Millimeter wave relay communication system hybrid beam forming method based on depth expansion
By decomposing the fully digital beamformer into an analog-digital-analog hybrid structure and using deep expansion network solutions, the problems of hybrid beamforming design complexity and high computational complexity in the prior art are solved, and efficient spectrum efficiency and Real-time.
Patent Information
- Application Number
- CN202410996731.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-24
- Publication Date
- 2024-11-26
AI Technical Summary
Existing hybrid beamforming designs are difficult to implement in real time in practice, with high complexity and high computational complexity, hindering the real-time application of traditional optimization-based beamforming solutions.
Adopting a hybrid beamforming method of millimeter wave relay communication system based on deep expansion, the fully digital beamformer is decomposed into an analog-digital-analog hybrid structure, and converted into a matrix decomposition optimization problem, and using a deep expansion network for solving, reducing the calculation Complexity, improve system real-time.
The spectrum efficiency of the relay communication system is improved, the system calculation complexity is significantly reduced, the system real-timeness is enhanced, and the application in real-time systems is realized.
Smart Images

Figure CN119030582A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of communication technology, and in particular to a hybrid beamforming method for a millimeter wave relay communication system based on depth expansion. Background Art
[0002] As wireless communication technology moves towards new heights, the millimeter wave band plays a vital role in promoting the next generation of communication networks with its huge spectrum resources and potential transmission rate advantages. Especially in millimeter wave relay systems, hybrid beamforming technology has become a research hotspot in the field of wireless communications because it can significantly improve spectrum efficiency and network coverage.
[0003] The current mainstream hybrid beamforming design uses knowledge from the field of traditional communication systems to establish mathematical models and form a model-driven hybrid transceiver solution. Specifically, the fundamental problem of this type of hybrid transceiver design is to use communication mathematical modeling, take the all-digital transceiver matrix as the target, and solve the digital and analog transceiver matrices through convex optimization methods while satisfying constant modulus constraints and transmit power constraints. This type of algorithm shows good performance, but involves a large number of alternating iterations and high complexity, which hinders the real-time implementation of traditional optimization-based beamforming solutions in practice.
[0004] In recent years, deep learning-based methods have opened up new ways to solve the optimization problem of analog-digital hybrid beamforming, which makes it possible to obtain the optimal hybrid beamformer in a low-complexity way. However, doing so is mainly limited by resource constraints, poor interpretability, and black-box nature. In the deep unfolding method, domain knowledge is used with deep learning to build an interpretable deep learning model, thereby achieving performance gains and being easier to implement. Based on this advantage, it is of practical significance to design a deep learning-based transceiver design method for millimeter-wave relay communication systems. Summary of the invention
[0005] In order to improve the system performance of the communication system and reduce the computational complexity so as to make it possible to apply it in a real-time system, the present invention provides a hybrid beamforming method for a millimeter wave relay communication system based on deep expansion. In the system, the base station adopts a full digital precoder F, the user adopts a full digital combiner W, and the relay node decomposes the full digital beamformer G into an analog-digital-analog hybrid structure. The full digital beamformer G includes a relay analog precoder G rrf , Relay digital precoder G bb and relay analog combiner G trf , the method comprising:
[0006] The problem of maximizing the system spectrum efficiency is solved under the constraint of system transmission power, and it is equivalently converted into the matrix decomposition optimization problem of solving the relay all-digital beamformer G;
[0007] Solving the matrix decomposition optimization problem is equivalently converted into a least squares problem similar to that under Gaussian noise, which is decomposed into two sub-problems, which are:
[0008] The first sub-problem: Solve the relay analog precoder G separately rrf and relay analog combiner G trf ;
[0009] Second sub-problem: According to the solved relay simulation precoder G rrf and relay analog combiner G trf , solve the relay digital precoder G bb ;
[0010] Solve the first and second sub-problems respectively to obtain the relay simulation precoder G rrf , Relay digital precoder G bb and relay analog combiner G trf , and then the hybrid beamforming matrix of the relay node is obtained.
[0011] Beneficial effects of the present invention: The relay node of the present invention adopts a fully digital beamformer G with an analog-digital-analog hybrid structure, solves the problem of maximizing the system spectrum efficiency under the constraint of the system transmission power, and converts it into a matrix decomposition optimization problem for solving the relay fully digital beamformer G, and converts the solution of the matrix decomposition optimization problem into a least squares problem similar to that under Gaussian noise, which is further decomposed into two sub-problems; a deep expansion network is constructed to solve the two sub-problems, and finally the hybrid beamforming matrix of the relay node is obtained. The present invention can not only improve the spectrum efficiency of the relay communication system, but also significantly reduce the system calculation complexity and improve the real-time performance of the system. BRIEF DESCRIPTION OF THE DRAWINGS
[0012] Figure 1 This is a schematic diagram of a millimeter wave relay system model in an embodiment of the present invention;
[0013] Figure 2 A diagram showing the training process of a deep expansion network in an embodiment of the present invention;
[0014] Figure 3 This is a comparison diagram of spectrum efficiency under different data streams and radio frequency chains during the simulation experiment of the present invention;
[0015] Figure 4 This is a comparison chart of the spectrum efficiency of different algorithms under different numbers of antennas during the simulation experiment of the present invention. DETAILED DESCRIPTION
[0016] The terms "first", "second", "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances. This is just a way of distinguishing objects with the same attributes when describing the embodiments of this application.
[0017] 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 in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Among them, the drawings are only used for exemplary illustrations, representing only schematic diagrams, not physical diagrams, and cannot be understood as limitations on the present invention; it is understandable to those skilled in the art that some well-known structures and their descriptions in the drawings may be omitted. It is known to those of ordinary skill in the art that with the development of technology and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.
[0018] The embodiment of the present invention provides a hybrid beamforming method for a millimeter wave relay communication system based on deep expansion. In the millimeter wave relay communication system based on deep expansion, the base station adopts a full digital precoder F, the user adopts a full digital combiner W, and the relay node decomposes the full digital beamformer G into an analog-digital-analog hybrid structure. The full digital beamformer G includes a relay analog precoder G rrf , Relay digital precoder G bb and relay analog combiner G trf , the method comprising:
[0019] The problem of maximizing the system spectrum efficiency is solved under the constraint of system transmission power, and it is equivalently converted into the matrix decomposition optimization problem of solving the relay all-digital beamformer G;
[0020] Solving the matrix decomposition optimization problem is equivalently converted into a least squares problem similar to that under Gaussian noise, which is decomposed into two sub-problems, which are:
[0021] The first sub-problem: Solve the relay analog precoder G separately rrf and relay analog combiner G trf ;
[0022] Second sub-problem: According to the solved relay simulation precoder G rrf and relay analog combiner Gtrf , solve the relay digital precoder G bb ;
[0023] Solve the first and second sub-problems respectively to obtain the relay simulation precoder G rrf , Relay digital precoder G bb and relay analog combiner G trf , and then the hybrid beamforming matrix of the relay node is obtained.
[0024] Figure 1 Schematic diagram of a millimeter wave relay communication system model in an embodiment of the present invention.
[0025] Reference Figure 1 As shown, the millimeter wave relay communication system includes a number of base stations, relays and users. The base station is provided with N t There are N transmitting antennas on both sides of the relay receiving station. r antennas, the user sets N d The base station has N receiving antennas. s A data stream signal x, N s dimensional column vector, the signal x obeys the statistical characteristics E represents the expectation, x H represents the Hermitian transpose of the signal x, N s The base station, user, and relay transceiver both use N rf Root radio links to support N s data streams, the number of which must satisfy N s ≤N rf ≤min(N t , N r , N d ). The base station and the user use a fully digital precoder F and a fully digital combiner W respectively. y represents the received signal of the user. The relay node is provided with a hybrid transceiver for hybrid coding, which is designed based on the fully digital beamformer G. The fully digital beamformer G of the relay node is decomposed into an analog-digital-analog hybrid structure, which includes a relay analog precoder G rrf , Relay digital precoder G bb and relay analog combiner G trf H1 represents the channel matrix from the base station to the relay; H2 represents the channel matrix from the relay to the user.
[0026] According to the base station-relay channel matrix H1 and the relay-user channel matrix H2, the relay's all-digital beamformer G, the base station's all-digital precoder F, and the user's all-digital combiner W are calculated respectively. Specifically:
[0027] S201: Perform singular value decomposition on the channel matrices H1 and H2 respectively to obtain left and right unitary matrices.
[0028] svd(H1)=[U1,S1,A1](1a)
[0029] svd(H2)=[U2,S2,A2](1b)
[0030] In the formula, svd(H1) represents the singular value decomposition of the channel matrix H1, U1 represents the left unitary matrix obtained after the singular value decomposition of the channel matrix H1, A1 represents the right unitary matrix obtained after the singular value decomposition of the channel matrix H1, and S1 represents the diagonal matrix obtained after the singular value decomposition of the channel matrix H1; svd(H2) represents the singular value decomposition of the channel matrix H2, U2 represents the left unitary matrix obtained after the singular value decomposition of the channel matrix H2, A2 represents the right unitary matrix obtained after the singular value decomposition of the channel matrix H2, and S2 represents the diagonal matrix obtained after the singular value decomposition of the channel matrix H2.
[0031] S202: Obtain the all-digital beamformer G of the relay, the all-digital precoder F of the base station and the all-digital combiner W of the user according to the left and right unitary matrices respectively. Specifically:
[0032] All rows and first N rows of the right unitary matrix A1 S The columns are assigned to the all-digital precoder F of the base station, namely:
[0033] F=A1([:,:N S ])(2a)
[0034] In the formula, the equal sign indicates assignment, A1([:,:N S ]) represents all rows and the first N right unitary matrix A1 S List;
[0035] All rows and first N of the left unitary matrix U2 S The column is assigned to the user's all-numeric combiner W, that is:
[0036] W=U2([:,:N S ])(2b)
[0037] In the formula, the equal sign indicates assignment, U2([:,:N S ]) represents all rows and the first N left unitary matrix U2 S List;
[0038] The expression of the relay's all-digital beamformer G is:
[0039] G=A2([:,:NS ])×(U1([:,:N S ])) H (2c)
[0040] In the formula, the equal sign indicates assignment, A2([:,:N S ]) represents all rows and the first N of the right unitary matrix A2 S Column, U1([:,:N S ]) represents all rows and the first N left unitary matrix U1 S List,(·) H represents the conjugate transpose.
[0041] In the embodiment of the present invention, the hybrid transceiver is divided into an analog part and a digital part for separate design; thus, the hybrid beamforming includes the analog part and the digital part. The design process of the hybrid transceiver is as follows:
[0042] Perform singular value decomposition on the channel matrices H1 and H2 respectively to obtain the left and right unitary matrices, and compensate the phase information in the unitary matrices to the relay analog combiner G trf With relay analog precoder G rrf , thus we get the relay analog combiner G trf With relay analog precoder G rrf , complete G rrf , G trf Initialization of
[0043] Construct the inputs of two deep expansion networks A and B (i.e. x, B, z);
[0044] Optimizing the analog combiner G using the deeply unfolded network A trf ;
[0045] Optimizing the analog precoder G using the deep unrolled network B rrf ;
[0046] G trf With G rrf Normalized to meet the relay power constraint, according to the relay simulation combiner G trf , relay analog precoder G rrf Solve the relay digital precoder G bb .
[0047] The network architectures of the deep expansion networks A and B are exactly the same. The difference is that the training parameters are based on different network initialization input data to optimize the analog combiner G respectively. trf With analog precoder G rrf .
[0048] Under the constraints of transmission and relay power, the problem of maximizing the system spectrum efficiency is converted into a matrix decomposition problem of the relay all-digital beamformer G. Specifically:
[0049] Considering the maximum transmit power limit of the base station and the relay, the power constraint of the base station's full digital precoder F and the relay's digital beamformer G are normalized to obtain:
[0050]
[0051] In the formula, the equal sign indicates the assignment, F represents the full digital precoder of the base station, and E t represents the transmission power of the base station, (·) H represents the Hermitian transpose, F H represents the Hermitian transpose of the fully digital precoder F, Tr(·) represents the trace, E r represents the transmit power of the relay, G represents the all-digital beamformer of the relay node, H1 represents the channel matrix from the base station to the relay, (·) -1 / 2 It means to seek -1 / 2 power, σ1 2 represents the noise variance between the base station and the relay.
[0052] The received signal of the user is expressed as:
[0053] y=WH2G trf G bb G rrf H1Fx+WH2G trf G bb G rrf n1+Wn2 (4)
[0054] Where y represents the user's received signal, W represents the user's all-digital combiner, H2 represents the channel matrix from the relay to the user, and G rrf Represents the relay analog precoder, G bb Represents the relay digital precoder, G trf represents the relay analog combiner, H1 represents the channel matrix from the base station to the relay, F represents the full digital precoder of the base station; x represents the signal sent by the base station, which has N s data streams; n1 indicates that the mean from base station to relay is 0 and the variance is σ1 2 The additive complex Gaussian white noise is 0, and the variance is σ2. 2 Additive complex Gaussian white noise.
[0055] The spectrum efficiency of the millimeter wave relay communication system is expressed as:
[0056]
[0057] R1=σ1 2 WH2G trf G bb G rrf (WH2G trf G bb G rrf ) H +σ2 2 WW H (5b)
[0058] Where R represents the spectrum efficiency of the millimeter wave relay communication system, det(·) represents the determinant, N s The unit matrix of order, W represents the user's all-digital combiner, G rrf Represents the relay analog precoder, G bb Represents the relay digital precoder, G trf represents the relay analog combiner, H2 represents the channel matrix from the relay to the user, H1 represents the channel matrix from the base station to the relay, F represents the full digital precoder of the base station, (·) H represents the Hermitian transpose, (·) -1 represents the inverse, R1 represents the intermediate expression about R, σ1 2 represents the noise variance between the base station and the relay, σ2 2 represents the noise variance between relay and user.
[0059] Convert the matrix decomposition problem into a least squares problem similar to that under Gaussian noise. Specifically:
[0060] In order to maximize the spectrum efficiency of the millimeter wave relay communication system, the optimization problem (5) is a very complex logarithmic-determinant optimization problem. The objective function of the optimization problem (5) is converted into a simpler norm form, which is:
[0061] Objective function:
[0062] Constraints:
[0063]
[0064] In the formula, minimize means minimization, ||·|| F represents the F-norm, and the objective function (6a) represents the value when ||GG trf G bb G rrf || F The independent variable G is minimum trf , G bb , G rrf The value of Gtrf Represents the repeater RF combiner, G bb Represents the relay baseband filter, G rrf represents the relay RF precoder, F represents the full digital beamformer of the base station, G represents the digital beamformer of the relay, Indicates any, G trf (m,n) represents the matrix G trf The element in the mth row and nth column of , |·| represents the modulo operation, represents the square of the F norm, H1 represents the channel matrix from the base station to the relay, E r represents the transmission power of the relay, E[·] represents the expectation, x represents the signal sent by the base station, n1 represents the mean of the base station to the relay is 0, and the variance is σ1 2 Additive complex Gaussian white noise.
[0065] The optimization problem (6) is a complex non-convex optimization problem. According to the properties of the matrix Klaunec product, the optimization problem (6) can be equivalently converted into a vector operation, which is specifically expressed as:
[0066]
[0067] In the formula,
[0068]
[0069] In the formula, represents the square of the F norm, and the objective function (7a) represents the Minimum independent variable The value of G trf Represents the initialized relay simulation combiner, Indicates vectorizing the matrix into a column vector; Indicates vectorizing the matrix G into a column vector, T indicates transposition, Indicates the search for G trf The pseudo-reversal, N s The unit matrix of order; vec(G) means to vectorize G into a column vector. represents the Klaunec product; G rrf Represents the initialized relay analog precoder, Indicates the search for G rrf The pseudo-inverse.
[0070] In order to meet the energy constraints of relay nodes, Perform normalization operation:
[0071]
[0072] The optimization problem (7) is a complex number problem. We can introduce new variable symbols to transform the optimization problem (7) into an equivalent real number domain problem. Let:
[0073]
[0074] in It means taking the real part of a vector or matrix. Let x represent the imaginary part of a vector or matrix, and [·] represent a matrix. i represents the i-th element of the real-valued vector x, then x i Consists of real values.
[0075] Through the newly introduced variables z, B, x and feasible set The optimization problem (7) can be transformed into:
[0076]
[0077] In the formula, the optimization problem (10) is to find The value of the independent variable x when it is minimum; Denotes the feasible set of the optimization problem (10). According to the variables z, B, x, a deep expansion network input (z, B, x) can be constructed.
[0078] The two deep unfolded networks A and B have L layers, and their network architectures are exactly the same, and their parameters are obtained after training with different inputs (z, B, x). Using the trained deep unfolded networks A and B, the deep neural network unfolded based on the projected gradient descent algorithm effectively solves the analog beamforming matrix in the analog-digital-analog relay hybrid structure.
[0079] The deep unfolding networks A and B adopt a sparse connection mode, in which only the nodes of each layer are allowed to connect with the nodes in the same vertical alignment position in the directly adjacent layer, thereby constructing a deep unfolding network with a specific sparse connection architecture. The activation function of the network is:
[0080]
[0081] In the formula, f(·) represents the activation function, x represents the input of the activation function f(·), t represents the training parameter, and σ(·) represents the ReLu activation function.
[0082] The relationship between the A and B layers of the deep expansion network is:
[0083]
[0084] Among them, u l-1 =BT ZB T Bx l-1
[0085] Where x l represents the output of the lth layer of the deep expansion network, x l-1 represents the output of the l-1th layer of the deep expansion network, ∏ l (·) represents x l-1 to x l A nonlinear projection of l represents the step size of the lth layer of the deep expansion network, w l represents the training parameters of the lth layer of the deep unfolded neural network, represents the Klaunec product, u l-1 is the intermediate expression, b l represents the training parameters of the lth layer of the deep expanded network, T represents the transpose, and B and z are the variables B and z in the least squares problem respectively.
[0086] Expand the network A according to the depth and each time the network outputs x L Refactoring the analog combinator G trf , according to the depth of network B, each network output Reconstruct the analog precoder G rrf , which are specifically expressed as follows:
[0087]
[0088] In the formula, G trf An analog combiner representing a relay, Set the size to N r N rf ×1 complex column vector is reconstructed into a size of N r ×N rf The matrix G trf ; G rrf represents the analog precoder of the relay, Will be N r N rf ×1 complex column vector is reconstructed into a size of N rf ×N r The matrix G rrf ;x L represents the output of the L-layer deep expansion network A, Represents a real number column vector x L Restore to the corresponding complex column vector, x L,i Represents vector x L The i-th element of , j represents the imaginary unit, |·| represents the modulo operation, represents the output of the L-layer deep expansion network B, Represents a real column vector Recover to the corresponding complex column vector.
[0089] The first sub-problem is solved by using the deep expansion network, which is specifically:
[0090] According to the formula: Construct the input of the first deep expansion network (i.e., network A), namely the variables x, B, and z;
[0091] Use the input to train the first deep expansion network, and use the output of the trained first deep expansion network (i.e., network A) to reconstruct the relay analog combiner G trf ;
[0092] According to the formula
[0093] Construct the input of the second deep expansion network (i.e., network B), namely the variables x, B, z;
[0094] The second deep expansion network is trained using the input, and the output of the trained second deep expansion network (i.e., network B) is used to reconstruct the relay simulation precoder G rrf .
[0095] The loss functions of the first deep expansion network (i.e., network A) and the second deep expansion network (i.e., network B) are respectively:
[0096]
[0097] In the formula, represents the loss function of the first deep expanded network, G trf Represents the output x of the network A according to the depth expansion L The reconstructed analog combiner G trf , represents the loss function of the second deep expanded network, G rrf Represents the output of network B according to the depth expansion The reconstructed analog precoder, Indicates the search for G trf The pseudo-reversal, Indicates the search for G rrf is the pseudo-inverse of , and G represents the all-digital beamformer of the relay.
[0098] Figure 2 This is a diagram of the training process of the deep expansion network in an embodiment of the present invention.
[0099] Reference Figure 2As shown in the figure, since the structures of the deep unfolded networks A and B are the same, they are trained by constructing different network inputs and using different loss functions. Therefore, the training process of the deep unfolded networks A and B is the same, which is specifically as follows:
[0100] (I) According to formulas (1a), (1b), and (1c), the channel matrix and are subjected to singular value decomposition to obtain left and right unitary matrices;
[0101] (II) Compensate the phase information in the unitary matrix to the relay analog combiner G trf Analog precoder G with relay rrf , the analog combiner G of the relay trf Analog precoder G with relay rrf , complete G rrf , G trf Initialization of
[0102] (III) According to equations (7), (8) and (9), the input of networks A and B are constructed respectively. After forward propagation, the training parameters of networks A and B are updated according to the loss function respectively.
[0103] (IV) Set the number of training channels H1 and H2 to N (for example, set N=4000), traverse the two training channels respectively, and repeat steps (I), (II), and (III);
[0104] (V) Set the number of training rounds to r (e.g., r=50), and repeat (IV) r times.
[0105] The training data of the deep expansion network A and B are z, B, x, and the loss function Update the parameters in network A Through the loss function Update the parameters in network B
[0106] After the training, the trained deep expansion networks A and B are obtained respectively.
[0107] According to the final output of the trained deep expansion network A, the final relay simulation combiner G is reconstructed. trf ;
[0108] The final relay simulation precoder G is obtained by reconstructing the final output of the trained deep expansion network B rrf .
[0109] According to the final analog combiner G trf and analog precoder G rrf Solve the digital precoder G bb , the problem is specifically:
[0110] Objective function: Constraints:
[0111] In the formula, minimize means minimization, represents the square of the F norm, and the objective function (15a) is to find The independent variable G is minimum bb The value of G bb represents the relay digital precoder, G represents the relay's all-digital beamformer, G trf represents the solved relay analog combiner, G rrf represents the relay analog precoder obtained by solving, E r represents the transmission power of the relay, E[·] represents the expectation, H1 represents the channel matrix from the base station to the relay, x represents the signal sent by the base station, n1 represents the mean of the base station to the relay is 0, and the variance is σ1 2 Additive complex Gaussian white noise. .
[0112] The optimization problem (15) is a convex optimization problem, and there exists a first-order optimal solution. According to the first-order optimality condition of the optimization problem (7), the relay digital precoder G bb It can be expressed as:
[0113]
[0114] In the formula, G trf represents the solved relay analog combiner, G rrf represents the relay analog precoder obtained by solution, Express request The inverse, Express request The inverse of (·) H It means to find the Hermitian transpose.
[0115] In order to meet the relay energy constraint, it is necessary to solve the relay digital precoder G bb After normalization, we get:
[0116]
[0117] In the formula, the equal sign indicates assignment, E r represents the transmission power of the relay, H1 represents the channel matrix from the base station to the relay, H2 represents the channel matrix from the relay to the user, σ1 2 represents the noise variance between the base station and the relay, Tr(·) represents the trace, G trf Represents a relay analog combiner, G rrf Represents the relay analog precoder, G bbrepresents the relay analog combiner, F represents the base station's all-digital precoder, (·) H represents the Hermitian transpose, (·) -1 / 2 It means to raise -1 / 2 to the power.
[0118] In the embodiment of the present invention, the relay node is provided with a hybrid transceiver for hybrid coding, and the hybrid transceiver is designed according to the all-digital beamformer G. The all-digital beamformer G includes an analog precoder G rrf , digital precoder G bb and analog combiner G trf The above steps have obtained the simulated precoder G rrf , digital precoder G bb and analog combiner G trf The hybrid beamforming matrix design of the relay node is completed.
[0119] Simulation settings: Base station transmission power E t , the transmission power of the relay E r Both are set to 2; n1 represents the additive complex Gaussian white noise with a mean of 0 and a variance of 2 from the base station to the relay; n2 represents the additive complex Gaussian white noise with a mean of 0 and a variance of 2 from the relay to the user.
[0120] Data stream N sent by the base station s The number is 2; the number of RF chains is N rf is 6; the channels between the base station and the relay, and between the relay and the user use the Saleh-Valenzuela channel model to generate 4000 training channels H1 and H2, and 1000 test channels, respectively. The number of scattering clusters for each channel is 5, the number of paths for each scattering cluster is 10, and the arrival angle and departure angle follow the Laplace distribution, with the average angle uniformly distributed between (0,2π], and the angle expansion is 10°. All simulation experiments are implemented using Python 3.9 and Pytorch library (version 1.8.1+cu101). In the training phase, the training round r is 50, the number of network layers L is 6, the learning rate is 0.0008, and the weight decay is 1×10 -6 , the initial value of the training parameter t is 0.1, w l 、b l The initial value of is a random number in (0,1], where l∈[1,L].
[0121] Simulation results:
[0122] 1. Comparison of spectrum efficiency of RF chain under different data flows
[0123] Figure 3 This is a comparison chart of spectrum efficiency under different data streams and radio frequency chains during the simulation experiment of the present invention. Figure 3 In the example, the number of base station transmitting antennas is N t The number of antennas is 32, the number of relay receiving and transmitting antennas is 16, and the number of user receiving antennas is 24. The "proposed algorithm" in the figure refers to the method of the present invention. The results show that increasing the number of data streams and RF chains can improve system performance; under different signal-to-noise ratios, the performance of the proposed method of the present invention is better than the deep unfolding network algorithm based on the weighted minimum mean square error algorithm (Weighted Minimum Mean-Square Error, WMMSE), and is close to the optimal performance under ideal conditions.
[0124] 2. Comparison of spectrum efficiency of different algorithms under different numbers of antennas;
[0125] Figure 4 This is a comparison chart of the spectrum efficiency of different algorithms under different numbers of antennas during the simulation experiment of the present invention. Figure 4 Medium, scene I:N t is 48, N r is 24, N d is 8, N s is 2, N rf 6, Scene II: N t is 24, N r is 8, N d is 8, N s is 2, N rf is 6, and the "proposed algorithm" in the figure refers to the method of the present invention. The spectrum efficiency of the all-digital relay beamforming is the theoretical upper limit of the simulation results of the hybrid beamforming system. The present invention is closest to the optimal spectrum rate, and its performance is better than the WMMSE algorithm in the scientific paper [1] and the WMMSE deep expansion algorithm based on the scientific paper [2]. Among them, the scientific papers [1] and [2] are respectively:
[0126] [1]Jiang L,Jafarkhani H.mmWave amplify-and-forward MIMO relaynetworks with hybrid precoding / combining design[J].IEEE Transactions onWireless Communications,2019,19(2):1333-1346.
[0127] [2] Luo Z, Wang X, Xie F, et al. Hybrid Beamforming for Millimeter WaveRelay Systems: A Deep Unfolding Approach, to be published.
[0128] 3. Comparison of running time of different algorithms
[0129] Table 1 is a comparison table of the running time of different algorithms in the simulation experiment of the present invention. In Table 1, under different antenna scenarios, the present invention is 115 to 310 times faster than the WMMSE algorithm in the scientific paper [1]. In most cases, the present invention shows better performance than the WMMSE deep expansion algorithm based on the scientific paper [2].
[0130]
[0131] Table 1 is a comparison table of the running time of different algorithms in the simulation experiment of the present invention
[0132] From the running time of different algorithms in Table 1 and the analysis of the simulation experiment results, it can be seen that the present invention can not only improve the spectrum efficiency of the relay communication system, but also significantly reduce the system calculation complexity, reduce the system running time, and improve the system real-time performance.
[0133] A person skilled in the art may understand that all or part of the steps in the various methods of the above embodiments may be completed by instructing the relevant hardware through a program, and the program may be stored in a computer-readable storage medium, which may include: ROM, RAM, disk or CD, etc.
[0134] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A hybrid beamforming method for a millimeter wave relay communication system based on deep expansion, characterized in that: In this system, the base station uses a fully digital precoder F, the user uses a fully digital combiner W, and the relay node decomposes the fully digital beamformer G into an analog-digital-analog hybrid structure. The fully digital beamformer G includes a relay analog precoder G rrf , Relay digital precoder G bb and relay analog combiner G trf , the method comprising: The problem of maximizing the system spectrum efficiency is solved under the constraint of system transmission power, and it is equivalently converted into the matrix decomposition optimization problem of solving the relay all-digital beamformer G; Solving the matrix decomposition optimization problem is equivalently converted into a least squares problem similar to that under Gaussian noise, which is decomposed into two sub-problems, which are: The first sub-problem: Solve the relay analog precoder G separately rrf and relay analog combiner G trf ; Second sub-problem: According to the solved relay simulation precoder G rrf and relay analog combiner G trf , solve the relay digital precoder G bb ; Solve the first and second sub-problems respectively to obtain the relay simulation precoder G rrf , Relay digital precoder G bb and relay analog combiner G trf , and then the hybrid beamforming matrix of the relay node is obtained.
2. The hybrid beamforming method for a millimeter wave relay communication system based on deep expansion according to claim 1, characterized in that: The system transmit power constraint includes the maximum transmit power constraint of the base station and the relay. To meet the system transmit power constraint, after normalization, the full digital precoder of the base station and the full digital beamformer of the relay node are respectively expressed as: In the formula, the equal sign indicates the assignment, F represents the full digital precoder of the base station, and E t represents the transmission power of the base station, (·) H represents the Hermitian transpose, F H represents the Hermitian transpose of the fully digital precoder F, Tr(·) represents the trace, E r represents the transmit power of the relay, G represents the all-digital beamformer of the relay node, H1 represents the channel matrix from the base station to the relay, (·) -1 / 2 It means to seek -1 / 2 power, σ1 2 represents the noise variance between the base station and the relay.
3. The hybrid beamforming method for a millimeter wave relay communication system based on deep expansion according to claim 1, characterized in that: The spectrum efficiency of the millimeter wave relay communication system is expressed as: Where R represents the spectrum efficiency of the millimeter wave relay communication system, det(·) represents the determinant, N s The unit matrix of order, W represents the user's all-digital combiner, G rrf Represents the relay analog precoder, G bb Represents the relay digital precoder, G trf represents the relay analog combiner, H2 represents the channel matrix from the relay to the user, H1 represents the channel matrix from the base station to the relay, F represents the full digital precoder of the base station, (·) H represents the Hermitian transpose, (·) -1 represents the inverse, R1 represents the intermediate expression about R, σ1 2 represents the noise variance between the base station and the relay, σ2 2 represents the noise variance between relay and user.
4. The hybrid beamforming method for a millimeter wave relay communication system based on deep expansion according to claim 1, characterized in that: The matrix decomposition optimization problem of the relay all-digital beamformer G is specifically: Objective function: Constraints: In the formula, ||·|| F Indicates the F-norm. Indicates seeking to be||GG trf G bb G rrf || F The independent variable G is minimum trf , G bb , G rrf The value of G trf Represents the repeater RF combiner, G bb Represents the relay baseband filter, G rrf represents the relay RF precoder, F represents the full digital beamformer of the base station, G represents the digital beamformer of the relay, Indicates any, G trf (m,n) represents the matrix G trf The element in the mth row and nth column of , |·| represents the modulo operation, represents the square of the F norm, H1 represents the channel matrix from the base station to the relay, E r represents the transmission power of the relay, E[·] represents the expectation, x represents the signal sent by the base station, n1 represents the mean of the base station to the relay is 0, and the variance is σ1 2 Additive complex Gaussian white noise.
5. The hybrid beamforming method for a millimeter wave relay communication system based on deep expansion according to claim 4 is characterized in that: The least squares problem is specifically: Objective function: in, or In the formula, It means to find the square of the F norm, Expressing the desire Minimum independent variable The value of G trf Represents the initialized relay simulation combiner, Indicates vectorizing the matrix into a column vector; Indicates vectorizing the matrix G into a column vector, T indicates transposition, Indicates the search for G trf The pseudo-reversal, N s The unit matrix of order; vec(G) means to vectorize G into a column vector. represents the Klaunec product; G rrf Represents the initialized relay analog precoder, Indicates the search for G rrf of pseudo-rebellion; make: In the formula, It means taking the real part of a vector or matrix. Take represents the imaginary part of a vector or matrix, [·] represents a matrix; The objective function of the least squares problem is equivalently transformed into: In the formula, Expressing the desire The value of the independent variable x when it is minimum; represents the feasible set of the objective function.
6. The hybrid beamforming method for a millimeter wave relay communication system based on deep expansion according to claim 5, characterized in that: Two deep neural unfolding networks with the same structure are constructed to solve the first sub-problem. The activation function of the deep unfolding network is: Where f(·) represents the activation function, x represents the input of the activation function f(·), t represents the training parameter, and σ(·) represents the ReLu activation function; The relationship between the layers of the deep expansion network is: u l-1 =B T z-B T Bx l-1 In the formula, x l represents the output of the lth layer of the deep expansion network, w l represents the training parameters of the lth layer of the deep unfolded neural network, x l-1 represents the output of the l-1th layer of the deep expanded network, represents the Klaunec product, u l-1 is the intermediate expression, b l represents the training parameters of the lth layer of the deep expanded network, T represents the transpose, and B and z are the variables B and z in the least squares problem respectively.
7. The hybrid beamforming method for a millimeter wave relay communication system based on depth expansion according to claim 6, characterized in that: The first sub-problem is solved by using the deep expansion network, which is specifically: according to the formula: Construct the input of the first deep expansion network, namely the variables x, B, z; Use the input to train the first deep expansion network, and use the output of the trained first deep expansion network to reconstruct the relay analog combiner G trf ; According to the formula Construct the input of the second deep expansion network, namely the variables x, B, z; The second deep expansion network is trained using the input, and the output of the trained second deep expansion network is used to reconstruct the relay simulation precoder G rrf .
8. The hybrid beamforming method for a millimeter wave relay communication system based on deep expansion according to claim 7, characterized in that: The loss functions of the first deep expansion network and the second deep expansion network during the training process are respectively: In the formula, represents the loss function of the first deep expanded network, G trf Indicates that the network output x is expanded according to the first depth L The reconstructed relay simulation combiner, Indicates the search for G trf of pseudo-rebellion; represents the loss function of the second deep expanded network, G rrf Indicates that the network output is expanded according to the second depth The reconstructed analog precoder, Indicates the search for G rrf is the pseudo-inverse of , and G represents the all-digital beamformer of the relay.
9. The hybrid beamforming method for a millimeter wave relay communication system based on depth expansion according to claim 5, characterized in that: The second sub-problem is expressed as: Objective function: Constraints: In the formula, minimize means minimization, represents the square of the F norm, Expressing the desire The independent variable G is minimum bb The value of G bb represents the relay digital precoder, G represents the relay's all-digital beamformer, G trf represents the solved relay analog combiner, G rrf represents the relay analog precoder obtained by solving, E r represents the transmission power of the relay, E[·] represents the expectation, H1 represents the channel matrix from the base station to the relay, x represents the signal sent by the base station, n1 represents the mean of the base station to the relay is 0, and the variance is σ1 2 Additive complex Gaussian white noise.
10. The hybrid beamforming method for a millimeter wave relay communication system based on deep expansion according to claim 9, characterized in that: Solve the second sub-problem and get the relay digital precoder G bb , which is specifically expressed as: In the formula, G trf represents the solved relay analog combiner, G rrf represents the relay analog precoder obtained by solution, Express request The inverse, Express request The inverse of (·) H It means to find the Hermitian transpose.
