ t = [(2t - 1)M - N] / 4; (4.5) Set Among them (4.6) Digital combining is expressed as Among them, (4.7) The received signal in the first stage is digitally combined and then expressed as Among them, (4.8) Compare all NS + N combined signals, and select the quantization position corresponding to the signal with the maximum energy, which is expressed as (4.9) The finally selected codeword for the most suitable channel is In step (5), based on the first-stage beam training in step (4), for multipath channel estimation, perform two-stage beam training on the very large-scale multiple-input multiple-output system, and the method is as follows: (5.1) On the basis of the first stage, express the signal after analog combining as Define where (5.2) Initialize the residual as R0 = Y, and the index set Y0 = φ, where φ represents the empty set; (5.3) For the l-th path, where is a preset number of path traversals, usually The specific steps are as follows: ① Calculate the correlation matrix Γ l = Ψ H R l-1 ; ② Obtain the index corresponding to the maximum modulus value in the correlation matrix where |·| represents taking the modulus; ③Update the index set Υ l = Υ l-1 ∪ p * , where U represents the union operation; ④ Update orthogonal functions wherein denotes finding the pseudo-inverse; ⑤ Update the residual (5.4) Repeat step (5.4) until all paths are traversed and ended, and the estimated value of the channel matrix is

Images