Parallel dynamic sub-array architecture adopting combined phase shifters and beam forming method

By adopting a flat dynamic subarray architecture and codebook design with a combined phase shifter in a millimeter wave large-scale multi-input and output system, the problems of system complexity and channel estimation overhead are solved, and beamforming with high resolution low gate lobes is realized , reducing system complexity.

CN119010973APending Publication Date: 2024-11-22HARBIN INSTITUTE OF TECHNOLOGY (SHENZHEN) (INSTITUTE OF SCIENCE AND TECHNOLOGY INNOVATION HARBIN INSTITUTE OF TECHNOLOGY SHENZHEN)
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411247638.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-05
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

When traditional millimeter wave large-scale multi-input output systems realize high resolution beamforming, the system complexity and channel estimation overhead are high, and the codebook-based beamforming scheme has problems with high user interference and system complexity.

Method used

A parallel dynamic subarray architecture with combined phase shifters is used to connect high-resolution and low-resolution phase shifters to each RF link through a PUPA architecture composed of two parallel UPAs, and utilizes exhaustive search and cuckoo search The algorithm obtains the best phase shifter selection strategy and phase distribution, and designs a codebook with high resolution and low gate lobes.

Benefits of technology

While maintaining high-resolution main lobes, it realizes flexible subarray layout, overcomes the limitations of simulated precoding hardware, reduces system complexity, and realizes beamforming of high-resolution low-gate lobes.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119010973A_ABST
    Figure CN119010973A_ABST
Patent Text Reader

Abstract

The invention discloses a parallel dynamic sub-array architecture adopting a combined phase shifter and a beam forming method, in the parallel dynamic sub-array architecture, each radio frequency link is dynamically connected to a sub-array consisting of two antenna subsets of a parallel array, and flexible sub-array layout is realized while a high-resolution main lobe is kept; a combined phase shifter architecture is adopted, so that the hardware limitation of an analog pre-coding constant amplitude can be overcome, and better beam performance can be obtained; a phase shifter activation strategy is obtained by adopting an exhaustion search scheme, and phase distribution is searched by adopting a cuckoo search algorithm, so that a final code word is obtained, and a high-resolution low-grating lobe beam can be realized on the basis of a low-complexity architecture.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a parallel dynamic subarray architecture using a combined phase shifter and a millimeter wave large-scale array beamforming method based on the parallel dynamic subarray. Background Art

[0002] Traditional millimeter-wave massive multi-input and multi-output systems use uniform arrays with half-wavelength spacing. They obtain high-resolution beams by increasing the number of antennas. However, configuring a large number of antennas makes the system more complex.

[0003] Specifically, there are mainly two beamforming (precoding) schemes in the current millimeter wave multiple-input and multiple-output systems: adaptive beamforming and codebook-based beamforming.

[0004] Adaptive beamforming technology: In time division duplex mode, the base station estimates the uplink channel information based on the received signal, and calculates the downlink channel information based on the reciprocity of the uplink and downlink channels. Finally, the base station obtains the precoding matrix based on the acquired downlink channel information. However, in frequency division duplex mode, the uplink and downlink channels are at different frequencies, and the channel gain and phase lag are different. Therefore, it is necessary to estimate the downlink channel at the user end and feed it back to the base station to obtain the final precoding matrix.

[0005] Codebook-based solution: This solution designs a codebook that contains multiple codewords, each corresponding to a physical direction. The system performs beam training on the codebook by sending pilot signals, and the receiving end finally matches the optimal codeword that adapts to the current channel environment based on the obtained pilot signal information.

[0006] Common adaptive beamforming schemes have excellent performance, but with the configuration of large-scale antenna arrays, it is difficult to obtain channel state information in real time and accurately in actual systems, especially in mobile environments, and it will generate high channel estimation (channel feedback) overhead. For this reason, codebook-based beamforming is becoming a promising alternative in the industry.

[0007] On the other hand, for codebook-based beamforming schemes, related work proposed a parallel co-prime array (PCA), which consists of two co-prime arrays to produce high-resolution beams. However, parallel co-prime arrays will produce higher grating lobes, resulting in greater user interference in the system. In addition, there is also work that proposes a parallel dual UPA (Parallel Uniform Plane Array, PUPA) architecture and designs the corresponding codebook. PUPA consists of two UPAs that are aligned in parallel, which can suppress unwanted beams and increase the gain of the desired main lobe. However, compared with PCA, PUPA needs to be configured with more antennas to ensure the same beam resolution, which ultimately leads to considerable system complexity problems. Summary of the invention

[0008] The main purpose of the present invention is to propose a parallel dynamic subarray architecture using a combined phase shifter to generate a high-resolution low grating lobe beam and a millimeter wave large-scale array beamforming method based on a parallel dynamic subarray for a millimeter wave multiple-input multiple-output communication system.

[0009] To achieve the above object, the present invention provides a parallel dynamic subarray architecture using a combined phase shifter, characterized in that the base station end of the parallel dynamic subarray architecture is equipped with a PUPA composed of two parallel UPAs to simultaneously provide services for K single-antenna users, and each UPA array has N = N x ×N z Antennas, N x and N z are the number of antennas along the x-axis and z-axis respectively. The number of users served by the base station at the same time is affected by the number of RF links N RF The restriction is K≤N RF Each RF link is connected via a high-resolution phase shifter and 2N low-resolution phase shifters, the high-resolution phase shifter and the low-resolution phase shifter are controlled by 2N+1 switches, and the number of phase shifters activated in the two UPAs is and The constraints are

[0010] The hybrid beamformer at the base station includes a digital beamformer at the baseband: N RF ×K-dimensional matrix F BB , and analog beamformer in the RF domain: 2N×N RF Dimensional matrix F RF , f k is a 2N×1 dimensional matrix, f kis the analog beamforming vector corresponding to RF link k, k = 1, ..., K;

[0011] The parallel dynamic subarray architecture is constructed by setting ||F RF F BB || 2 =K to normalize the power gain of the hybrid beamformer.

[0012] To achieve the above object, the present invention further proposes a millimeter wave large-scale array beamforming method based on parallel dynamic subarrays, the method being applied to the parallel dynamic subarray architecture using combined phase shifters as described above, the method comprising the following steps:

[0013] The base station sends a training signal s to K users and obtains the beam vector f through training. k , the base station can obtain an effective channel between it and user k.

[0014] A further technical solution of the present invention is that the base station sends a training signal s to K users, and obtains a beam vector f through training. k , the steps by which the base station can obtain an effective channel between itself and user k include:

[0015] The base station sends a training signal s to K users, where s k =[0, ..., s k , ...0] T represents the symbol vector sent to user k, (.) T is the vector transpose. In addition P k represents power and is equal to all users, then the signal received by user k is expressed as

[0016]

[0017] in, is Gaussian noise with mean 0 and variance h k represents the channel information between the base station and user k, (.) H It means to find the conjugate of the matrix, s u Represents the symbol sent by user u;

[0018] Assume that a k The geometric channel model of independent propagation paths, the channel vector h between the base station and user k k It can be defined as:

[0019]

[0020] where αkl ,θ kl ,φ kl are defined as the gain, elevation angle, and azimuth angle of the lth path respectively; due to the significant path loss caused by the millimeter wave band, the number of transmitting antennas 2N is much higher than the number of resolvable paths L k , that is, 2N>>L k ;function represents the antenna array steering vector, and a(θ, φ) is a UPA steering vector and can be expressed as:

[0021]

[0022] in, d represents the distance between antennas, y represents the spacing between two UPAs, λ is the wavelength, N=N x ×N z represents the number of antennas in a UPA, e is the natural logarithm base, j is the imaginary unit, (.) T is the vector transpose;

[0023] The beam vector fk is obtained through beam training, and the base station obtains the effective channel between it and user k as When zero-forcing precoding is used, the digital baseband precoder is in is the combination of the effective channels of all users; then, the signal to interference and noise ratio at user k is expressed as:

[0024]

[0025] P k and P u denote the transmit power serving users k and u respectively, represents the channel information h between the base station and user k k The conjugate of RF represents analog precoding, [F BB ] :,k Indicates digital precoding F BB The kth column vector of represents the variance of Gaussian noise. The sum rate is expressed as:

[0026]

[0027] Among them, K is the total number of users, SINR k is the signal to interference and noise ratio corresponding to user k, and log2(.) represents the logarithm with base 2.

[0028] A further technical solution of the present invention is that the base station sends a training signal s to K users, obtains a beam vector fk through training, and the base station can obtain an effective channel between itself and user k, including:

[0029] The optimal phase shifter selection strategy T is found by exhaustive search, and the phase distribution of the phase shifter is obtained by using the cuckoo search algorithm.

[0030] A further technical solution of the present invention is that the step of using exhaustive search to find the best phase shifter selection strategy T and using the cuckoo search algorithm to obtain the phase distribution of the phase shifter comprises:

[0031] Obtain the phase shifter selection strategy T through exhaustive search;

[0032] The phase is updated through a global search mechanism;

[0033] The phase is updated through a local search mechanism;

[0034] Obtain final phase distribution information;

[0035] Output codeword.

[0036] A further technical solution of the present invention is that, when considering the combined phase shifter architecture, the direction (θ p ,φ q ) corresponds to the code word f(θ p ,φ q ) by the high-resolution part f h and the low resolution part f l composition:

[0037] f(θ p ,φ q ) = f l (θ p ,φ q )+f h (θ p ,φ q ),

[0038] in, represents the low-resolution part of the codeword, where represents the phase of the i-th low-resolution phase shifter, N represents the number of antennas in a UPA, N RF represents the number of radio frequency chains, e is the base of natural logarithm, j is the imaginary unit, (.) T is the vector transpose. represents the high-resolution part of the codebook, where represents the phase of the high phase shifter, e is the base of the natural logarithm, j is the imaginary unit, (.) T is the vector transpose. B∈{k l , k h}, and They are respectively h Hardware-limited set of phase shifters with resolution k l Hardware-limited set of phase shifter phases.

[0039] A further technical solution of the present invention is that, in order to express the phase shifter selection strategy, T is defined as the activated phase shifter state expressed as:

[0040] T=diag(t1,t2,…,t 2N ), t i =0or 1,i=1,2,…,2N,

[0041] Among them, t i =0 means the phase shifter is not activated, otherwise the phase shifter is activated, diag(.) represents the diagonal element, based on the codeword f(θ p ,φ q ) and matrix T, a 90×180-dimensional matrix G is used to represent the discrete beam gain distribution:

[0042]

[0043] i=1,2,…,90, j=1,2,…,180,

[0044] in Indicates the proper amount of guidance of the PUPA array, (.) H represents conjugate. For the matrix G, select all beam peaks that satisfy the condition G[i, j]>G[i±1, j±1], that is, the gains of all lobes. We define these peak sets as {g1, g2, ..., g M}, where all peaks are sorted in descending order, i.e. g m-1 ≥g m , m = 1, ..., M, the mth peak g m The index in the matrix G is [i m , j m ];

[0045] The ratio of the maximum grating lobe gain to the target direction gain is η(θ p ,φ q ) is expressed as

[0046]

[0047] in Indicates the direction corresponding to the maximum peak g1 Approximate to (θ p ,φq ), δ θ (δ φ ) is the direction edge error predefined by the direction angle, that is [i p , j q ] is defined as the target direction (θ p ,φ q ) in the matrix G; the precision of g1 is determined by δ θ and δ φ When the direction of the maximum peak Away from the target direction (θ p ,φ q ), the target direction gain is G[i p , j q ], the maximum grating lobe gain is g1.

[0048] A further technical solution of the present invention is that the step of using exhaustive search to find the best phase shifter selection strategy T and using the cuckoo search algorithm to obtain the phase distribution of the phase shifter comprises:

[0049] Step S1, for the phase shifter selection strategy problem, construct a single phase shifter structure and use the steering vector as a beamforming vector, deactivating the high-resolution phase shifter, h = 0. Therefore, the main lobe gain g1 in the target direction is An exhaustive search method is used to obtain the optimal activation strategy T to minimize the maximum grating lobe value g2;

[0050] Step S2, YesN n ×(N t +1)-dimensional matrix, representing the random initial value of the solution, represents the total number of activated phase shifters in the two UPAs. First, a global search mechanism is used to update the solution:

[0051]

[0052] m=1, ..., N n , n=1,...,N t +1, γ g Step size scale factor, ω t YesN n ×1-dimensional matrix, [ω t ] m represents the inertia weight of the mth solution in the tth iteration, S t YesN n ×(N t +1)-dimensional matrix, [S t ] m,n represents the random step size of the mth solution in the tth iteration;

[0053] Among them, the inertia weight of the mth solution at the tth iteration is [ω t ] m , which depends on the number of times the objective function value in the historical iteration is equal to the objective function value of the t-1 iteration

[0054]

[0055] Where exp(.) represents e (.) , e is the base of natural logarithm, represents t-1 iterations N n The objective function value, (.) T is the vector transpose, YesN n The mean of the target values, ∑(.) represents the summation, and ρ is the step index;

[0056] Using the matrix S t As the random step size in t iterations, its elements follow the Lévy flight mechanism:

[0057]

[0058] where β is the distribution factor of the Levy flight, It conforms to the standard normal distribution, σ v =1, Γ(.) is the gamma function;

[0059] Step S3, updating the solution obtained by the global search mechanism, wherein the solution obtained by the global search mechanism is limited to [-π, π],

[0060]

[0061] The relevant objective function is Furthermore, the objective function value after local search in the t-1 iteration is Compare and select the better solution:

[0062]

[0063] Step 4: Further, the algorithm performs a local search mechanism:

[0064]

[0065] in, represents the inner product, N n ×(N t+1)-dimensional matrix, whose elements are randomly selected in the range [0,1]; m1,m2 is N n dimensional vector, is the Heaviside function:

[0066]

[0067] m=1, ..., N n , n=1,...,N t +1, in the local random exploration mechanism, only the random p of the solution a Some parts have been updated to promote the convergence of the algorithm;

[0068] Step S5, repeat step S3 to obtain the corresponding and

[0069] Step S6, repeating steps S2 to S5 until the number of iterations is greater than a predefined maximum number of iterations;

[0070] Step S7, obtaining in, index[.] represents the index symbol;

[0071] Step S8, selecting a phase that satisfies the hardware constraints in a near-optimal manner:

[0072]

[0073] Among them, |.| means to find the absolute value, [.] n represents the nth element of a vector, is to satisfy the resolution k h Hardware-limited phaser phase sets, is to satisfy the resolution k l Hardware-limited phaser phase set;

[0074] Step 9, obtain the final codeword f(θ p ,φ q ), f(θ p ,φ q ) corresponds to the element of the nth activated phase shifter is

[0075] The beneficial effects of the parallel dynamic subarray architecture using combined phase shifters and the millimeter wave large-scale array beamforming method based on parallel dynamic subarrays in the present invention are:

[0076] The present invention can achieve flexible sub-array layout while maintaining high-resolution main lobe. The combined phase shifter architecture can overcome the hardware limitation of analog precoding constant amplitude. An exhaustive search scheme is used to obtain the phase shifter activation strategy, and a cuckoo search algorithm is used to search the phase distribution to obtain the final codeword. A high-resolution and low-grating-lobe beam can be achieved on the basis of a low-complexity architecture. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0078] Figure 1 It is a schematic diagram of the structure of PUPA in the parallel dynamic subarray architecture using combined phase shifters of the present invention;

[0079] Figure 2 It is a schematic diagram of the structure of the millimeter wave large-scale multi-input and output system configured with PUPA of the present invention;

[0080] Figure 3 It is a flow chart of the millimeter wave large-scale array beamforming method based on parallel dynamic subarrays of the present invention.

[0081] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0082] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0083] In order to solve the beamforming problem of millimeter-wave large-scale multi-input multi-output systems, the present invention proposes a new parallel dynamic subarray architecture, in which each RF link is dynamically connected to a subarray consisting of two parallel array antenna subsets, achieving flexible subarray layout while maintaining high-resolution main lobes. Then, in order to obtain better beam performance, the present invention adopts a combined phase shifter architecture to overcome the hardware limitations of analog precoding constant amplitude. In this structure, each RF chain is connected to a high-resolution phase shifter, and each antenna is connected in parallel to a low-resolution phase shifter. Finally, in order to design a high-resolution low-grating lobe beam, the present invention proposes a codebook design scheme. The scheme adopts an exhaustive search scheme to obtain the phase shifter activation strategy, and uses a cuckoo search algorithm to search the phase distribution to obtain the final codeword. Compared with the existing codebook design scheme, the present invention can achieve a high-resolution low-grating lobe beam on the basis of a low-complexity architecture.

[0084] Specifically, the present invention aims at the beamforming problem of a millimeter wave large-scale multiple-input multiple-output system, and proposes a parallel dynamic subarray architecture using a combined phase shifter, specifically a millimeter wave large-scale multiple-input multiple-output architecture based on a parallel dynamic subarray architecture, such as Figure 1 and Figure 2 shown.

[0085] The present invention takes into account the downlink transmission of a large-scale multi-input and multi-output system. The base station of the parallel dynamic subarray architecture is equipped with a PUPA composed of two parallel UPAs to provide services for K single-antenna users at the same time. Each UPA array has N=N x ×N z Antennas, N x and N z are the number of antennas along the x-axis and z-axis respectively. The number of users served by the base station at the same time is affected by the number of RF links N RF The restriction is K≤N RF For simplicity, the present invention assumes that K = N RF Each RF link is connected via a high-resolution phase shifter and 2N low-resolution phase shifters, which are controlled by 2N+1 switches. The number of phase shifters activated in the two UPAs is and The constraints are

[0086] The hybrid beamformer at the base station includes a digital beamformer at the baseband: N RF ×K-dimensional matrix F BB , and analog beamformer in the RF domain: 2N×N RF Dimensional matrix F RF , f k is a 2N×1 dimensional matrix, f k is the analog beamforming vector corresponding to RF link k, k = 1, ..., K;

[0087] The parallel dynamic subarray architecture of the present invention is configured by setting ||F RF F BB || 2 = K to normalize the power gain of the hybrid beamformer to ensure that the hybrid beamformer does not provide any power gain.

[0088] Please refer to Figure 3 To achieve the above object, the present invention further proposes a millimeter wave large-scale array beamforming method based on parallel dynamic subarrays, the method being applied to the parallel dynamic subarray architecture using combined phase shifters as described above, the method comprising:

[0089] The base station sends a training signal s to K users and obtains the beam vector f through training. k , the base station can obtain an effective channel between it and user k.

[0090] Specifically, the base station sends a training signal s to K users, where represents the symbol vector sent to user k, (.) T is the vector transpose. In addition P k represents power and is equal to all users, then the signal received by user k is expressed as:

[0091]

[0092] in, is Gaussian noise with mean 0 and variance h k represents the channel information between the base station and user k, (.) H It means to find the conjugate of the matrix, s u represents the symbol sent by user u. The first term is useful information, and the second term is the interference caused by other users' data to user k.

[0093] According to the widely used Saleh-Valenzuela model, considering the finite scattering characteristics in the mmWave channel, it is assumed that a k The geometric channel model of independent propagation paths, the channel vector h between the base station and user k k It can be defined as:

[0094]

[0095] where α kl ,θ kl ,φ kl are defined as the gain, elevation angle, and azimuth angle of the lth path, respectively. Due to the significant path loss caused by the millimeter wave band, the number of transmitting antennas 2N is much higher than the number of resolvable paths L k , that is, 2N>>L k .function represents the antenna array steering vector, and a(θ, φ) is a UPA steering vector and can be expressed as:

[0096]

[0097] in, d represents the distance between antennas, y represents the spacing between two UPAs, λ is the wavelength, N=N x ×N z represents the number of antennas in a UPA, e is the natural logarithm base, j is the imaginary unit, (.) T is the vector transpose.

[0098] The beam vector f is obtained through beam training k , the effective channel between the base station and user k is When zero-forcing precoding is used, the digital baseband precoder is in is the combination of the effective channels of all users. Then, the signal to interference and noise ratio at user k is expressed as:

[0099]

[0100] P k and P u denote the transmit power serving users k and u respectively, represents the channel information h between the base station and user k k The conjugate of RF represents analog precoding, [F BB ] :,k Indicates digital precoding F BB The kth column vector of represents the variance of Gaussian noise. The sum rate is expressed as:

[0101]

[0102] Among them, K is the total number of users, SINR k is the signal to interference and noise ratio corresponding to user k, and log2(.) represents the logarithm with base 2.

[0103] It should be noted that the base station adopts a partially connected structure inside each UPA, where each RF link is connected to two parallel subarrays through a phase shifter, and all phase shifters of each subarray are controlled by an ON-OFF switch to activate their state. In principle, after some phase shifters are deactivated, the achievable spectral efficiency will be reduced due to the wide beam and greater interference. We use a bisection method to design a phase shifter adaptive codebook for beam matching, which minimizes the number of activated phase shifters under a given spectral efficiency constraint.

[0104] The optimization process for the number of activated phase shifters can be expressed as follows: the base station first selects a beam vector from a codebook where all phase shifters are activated. The user then calculates the corresponding spectral efficiency through beam matching, compares it with its spectral efficiency threshold, and feeds it back to the base station. Based on the received spectral efficiency feedback, the base station updates the number of activated phase shifters using a binary search method and selects a beam vector from the corresponding codebook. This process continues to run, but always ensures that the current spectral efficiency value of each user is greater than its threshold (i.e., the minimum required spectral efficiency). In any case where the spectral efficiency is lower than the threshold, the base station will increase the number of activated phase shifters.

[0105] In the process of optimizing the activated phase shifters, the key issue is how to improve the beam performance when updating the number of activated phase shifters of each RF chain in each round. Therefore, the present invention focuses on the design of the phase shifter activation strategy and the corresponding codebook after each update of the number of activated phase shifters.

[0106] Furthermore, the base station sends a training signal s to K users, and obtains a beam vector fk through training. The base station can obtain an effective channel between itself and user k, which includes:

[0107] The optimal phase shifter selection strategy T is found by exhaustive search, and the phase distribution of the phase shifter is obtained by using the cuckoo search algorithm.

[0108] Specifically, the step of using exhaustive search to find the best phase shifter selection strategy T and using the cuckoo search algorithm to obtain the phase distribution of the phase shifter includes:

[0109] Obtain the phase shifter selection strategy T through exhaustive search;

[0110] The phase is updated through a global search mechanism;

[0111] The phase is updated through a local search mechanism;

[0112] Obtain final phase distribution information;

[0113] Output codeword.

[0114] Among them, when considering the combined phase shifter architecture, the direction (θ p ,φq ) corresponds to the code word f(θ p ,φ q ) consists of the high-resolution part, h and the low resolution part f l composition:

[0115] f(θ p ,φ q ) = f l (θ p ,φ q )+f h (θ p ,φ q ).

[0116] in, represents the low-resolution part of the codeword, where represents the phase of the i-th low-resolution phase shifter, N represents the number of antennas in a UPA, N RF represents the number of radio frequency chains, e is the base of natural logarithm, j is the imaginary unit, (.) T is the vector transpose. represents the high-resolution part of the codebook, where represents the phase of the high phase shifter, e is the base of the natural logarithm, j is the imaginary unit, (.) T is the vector transpose. B∈{k l , k h}, and They are respectively h Hardware-limited set of phase shifters with resolution k l Hardware-limited set of phase shifter phases.

[0117] For the cuckoo search-based codebook design, in order to express the phase shifter selection strategy, T is defined as the activated phase shifter state expressed as:

[0118] T=diag(t1,t2,…,t 2N ), t i =0or 1,i=1,2,…,2N,

[0119] Among them, t i =0 means the phase shifter is not activated, otherwise the phase shifter is activated, diag(.) represents the diagonal element, based on the codeword f(θ p ,φ q ) and matrix T, a 90×180-dimensional matrix G is used to represent the discrete beam gain distribution:

[0120]

[0121] i=1,2,…,90, j=1,2,…,180,

[0122] in Indicates the proper amount of guidance of the PUPA array, (.) H represents conjugate. For the matrix G, select all beam peaks that satisfy the condition G[i, j]>G[i±1, j±1], that is, the gains of all lobes. We define these peak sets as {g1, g2, ..., g M}, where all peaks are sorted in descending order, i.e. g m-1 ≥g m , m = 1, ..., M, the mth peak g m The index in the matrix G is [i m , j m ].

[0123] The present invention calculates the ratio η(θ p ,φ q ) is expressed as:

[0124]

[0125] in Indicates the direction corresponding to the maximum peak g1 Approximate to (θ p ,φ q ), δ θ (δ φ ) is the direction edge error predefined by the direction angle, that is [i p , j q ] is defined as the target direction (θ p ,φ q ) is the index in the matrix G. This optimization problem is calculated based on the distribution of beam peaks, and the accuracy of g1 is determined by δ θ and δ φ When the direction of the maximum peak Away from the target direction (θ p ,φ q ), the target direction gain is G[i p , j q ], the maximum grating lobe gain is g1.

[0126] In order to generate a high-resolution beam with low grating lobes, the codebook problem can be divided into two sub-problems: the phase shifter selection strategy T and the phase distribution of the combined phase shifter. First, an exhaustive search is used to find the best switching strategy, and then the cuckoo search algorithm is used to obtain the phase distribution of the combined phase shifter. The specific process is as follows:

[0127] Step S1, for the phase shifter selection strategy problem, construct a single phase shifter structure and use the steering vector As the beamforming vector, deactivate the high-resolution phase shifter f h = 0. Therefore, the main lobe gain g1 in the target direction is The present invention adopts an exhaustive search method to obtain the optimal activation strategy T to minimize the maximum grating lobe value g2.

[0128] In order to overcome the hardware limitation of the constant amplitude of analog precoding and obtain better beam performance, a high-resolution phase shifter is used in each RF chain according to the acquired phase shifter activation strategy, and the cuckoo search algorithm is used to obtain the phase distribution of the combined phase shifter so that the objective function value η(θ p ,φ q ) is the smallest. The cuckoo search algorithm adopts a balanced combination of global random exploration mechanism and local random exploration mechanism to update the solution.

[0129] Step S2, YesN n ×(N t +1)-dimensional matrix, representing the random initial value of the solution, represents the total number of activated phase shifters in the two UPAs. First, a global search mechanism is used to update the solution:

[0130]

[0131] m=1, ..., N n , n=1,...,N t +1, γ g Step size scale factor, ω t YesN n ×1-dimensional matrix, [ω t ] m represents the inertia weight of the mth solution in the tth iteration, S t YesN n ×(N t +1)-dimensional matrix, [S t ] m,n represents the random step size of the m-th solution in the t-th iteration.

[0132] Among them, the inertia weight of the mth solution at the tth iteration is [ω t ] m , which depends on the number of times the objective function value in the historical iteration is equal to the objective function value of the t-1 iteration

[0133]

[0134] Where exp(.) represents e (.), e is the base of natural logarithm, represents t-1 iterations N n The objective function value, (.) T is the vector transpose, YesN n The inertial weight selection procedure uses a larger weight for poor solutions to obtain stronger global search performance, and a weight of 1 for better solutions, thereby speeding up the convergence speed and search accuracy of the algorithm.

[0135] In order to improve the optimization accuracy and avoid falling into local optimization, the present invention uses the matrix S t As the random step size in t iterations, its elements follow the Lévy flight mechanism:

[0136]

[0137] where β is the distribution factor of the Levy flight, It conforms to the standard normal distribution, σ v =1, Γ(.) is the gamma function.

[0138] Step S3, updating the solution obtained by the global search mechanism, wherein the solution obtained by the global search mechanism is limited to [-π, π],

[0139]

[0140] The relevant objective function is Furthermore, the objective function value after local search in the t-1 iteration is Compare and select the better solution:

[0141]

[0142] Step 4: Further, the algorithm performs a local search mechanism:

[0143]

[0144] in, represents the inner product, N n ×(N t +1)-dimensional matrix, whose elements are randomly selected in the range [0, 1]. m1,m2 is N n dimensional vector, is the Heaviside function:

[0145]

[0146] m=1, ..., N n , n=1,...,N t +1, in the local random exploration mechanism, only the random p of the solution a Some parts have been updated to promote the convergence of the algorithm.

[0147] Step S5, repeat step S3 to obtain the corresponding and

[0148] Steps S2 to S5 are repeated until the number of iterations is greater than a predefined maximum number of iterations.

[0149] Step S6, obtaining in, index[.] represents the index symbol;

[0150] Step S7, selecting a phase that satisfies the hardware constraints by taking an approximately optimal approach:

[0151]

[0152] Among them, |.| means to find the absolute value, [.] n represents the nth element of a vector, is to satisfy the resolution k h Hardware-limited phaser phase sets, is to satisfy the resolution k l Hardware-limited set of phase shifter phases.

[0153] Step 8, obtain the final codeword f(θ p ,φ q ), f(θ p ,φ q ) corresponding to the nth activated phase shifter is

[0154] The beneficial effects of the parallel dynamic subarray architecture using combined phase shifters and the millimeter wave large-scale array beamforming method based on parallel dynamic subarrays in the present invention are:

[0155] The present invention can achieve flexible sub-array layout while maintaining high-resolution main lobe. The combined phase shifter architecture can overcome the hardware limitation of analog precoding constant amplitude. An exhaustive search scheme is used to obtain the phase shifter activation strategy, and a cuckoo search algorithm is used to search the phase distribution to obtain the final codeword. A high-resolution and low-grating-lobe beam can be achieved on the basis of a low-complexity architecture.

[0156] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. All equivalent structural changes made by using the contents of the present invention specification and drawings under the concept of the present invention, or directly / indirectly applied in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. A parallel dynamic subarray architecture using a combined phase shifter, characterized in that: The base station of the parallel dynamic subarray architecture is equipped with a PUPA composed of two parallel UPAs to provide services for K single-antenna users at the same time. Each UPA array has N = N x ×N z antennas, where N x and N z are the number of antennas along the x-axis and z-axis respectively. The number of users served by the base station at the same time is affected by the number of RF links N RF The restriction is K≤N RF Each RF link is connected via a high-resolution phase shifter and 2N low-resolution phase shifters, the high-resolution phase shifter and the low-resolution phase shifter are controlled by 2N+1 switches, and the number of phase shifters activated in the two UPAs is and The constraints are The hybrid beamformer at the base station includes a digital beamformer at the baseband: N RF ×K-dimensional matrix F BB , and analog beamformer in the RF domain: 2N×N RF Dimensional matrix F RF , f k is a 2N×1 dimensional matrix, f k is the analog beamforming vector corresponding to RF link k, k = 1, ..., K; The parallel dynamic subarray architecture is constructed by setting ||F RF F BB || 2 =K to normalize the power gain of the hybrid beamformer.

2. A millimeter wave large-scale array beamforming method based on parallel dynamic subarrays, characterized in that: The method is applied to the parallel dynamic subarray architecture using combined phase shifters as claimed in claim 1, and the method comprises the following steps: The base station sends a training signal s to K users, and obtains the beam vector fk through training. The base station can obtain an effective channel between it and user k.

3. The millimeter wave large-scale array beamforming method based on parallel dynamic subarray according to claim 2 is characterized in that: The base station sends a training signal s to K users, obtains a beam vector fk through training, and the base station can obtain an effective channel between itself and user k, including: The base station sends a training signal s to K users, where s k =[0, ..., s k , ...0] T represents the symbol vector sent to user k, (.) T is the vector transpose. In addition P k represents power and is equal to all users, then the signal received by user k is expressed as in, is Gaussian noise with mean 0 and variance h k represents the channel information between the base station and user k, (.) H It means to find the conjugate of the matrix, s u Represents the symbol sent by user u; Assume that a k The geometric channel model of independent propagation paths, the channel vector h between the base station and user k k It can be defined as: where α kl ,θ kl ,φ kl are defined as the gain, elevation angle, and azimuth angle of the lth path respectively; due to the significant path loss caused by the millimeter wave band, the number of transmitting antennas 2N is much higher than the number of resolvable paths L k , that is, 2N>>L k ;function represents the antenna array steering vector, and a(θ, φ) is a UPA steering vector and can be expressed as: in, d represents the distance between antennas, y represents the spacing between two UPAs, λ is the wavelength, N=N x ×N z represents the number of antennas in a UPA, e is the natural logarithm base, j is the imaginary unit, (.) T is the vector transpose; The beam vector fk is obtained through beam training, and the base station obtains the effective channel between it and user k as When zero-forcing precoding is used, the digital baseband precoder is in is the combination of the effective channels of all users; then, the signal to interference and noise ratio at user k is expressed as: P k and P u denote the transmit power serving users k and u respectively, represents the channel information h between the base station and user k k The conjugate of RF represents analog precoding, [F BB ] :,k Indicates digital precoding F BB The kth column vector of represents the variance of Gaussian noise. The sum rate is expressed as: Among them, K is the total number of users, SINR k is the signal to interference and noise ratio corresponding to user k, and log2(.) represents the logarithm with base 2.

4. The millimeter wave large-scale array beamforming method based on parallel dynamic subarray according to claim 3 is characterized in that: The base station sends a training signal s to K users and obtains the beam vector f through training. k , the base station can obtain the effective channel between it and user k, including: The optimal phase shifter selection strategy T is found by exhaustive search, and the phase distribution of the phase shifter is obtained by using the cuckoo search algorithm.

5. The millimeter wave large-scale array beamforming method based on parallel dynamic subarray according to claim 4 is characterized in that: The step of using exhaustive search to find the best phase shifter selection strategy T and using the cuckoo search algorithm to obtain the phase distribution of the phase shifter includes: Obtain the phase shifter selection strategy T through exhaustive search; The phase is updated through a global search mechanism; The phase is updated through a local search mechanism; Obtain final phase distribution information; Output codeword.

6. The millimeter wave large-scale array beamforming method based on parallel dynamic subarray according to claim 5, characterized in that: When considering the combined phase shifter architecture, the direction (θ p ,φ q ) corresponds to the code word f(θ p ,φ q ) by the high-resolution part f h and the low resolution part f l composition: f(θ p ,f q )=f l (i p ,f q )+f h (i p ,f q ), in, represents the low-resolution part of the codeword, where represents the phase of the i-th low-resolution phase shifter, N represents the number of antennas in a UPA, N RF represents the number of radio frequency chains, e is the base of natural logarithm, j is the imaginary unit, (.) T is the vector transpose. represents the high-resolution part of the codebook, where represents the phase of the high phase shifter, e is the base of the natural logarithm, j is the imaginary unit, (.) T is the vector transpose. and They are respectively h Hardware-limited set of phase shifters with resolution k l Hardware-limited set of phase shifter phases.

7. The millimeter wave large-scale array beamforming method based on parallel dynamic subarray according to claim 6, characterized in that: To express the phase shifter selection strategy, T is defined as the activated phase shifter state expressed as: T=diag(t1,t2,…,t 2N ),t i =0or 1,i=1,2,…,2N, Among them, t i =0 means the phase shifter is not activated, otherwise the phase shifter is activated, diag(.) represents the diagonal element, based on the codeword f(θ p ,φ q ) and matrix T, a 90×180-dimensional matrix G is used to represent the discrete beam gain distribution: in Indicates the proper amount of guidance of the PUPA array, (.) H represents conjugate. For the matrix G, select all beam peaks that satisfy the condition G[i, j]>G[i±1, j±1], that is, the gains of all lobes. We define these peak sets as {g1, g2, ..., g M }, where all peaks are sorted in descending order, i.e. g m-1 ≥g m , m = 1, ..., M, the mth peak g m The index in the matrix G is [i m , j m ]; The ratio of the maximum grating lobe gain to the target direction gain is η(θ p ,φ q ) is expressed as in Indicates the direction corresponding to the maximum peak g1 Approximate to (θ p ,φ q ), δ θ (δ φ ) is the direction edge error predefined by the direction angle, that is [i p , j q ] is defined as the target direction (θ p ,φ q ) in the matrix G; the precision of g1 is determined by δ θ and δ φ When the direction of the maximum peak Away from the target direction (θ p ,φ q ), the target direction gain is G[i p , j q ], the maximum grating lobe gain is g1.

8. The millimeter wave large-scale array beamforming method based on parallel dynamic subarray according to claim 7, characterized in that: The step of using exhaustive search to find the best phase shifter selection strategy T and using the cuckoo search algorithm to obtain the phase distribution of the phase shifter includes: Step S1, for the phase shifter selection strategy problem, construct a single phase shifter structure and use the steering vector As the beamforming vector, the high-resolution phase shifter is disabled, h = 0. Therefore, the main lobe gain g1 in the target direction is An exhaustive search method is used to obtain the optimal activation strategy T to minimize the maximum grating lobe value g2; Step S2, YesN n ×(N t +1)-dimensional matrix, representing the random initial value of the solution, represents the total number of activated phase shifters in the two UPAs. First, a global search mechanism is used to update the solution: m=1,…,N n , n=1,...,N t +1, γ g Step size scale factor, ω t YesN n ×1-dimensional matrix, [ω t ] m represents the inertia weight of the mth solution in the tth iteration, S t YesN n ×(N t +1)-dimensional matrix, [S t ] m,n represents the random step size of the mth solution in the tth iteration; Among them, the inertia weight of the mth solution at the tth iteration is [ω t ] m , which depends on the number of times in the historical iteration that the objective function value is equal to the objective function value of the t-1 iteration Where exp(.) represents e (·) , e is the base of natural logarithm, represents t-1 iterations N n The objective function value, (.) T is the vector transpose, YesN n The mean of the target values, ∑(.) represents the summation, and ρ is the step index; Using the matrix S t As the random step size in t iterations, its elements follow the Lévy flight mechanism: Where β is the distribution factor of the Levy flight, It conforms to the standard normal distribution, σ v =1, Γ(.) is the gamma function; Step S3, updating the solution obtained by the global search mechanism, wherein the solution obtained by the global search mechanism is limited to [-π, π], The relevant objective function is Step = step, and the objective function value after local search in t-1 iterations Compare and select the better solution: Step 4: Going forward, the algorithm performs a local search mechanism: in, represents the inner product, N n ×(N t +1)-dimensional matrix, whose elements are randomly selected in the range [0, 1]; m1,m2 is N n dimensional vector, is the Heaviside function: m=1, ..., N n , n=1,...,N t +1, in the local random exploration mechanism, only the random p of the solution a Some parts have been updated to promote the convergence of the algorithm; Step S5, repeat step S3 to obtain the corresponding and Step S6, repeating steps S2 to S5 until the number of iterations is greater than a predefined maximum number of iterations; Step S7, obtaining in, index[.] represents the index symbol; Step S8, selecting a phase that satisfies the hardware constraints in a near-optimal manner: Among them, |.| means to find the absolute value, [.] n represents the nth element of a vector, is to satisfy the resolution k h Hardware-limited phaser phase sets, is to satisfy the resolution k l Hardware-limited phaser phase set; Step 9, obtain the final codeword f(θ p ,φ q ), f(θ p ,φ q ) corresponds to the element of the nth activated phase shifter is