Large-scale MIMO deep unfolding precoding method and device based on conjugate gradient method

Abstract

The application discloses a large-scale MIMO deep unfolding precoding method and device based on a conjugate gradient method, and belongs to the wireless communication field, wherein the method comprises the following steps: obtaining a downlink air interface channel vector and a transmission vector; inputting the downlink air interface channel vector and the transmission vector into a pre-trained nonlinear neural network to obtain a precoded transmission signal. The nonlinear neural network is a combination of a deep unfolding multi-layer nonlinear subnetwork with a step length in a conjugate gradient algorithm as a training parameter and a single-layer linear subnetwork at a receiver with receiver coefficients as a training parameter, and is obtained by training a downlink air interface channel vector and a transmission vector training set. The internal structure of the conjugate gradient MMSE precoding algorithm is combined with an advanced DNN network, a model-driven supervised neural network is constructed by using deep unfolding, and compared with a data-driven 'black box' DNN network, the network has better generalization ability and interpretability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless communication, in particular to a large-scale MIMO deep unfolding precoding method and device based on conjugate gradient method. BACKGROUND

[0002] Large-scale MIMO (Multiple-Input Multiple-Output) technology is the key to improve the spectrum efficiency of the next generation of wireless networks, the main idea is to equip the base station with hundreds of antennas, which serve a relatively small number of users in the same frequency band. Low-complexity precoding algorithm has always been a research hotspot to solve the problem of multi-user interference. The precoding algorithm can obtain the corresponding precoding matrix based on the obtained channel state information, and use the precoding matrix to preprocess the transmitted signal before transmission to improve the transmission rate and link reliability of the system. For the downlink of the multi-user MIMO system, the use of precoding technology not only reduces the interference of other users, so that the signal energy is concentrated in a certain receiving end direction, improving the spectrum utilization, but also places a large amount of complex calculation at the sending end with better computing performance, which can reduce the complexity of the receiver.

[0003] The precoding technology can be divided into linear precoding and nonlinear precoding according to the design scheme. In the conventional scene, the nonlinear precoding performance is better than the linear precoding, but the linear precoding has the advantage of low complexity and is more easily applied in practical scenarios. And the physical characteristics of large-scale MIMO make the linear precoding technology can be wireless approximation to the performance of nearly optimal nonlinear precoding technology, so the linear precoding technology of large-scale MIMO has become a research focus and hotspot in 5G communication technology.

[0004] The existing traditional minimum mean square error (MMSE) precoding method is an improved form of zero forcing (ZF) precoding, but it needs to invert the channel. Due to the use of large-scale antennas, the dimension of the channel matrix and the precoding matrix is increased, and the complexity of the precoding algorithm, the system hardware cost and the implementation difficulty will be increased. SUMMARY

[0005] The present application provides a large-scale MIMO deep unfolding precoding method and device based on conjugate gradient method, which can achieve a better compromise between performance and complexity compared to the traditional MMSE precoding method, which is beneficial to engineering implementation.

[0006] The first aspect embodiment of the present application provides a large-scale MIMO deep unfolding precoding method based on a conjugate gradient method, comprising the following steps: obtaining a downlink air interface channel vector and a transmission vector; inputting the downlink air interface channel vector and the transmission vector into a pre-trained nonlinear neural network to obtain a precoded transmission signal; wherein the nonlinear neural network is a combination of a deep unfolding multi-layer nonlinear subnetwork with a step length in the conjugate gradient algorithm as a training parameter and a single-layer linear subnetwork at a receiver with receiver coefficients as a training parameter, and is trained by using a downlink air interface channel vector and transmission vector training set.

[0007] Optionally, in an embodiment of the present application, the input of the deep unfolding multi-layer nonlinear subnetwork is the downlink air interface channel vector and transmission vector training set, the output is the precoded transmission signal, each layer in the subnetwork corresponds to one iteration of the conjugate gradient algorithm, and a normalization processing layer is added after the deep unfolding multi-layer nonlinear subnetwork; the input of the single-layer linear subnetwork at the receiver is the output of the normalization processing layer, which is used to generate receiver coefficients configured for each user, and before the downlink air interface channel vector and the transmission vector are input into the pre-trained nonlinear neural network, the method further comprises: constructing the downlink air interface channel vector and transmission vector training set; training the nonlinear neural network by using the downlink air interface channel vector and transmission vector training set to obtain a step length parameter of the nonlinear neural network and receiver coefficients configured for each user; and updating the parameters of the nonlinear neural network by using the step length parameter and the receiver coefficients configured for each user to obtain the pre-trained nonlinear neural network.

[0008] Optionally, in an embodiment of the present application, the loss objective function for training the nonlinear neural network by using the downlink air interface channel vector and transmission vector training set is:

[0009]

[0010] The constraint condition is:

[0011]

[0012] wherein E[·] is an expectation, |·| is a modulus of a complex number, ‖·‖ is a vector two-norm, (·) H is a conjugate transpose of a vector, h k ∈C N×1 is a downlink channel between a base station and a user k, W=[w1,w2…w k ]∈C N×K is a precoding matrix, N and K are respectively a number of antennas at a base station end and a number of users at a receiving end, w k ∈C N×1is the precoding vector for the kth user, q∈C N×1 is the precoded transmit signal, a k is the receiver coefficient configured for the kth user, q∈C is a complex Gaussian noise, is the noise power, P is the base station end transmit power constraint.

[0013] The second aspect embodiment of the present application provides a large-scale MIMO deep expansion precoding device based on the conjugate gradient method, comprising: an acquisition module, configured to acquire a downlink air interface channel vector and a transmit vector; a precoding module, configured to input the downlink air interface channel vector and the transmit vector into a pre-trained nonlinear neural network to obtain a precoded transmit signal; wherein the nonlinear neural network is a combination of a deep expansion multi-layer nonlinear subnetwork with a step length in the conjugate gradient algorithm as a training parameter and a single-layer linear subnetwork at a receiver with a receiver coefficient as a training parameter, and is trained by using a downlink air interface channel vector and transmit vector training set.

[0014] Optionally, in an embodiment of the present application, the input of the deep expansion multi-layer nonlinear subnetwork is the downlink air interface channel vector and transmit vector training set, the output is the precoded transmit signal, each layer in the subnetwork corresponds to one iteration of the conjugate gradient algorithm, and a normalization processing layer is added after the deep expansion multi-layer nonlinear subnetwork; the input of the single-layer linear subnetwork at the receiver is the output of the normalization processing layer, which is used to generate the receiver coefficient configured for each user, and the device further comprises:

[0015] a training module, configured to, before inputting the downlink air interface channel vector and the transmit vector into the pre-trained nonlinear neural network, construct the downlink air interface channel vector and transmit vector training set; train the nonlinear neural network by using the downlink air interface channel vector and transmit vector training set to obtain the step length parameter of the nonlinear neural network and the receiver coefficient configured for each user; and update the parameters of the nonlinear neural network by using the step length parameter and the receiver coefficient configured for each user to obtain the pre-trained nonlinear neural network.

[0016] Optionally, in an embodiment of the present application, the loss objective function for training the nonlinear neural network by using the downlink air interface channel vector and transmit vector training set is:

[0017]

[0018] The constraint condition is:

[0019]

[0020] wherein E[·] is the expectation, |·| is the modulus of a complex number, ‖·‖ is the vector two-norm, (·) H is the conjugate transpose of a vector, h k ∈C N×1 is the downlink channel between the base station and user k, W = [w1, w2…w k ]∈C N×K is the precoding matrix, N and K are the number of antennas at the base station end and the number of receiving end users, w k ∈C N×1 is the precoding vector of the kth user, q ∈ C N×1 is the precoded transmission signal, a k ∈C is the receiver coefficient configured for the kth user, is a complex Gaussian noise, is the noise power, and P is the transmission power constraint at the base station end.

[0021] The third aspect embodiment of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the program to perform the large-scale MIMO deep unfolding precoding method based on the conjugate gradient method as described in the above embodiments.

[0022] The fourth aspect embodiment of the present application provides a computer readable storage medium having a computer program stored thereon, and the program is executed by a processor to perform the large-scale MIMO deep unfolding precoding method based on the conjugate gradient method as described in the above embodiments.

[0023] The large-scale MIMO deep unfolding precoding method and device based on the conjugate gradient method of the embodiments of the present application have the following beneficial effects:

[0024] 1) For a multi-user large-scale MIMO wireless communication system, the neural network constructed by using the deep learning method can effectively estimate the receiving coefficient at each user of the receiving end.

[0025] 2) The neural network combines the internal structure of the conjugate gradient MMSE precoding algorithm and the advanced DNN network, and constructs a model-driven supervised neural network by using deep unfolding. Compared with the data-driven “black box” DNN network, the network has better generalization ability and interpretability.

[0026] 3) While ensuring the system performance, the computational complexity of the traditional MMSE precoding is significantly reduced, which is convenient for engineering implementation.

[0027] Additional aspects and advantages of the present application will be partially given in the following description, partially will become obvious from the following description, or will be understood by practicing the present application. BRIEF DESCRIPTION OF DRAWINGS

[0028] The above and / or additional aspects and advantages of the present application will become apparent and be readily appreciated from the following description, including the references to the figures, in which:

[0029] Figure 1 A flow chart of a large-scale MIMO deep unfolding precoding method based on a conjugate gradient method according to an embodiment of the present application;

[0030] Figure 2 A non-linear neural network structure block diagram according to an embodiment of the present application;

[0031] Figure 3 A conjugate gradient algorithm each iteration structure diagram according to an embodiment of the present application;

[0032] Figure 4 A simulation experiment comparison result diagram according to an embodiment of the present application;

[0033] Figure 5 A large-scale MIMO deep unfolding precoding device structure schematic diagram according to an embodiment of the present application;

[0034] Figure 6 A structure schematic diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0035] Embodiments of the present application are described in detail below with reference to the accompanying drawings, in which like or similar elements are denoted by the same or similar reference signs, and examples of the embodiments are described below by referring to the accompanying drawings. The embodiments described below by referring to the accompanying drawings are exemplary and are intended to explain the present application, and should not be understood as limiting the present application.

[0036] A large-scale MIMO deep unfolding precoding method and device based on a conjugate gradient method according to an embodiment of the present application are described below with reference to the accompanying drawings. The conventional minimum mean square error (MMSE) precoding method mentioned in the above background technology is an improved form of zero forcing (ZF) precoding, but requires channel inversion. Due to the use of large-scale antennas, the dimensions of the channel matrix and the precoding matrix are increased, and the problems of increased precoding algorithm complexity, system hardware cost and implementation difficulty are caused. The present application provides a large-scale MIMO deep unfolding precoding method based on a conjugate gradient method. In the method, a non-linear neural network composed of a deep unfolding multi-layer non-linear sub-network with a step size as a training parameter and a single-layer linear sub-network at a receiver with a receiver coefficient as a training parameter is used to precode a transmit signal. Compared with the conventional MMSE precoding method, the present application can achieve a better compromise between performance and complexity, which is conducive to engineering implementation.

[0037] Specifically, Figure 1 A flowchart of a large-scale MIMO deep unfolding precoding method based on a conjugate gradient method according to an embodiment of the present application is provided.

[0038] As Figure 1 shown, the large-scale MIMO deep unfolding precoding method based on the conjugate gradient method includes the following steps:

[0039] In step S101, a downlink air interface channel vector and a transmission vector are obtained.

[0040] In an embodiment of the present application, the downlink air interface channel vector and the transmission vector are used for precoding.

[0041] In step S102, the downlink air interface channel vector and the transmission vector are input into a pre-trained nonlinear neural network to obtain a precoded transmission signal; wherein the nonlinear neural network is a combination of a deep unfolding multi-layer nonlinear subnetwork with a step length in the conjugate gradient algorithm as a training parameter and a single-layer linear subnetwork at the receiver with a receiver coefficient as a training parameter, and is trained by using a downlink air interface channel vector and transmission vector training set.

[0042] Optionally, in an embodiment of the present application, the input of the deep unfolding multi-layer nonlinear subnetwork is a downlink air interface channel vector and transmission vector training set, the output is a precoded transmission signal, each layer in the subnetwork corresponds to one iteration of the conjugate gradient algorithm, and a normalization processing layer is added after the deep unfolding multi-layer nonlinear subnetwork; the input of the single-layer linear subnetwork at the receiver is the output of the normalization processing layer, which is used to generate a receiver coefficient configured for each user, and before the downlink air interface channel vector and the transmission vector are input into the pre-trained nonlinear neural network, it further includes: constructing a downlink air interface channel vector and transmission vector training set; training the nonlinear neural network by using the downlink air interface channel vector and transmission vector training set to obtain a step length parameter of the nonlinear neural network and a receiver coefficient configured for each user; updating the parameters of the nonlinear neural network by using the step length parameter and the receiver coefficient configured for each user to obtain the pre-trained nonlinear neural network.

[0043] Specifically, the nonlinear neural network is a supervised nonlinear neural network, and the training data set required for training is a downlink air interface channel vector H d and a transmission vector t.

[0044] The MMSE precoding algorithm based on the conjugate gradient is used to build a deep unfolding multi-layer nonlinear subnetwork with a step length as a training parameter and a single-layer linear subnetwork at the receiver with a receiver coefficient as a training parameter, and a supervised nonlinear neural network composed of the training data set generated by the step length is jointly trained.

[0045] The nonlinear neural network is composed of a deep unfolded multi-layer nonlinear subnetwork and a single-layer linear subnetwork at the receiver, wherein an input of the deep unfolded multi-layer nonlinear subnetwork is a training data set of the nonlinear neural network, an output is a precoded transmission signal, and each layer in the subnetwork corresponds to one iteration of the conjugate gradient method; a normalization processing layer is added after the deep unfolded multi-layer nonlinear subnetwork; an input of the single-layer linear subnetwork at the receiver is an output of the normalization processing layer, and the single-layer linear subnetwork is designed to train receiver coefficients configured for each user at the receiver.

[0046] The received signal of the kth user is:

[0047]

[0048] The base station jointly trains the neural network by using the generated training data set and the following training target:

[0049] The loss target function of the training is to minimize

[0050] The constraint condition is:

[0051] wherein E[·] is an expectation, |·| represents a modulus of a complex number, ‖·‖ represents a two-norm of a vector, (·) H represents a conjugate transpose of a vector, h k ∈C N×1 represents a downlink channel between the base station and the kth user, W = [w1, w2, …, wN] ∈ C k ]∈C N×K represents a precoding matrix, N and K represent a number of antennas at the base station end and a number of users at the receiving end respectively, w k ∈C N×1 represents a precoding vector of the kth user, q ∈ C N×1 represents a precoded transmission signal, a k ∈C represents receiver coefficients configured for the kth user, represents a complex Gaussian noise, represents a noise power, and P is a transmission power constraint at the base station end.

[0052] The base station performs offline stage training on the nonlinear neural network to obtain a step size parameter in the nonlinear neural network and the receiver coefficients configured for each user, and then performs online stage calculation to obtain the precoded transmission signal.

[0053] The conjugate gradient algorithm (CG) is an effective iterative method for solving a linear equation system, and the problem to be solved is in the form of: wherein A ∈ C K×Kis a positive definite matrix, and ||·|| denotes the vector two-norm, b∈C K ×1 Compared with the direct calculation CG algorithm iteratively obtains and each iteration requires low complexity. The algorithm can converge after multiple iterations, and the iteration process can be terminated in advance, while still obtaining a solution close to the accurate result.

[0054] The conjugate gradient algorithm for downlink precoding is in the form of:

[0055] where The precoding vector is q = H d v.

[0056] The specific iteration process of the designed conjugate gradient algorithm is as follows:

[0057]

[0058] In the above iteration, α m and β m are step size parameters in the conjugate gradient algorithm.

[0059] The nonlinear neural network is composed of a deep expansion of a multi-layer nonlinear subnetwork and a single-layer linear subnetwork at the receiver, as shown in Figure 2 , wherein the input of the deep expansion of the multi-layer nonlinear subnetwork is the training data set of the nonlinear neural network, and the output is the precoded transmission signal. The multi-layer nonlinear subnetwork takes the step size in the conjugate gradient algorithm as a training parameter, and the conjugate gradient algorithm is designed in depth. Each layer in the subnetwork corresponds to one iteration of the conjugate gradient algorithm, and a normalization processing layer is added to meet the power constraint. The iteration process of the algorithm is shown in Figure 3 .

[0060] The output of the normalization processing layer after the multi-layer nonlinear subnetwork is taken as the input of the single-layer linear subnetwork at the receiver. The single-layer linear subnetwork at the receiver is designed for the receiver coefficients provided for each user at the training receiver. The training goal is to minimize the MSE at the receiver, and the generated data set is used to jointly train the composed nonlinear neural network. The network constructed by using the deep learning method can estimate the receiving coefficients of each user at the receiving end.

[0061] In order to verify the technical effect of the present application, a simulation experiment is carried out, and the parameters involved in the simulation experiment are shown in Table 1:

[0062] Table 1 Simulation experiment parameter table

[0063] Parameter Value Number of base station transmit antennas 128 Number of user receive antennas 1 Number of users 16 Base station transmit power 30 dBm Channel model Rayleigh channel Modulation scheme 16 QAM Number of training sets 10000 Number of test sets 1000

[0064] Table 2 comparison table of calculation complexity

[0065]

[0066] To further illustrate the effect of the present application, the performance simulation and complexity calculation of the traditional minimum mean square error (MMSE) algorithm and the original conjugate gradient (CG) iterative algorithm are also carried out, and compared with the present application, as shown in Figure 4 .

[0067] Specifically, Table 1 is a simulation experiment parameter table, and Table 2 is a comparison table of the MMSE method, the CG original iterative algorithm and the method of the present application in terms of calculation complexity. The calculation method of complexity is to calculate the number of real multiplication, RDiv represents real division, represents the number of iterations of the CG algorithm, represents the number of iterations of the CG algorithm. The simulation results show that the method of the present embodiment can approach the MMSE performance after four iterations, and thus can significantly reduce the calculation complexity.

[0068] According to the large-scale MIMO deep unfolding precoding method based on the conjugate gradient method provided by the embodiment of the present application, for a multi-user large-scale MIMO wireless communication system, the neural network constructed by using the deep learning method can effectively estimate the receiving coefficient at each user of the receiving end. The internal structure of the conjugate gradient MMSE precoding algorithm is combined with the advanced DNN network, and a supervised neural network based on model driving is constructed by using deep unfolding. Compared with the data-driven "black box" DNN network, the network has better generalization ability and interpretability. While ensuring the system performance, the calculation complexity of the traditional MMSE precoding is significantly reduced, which is convenient for engineering implementation.

[0069] Secondly, the large-scale MIMO deep unfolding precoding device based on the conjugate gradient method according to the embodiment of the present application is described with reference to the accompanying drawings.

[0070] Figure 5 The large-scale MIMO deep unfolding precoding device based on the conjugate gradient method according to the embodiment of the present application is provided.

[0071] As Figure 5 shown, the large-scale MIMO deep unfolding precoding device 10 based on the conjugate gradient method includes an acquisition module 100 and a precoding module 200.

[0072] The acquisition module 100 is configured to acquire a downlink air interface channel vector and a transmission vector.

[0073] Optionally, in an embodiment of the present application, the input of the deep unfolded multi-layer nonlinear subnetwork is a downlink air interface channel vector and transmission vector training set, the output is a precoded transmission signal, each layer in the subnetwork corresponds to one iteration of the conjugate gradient algorithm, and a normalization processing layer is added after the deep unfolded multi-layer nonlinear subnetwork; the input of the single-layer linear subnetwork at the receiver is the output of the normalization processing layer, which is configured to generate receiver coefficients configured for each user, and the device further comprises a training module configured to, before the downlink air interface channel vector and the transmission vector are input into the pre-trained nonlinear neural network, construct the downlink air interface channel vector and the transmission vector training set; train the nonlinear neural network using the downlink air interface channel vector and the transmission vector training set to obtain a step parameter of the nonlinear neural network and the receiver coefficients configured for each user; and update the parameters of the nonlinear neural network using the step parameter and the receiver coefficients configured for each user to obtain the pre-trained nonlinear neural network.

[0074] Optionally, in an embodiment of the present application, the loss objective function for training the nonlinear neural network using the downlink air interface channel vector and the transmission vector training set is:

[0075]

[0076] The constraint condition is:

[0077]

[0078] wherein E[·] is an expectation, |·| is a modulus of a complex number, ‖·‖ is a two-norm of a vector, (·) H is a conjugate transpose of a vector, h k ∈C N×1 is a downlink channel between a base station and a user k, W=[w1,w2…w k ]∈C N×K is a precoding matrix, N and K are respectively a number of antennas at a base station end and a number of users at a receiving end, w k ∈C N×1 is a precoding vector of the kth user, q∈C N×1 is a precoded transmission signal, a kis the receiver coefficient configured by the kth user, is a complex Gaussian noise, is the noise power, and P is the base station end transmission power constraint.

[0079] It should be noted that the foregoing explanation and description of the embodiment of the large-scale MIMO deep unfolding precoding method based on the conjugate gradient method also applies to the embodiment of the large-scale MIMO deep unfolding precoding device based on the conjugate gradient method, which will not be repeated here.

[0080] The large-scale MIMO deep unfolding precoding device based on the conjugate gradient method according to the embodiment of the present application can effectively estimate the receiving coefficient at each user of the receiving end by using the neural network trained and constructed by the deep learning method for the multi-user large-scale MIMO wireless communication system. The internal structure of the conjugate gradient MMSE precoding algorithm is combined with the advanced DNN network, and a model-driven supervised neural network is constructed by using deep unfolding. Compared with the data-driven "black box" DNN network, the network has better generalization ability and interpretability. While ensuring the system performance, the computational complexity of the traditional MMSE precoding is significantly reduced, which is convenient for engineering implementation.

[0081] Figure 6 The electronic device provided by the embodiment of the present application is shown in the structural schematic diagram of the electronic device. The electronic device can include:

[0082] The memory 601, the processor 602, and the computer program stored in the memory 601 and executable on the processor 602.

[0083] The processor 602 implements the large-scale MIMO deep unfolding precoding method based on the conjugate gradient method provided in the above embodiments when executing the program.

[0084] Further, the electronic device further includes:

[0085] The communication interface 603 is used for communication between the memory 601 and the processor 602.

[0086] The memory 601 is used to store the computer program executable on the processor 602.

[0087] The memory 601 can include a high-speed RAM memory, and can also include a non-volatile memory, such as at least one disk memory.

[0088] If the memory 601, the processor 602 and the communication interface 603 are implemented independently, the communication interface 603, the memory 601 and the processor 602 can be connected with each other through a bus and complete communication between each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, Figure 6 Only one thick line is used in the figure to represent the bus, but it does not mean that there is only one bus or only one type of bus.

[0089] Optionally, in a specific implementation, if the memory 601, the processor 602 and the communication interface 603 are integrated on a chip, the memory 601, the processor 602 and the communication interface 603 can complete communication between each other through an internal interface.

[0090] The processor 602 can be a Central Processing Unit (CPU), or an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present application.

[0091] The embodiment also provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to implement the large-scale MIMO deep expansion precoding method based on the conjugate gradient method.

[0092] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example" or "some examples" means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or N embodiments or examples in a suitable manner. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the present specification and the features of the different embodiments or examples without contradiction.

[0093] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0094] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of the invention pertain.

[0095] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0096] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it includes one or a combination of the steps of the method embodiments.

Claims

1. A deep unrolling precoding method for large-scale MIMO based on the conjugate gradient method, characterized in that, Includes the following steps: Obtain the downlink air interface channel vector and transmit vector; The downlink air interface channel vector and the transmit vector are input into a pre-trained nonlinear neural network to obtain a pre-coded transmit signal; wherein, the nonlinear neural network is a combination of a deep unfolded multilayer nonlinear subnetwork with the step size in the conjugate gradient algorithm as the training parameter and a receiver single-layer linear subnetwork with the receiver coefficients as the training parameter, and is trained using the downlink air interface channel vector and the transmit vector training set. The input to the deeply expanded multi-layer nonlinear subnetwork is the downlink air interface channel vector and the transmission vector training set, and the output is the pre-coded transmission signal. Each layer in the subnetwork corresponds to one iteration of the conjugate gradient algorithm. A normalization layer is added after the deeply expanded multi-layer nonlinear subnetwork. The input to the single-layer linear subnetwork at the receiver is the output of the normalization layer, used to generate receiver coefficients configured for each user. Before inputting the downlink air interface channel vector and the transmission vector into the pre-trained nonlinear neural network, the following steps are also included: Construct the training sets for the downlink air interface channel vector and transmit vector; The nonlinear neural network is trained using the downlink air interface channel vector and transmit vector training set to obtain the step size parameter of the nonlinear neural network and the receiver coefficients configured for each user; The parameters of the nonlinear neural network are updated using the step size parameter and the receiver coefficients configured for each user to obtain the pre-trained nonlinear neural network. The loss objective function for training the nonlinear neural network using the downlink air interface channel vector and transmit vector training set is: The constraints are: Where E[·] represents the expectation, |·| represents the modulus of the complex number, and ‖·‖ represents the vector 2 norm. H h is the conjugate transpose of a vector. k Let W be the downlink channel between the base station and user k, where W = [w1, w2, ... wk]. k [ ] represents the precoding matrix, N and K are the number of antennas at the base station and the number of users at the receiver, respectively, w k Let q be the precoded vector of the k-th user, and a be the precoded transmitted signal. k The receiver coefficients configured for the k-th user. Let P be the noise power, and P be the base station transmit power constraint.

2. A large-scale MIMO deep unrolling precoding device based on the conjugate gradient method, characterized in that, include: The acquisition module is used to acquire the downlink air interface channel vector and transmit vector; A precoding module is used to input the downlink air interface channel vector and the transmit vector into a pre-trained nonlinear neural network to obtain a pre-coded transmit signal. The nonlinear neural network is a combination of a deeply expanded multilayer nonlinear subnetwork trained using the step size of the conjugate gradient algorithm as a training parameter and a receiver-level single-layer linear subnetwork trained using receiver coefficients as training parameters, trained using a training set of the downlink air interface channel vector and the transmit vector. The input to the deeply expanded multilayer nonlinear subnetwork is the downlink air interface channel vector and the transmit vector training set, and the output is the pre-coded transmit signal. Each layer in the subnetwork corresponds to one iteration of the conjugate gradient algorithm. A normalization layer is added after the deeply expanded multilayer nonlinear subnetwork. The input to the receiver-level single-layer linear subnetwork is the output of the normalization layer, used to generate receiver coefficients configured for each user. The device further includes: The training module is used to construct a training set of the downlink air interface channel vector and the transmission vector before inputting the downlink air interface channel vector and the transmission vector into the pre-trained nonlinear neural network; to train the nonlinear neural network using the training set of the downlink air interface channel vector and the transmission vector to obtain the step size parameter of the nonlinear neural network and the receiver coefficients configured for each user; and to update the parameters of the nonlinear neural network using the step size parameter and the receiver coefficients configured for each user to obtain the pre-trained nonlinear neural network. The loss objective function for training the nonlinear neural network using the downlink air interface channel vector and transmit vector training set is: The constraints are: Where E[·] represents the expectation, |·| represents the modulus of the complex number, and ‖·‖ represents the vector 2 norm. H h is the conjugate transpose of a vector. k Let W be the downlink channel between the base station and user k, where W = [w1, w2, ... wk]. k [ ] represents the precoding matrix, N and K are the number of antennas at the base station and the number of users at the receiver, respectively, w k Let q be the precoded vector of the k-th user, and a be the precoded transmitted signal. k The receiver coefficients configured for the k-th user. Let P be the noise power, and P be the base station transmit power constraint.

3. An electronic device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, the processor executing the program to implement the large-scale MIMO deep unrolling precoding method based on the conjugate gradient method as described in claim 1.

4. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the large-scale MIMO deep unrolling precoding method based on the conjugate gradient method as described in claim 1.

Images