Precoding Matrix Generation Method and Device for Massive MIMO System

Abstract

The present application proposes a precoding matrix generation method for large-scale MIMO systems, including: obtaining base station data and user terminal data, and inputting them into a trained precoding matrix design network to output a precoding matrix corresponding to each user. The precoding matrix design network is a multi-level network. After inputting the data, the calculation process of each level of the network is as follows: calculating the expectation of the second-order term about the channel by using the base station data and user terminal data, and approximately calculating the expectation of the inverse matrix of the second-order term, where the expectation of the inverse matrix of the second-order term includes a learnable matrix; approximately calculating the expectation of the high-order term about the channel by using the expectation of the inverse matrix of the second-order term, where the expectation of the high-order term includes a learnable matrix; approximately calculating the precoding matrix based on the expectation of the high-order term of the channel according to a closed-form expression. The present invention adopting the above solution can achieve a user ergodic sum rate close to the optimal with less computational overhead and a faster convergence rate.

Description

Technical Field

[0001] This application relates to the field of wireless communication technologies, and in particular, to a method and apparatus for generating a precoding matrix for a large-scale MIMO system. Background Art

[0002] Precoding is a key technology in large-scale MIMO systems to overcome severe path loss, noise pollution, and inter-user interference problems during transmission. However, in large-scale MIMO systems, due to limited pilot resources and severe channel aging, the transmitter cannot obtain accurate channel state information (CSI), which poses higher requirements for the robustness of precoding design. Additionally, when designing precoding, the requirements of the actual system for the algorithm execution time and computational overhead also need to be considered.

[0003] The inaccurate CSI at the transmitter is widely modeled as a Gaussian distributed posterior channel model. Based on this model, the robust precoder design can be expressed as a problem of maximizing the ergodic weighted sum rate (EWSR) of users under the total power constraint. There are two ways to solve this problem: one is the method based on iterative optimization, such as iteratively obtaining the deterministic equivalent of the EWSR and then using majorize-minimization (MM) to optimize the lower bound of the EWSR, or generating some instantaneous CSI samples according to the distribution of the channel posterior estimate, and then using the method of stochastic weighted minimum mean-squared-error (SWMMSE) to iteratively optimize the EWSR; the other is to develop deep learning algorithms to design robust precoding, such as using a black-box neural network to learn the low-dimensional features of the robust precoder.

[0004] However, the related technologies for the first approach have the characteristics of large computational overhead and slow convergence speed, which hinders their application in actual systems; although the related technologies for the second approach have relatively small computational overhead, their performance and interpretability are generally poor. Summary of the Invention

[0005] This application aims to at least solve one of the technical problems in the related technologies to some extent.

[0006] To this end, the first objective of this application is to propose a method for generating a precoding matrix for a large-scale MIMO system, which achieves an almost optimal ergodic sum rate of users with less computational overhead and a faster convergence speed.

[0007] The second object of this application is to propose a precoding matrix generation device for large-scale MIMO systems.

[0008] To achieve the above object, an embodiment of the first aspect of this application proposes a precoding matrix generation method for large-scale MIMO systems. The large-scale MIMO system includes a base station and user terminals. The method includes: obtaining base station data and user terminal data, and inputting them into a trained precoding matrix design network to output a precoding matrix corresponding to each user. Among them, the precoding matrix design network is a multi-level network. After inputting the data, the calculation process of each level of the network is as follows: calculating the expectation of the second-order term about the channel by using the base station data and user terminal data, and approximately calculating the expectation of the inverse matrix of the second-order term, where the expectation of the inverse matrix of the second-order term includes a learnable matrix; approximately calculating the expectation of the high-order term about the channel by using the expectation of the inverse matrix of the second-order term, where the expectation of the high-order term includes a learnable matrix; approximately calculating the precoding matrix based on the expectation of the high-order term of the channel according to a closed-form expression.

[0009] The precoding matrix generation method for large-scale MIMO systems according to the embodiments of this application combines the advantages of optimization-based methods and deep learning methods. By using the mean and variance of channel posterior estimation, it achieves a near-optimal user ergodic sum rate with less computational overhead and faster convergence speed.

[0010] Optionally, in an embodiment of this application, the base station data includes the base station transmit power, and the user terminal data includes the channel estimation mean, channel estimation variance, and average noise power.

[0011] Optionally, in an embodiment of this application, in the i-th level network, the second-order term is expressed as:

[0012]

[0013] where H k represents the channel of user terminal k, represents the conjugate transpose matrix of H k , represents the precoding matrix of user terminal m output by the (i - 1)-th layer network, P max represents the base station transmit power, represents the average noise power at user terminal m, and I represents the identity matrix;

[0014] The expectation of the second-order term is expressed as:

[0015]

[0016] where the operation vec(·) represents the vectorization of a matrix, represents the matrix vectorization, and the matrix matrix Q k the (i, j)-th element of is expressed as:

[0017]

[0018] wherein, represents the mean of the channel estimation of user terminal k, and ΔH k represents the variance of the channel estimation of user terminal k, and the subscript r2 = i % 64, t2 = j % 2.

[0019] Optionally, in an embodiment of the present application, the expectation of the inverse matrix of the second-order term is expressed as:

[0020]

[0021] wherein, is the first-order Taylor expansion of is used to compensate for and the error between is expressed as:

[0022]

[0023] wherein, [·] + operates to replace the diagonal elements of a matrix with the reciprocal of the element and set the other elements of the matrix to 0.

[0024] Optionally, in an embodiment of the present application, the expectation of the high-order term regarding the channel is expressed as:

[0025]

[0026] wherein, the operator is a learnable matrix introduced to compensate for the high-order expectation approximation error.

[0027] Optionally, in an embodiment of the present application, the closed-form expression is expressed as:

[0028]

[0029] wherein, represents the average noise power at user terminal m.

[0030] Optionally, in an embodiment of the present application, approximating the precoding matrix based on the expectation of the high-order term of the channel according to the closed-form expression includes:

[0031] Calculate the symmetric matrix B i and its inverse matrix, where

[0032] Determine the precoding matrix according to the inverse matrix of the calculated symmetric matrix where

[0033] Perform power scaling on the precoding matrix to obtain the final precoding matrix, where the power scaling is expressed as:

[0034]

[0035] where

[0036] Optionally, in an embodiment of the present application, the training process of the precoding matrix design includes:

[0037] Generate several groups of instantaneous channel samples as training data according to the distribution of a group of beam delay domain channel estimates, and train the learnable matrix using the stochastic gradient descent method. When training, the complex number of the ergodic sum rate of all users is used as the loss function.

[0038] Optionally, in an embodiment of the present application, the loss function is expressed as:

[0039]

[0040] where S represents the number of groups of instantaneous channel samples, and ω k represents the weight of the k-th user.

[0041] To achieve the above object, an embodiment of the second aspect of the present invention proposes a precoding matrix generation device for a large-scale MIMO system. The large-scale MIMO system includes a base station and user terminals. The device includes a data acquisition module and a precoding matrix generation module, where

[0042] The data acquisition module is used to acquire base station data and user terminal data,

[0043] The precoding matrix generation module is used to input the acquired data into the trained precoding matrix design network and output the precoding matrix corresponding to each user,

[0044] where the precoding matrix design network is a multi-level network. After inputting the data, the calculation process of each level of the network is:

[0045] Calculate the expectation of the second-order term about the channel using the base station data and user terminal data, and approximately calculate the expectation of the inverse matrix of the second-order term, where the expectation of the inverse matrix of the second-order term includes the learnable matrix;

[0046] Approximately calculate the expectation of the high-order terms regarding the channel by using the expectation of the inverse matrix of the second-order terms, where the expectation of the high-order terms includes a learnable matrix;

[0047] Approximately calculate the precoding matrix based on the expectation of the high-order terms of the channel according to a closed-form expression.

[0048] Additional aspects and advantages of the present application will be given in part in the following description, become apparent in part from the following description, or be learned through the practice of the present application. Description of the Drawings

[0049] The above and / or additional aspects and advantages of the present application will become apparent and easy to understand from the following description of the embodiments in conjunction with the drawings, where:

[0050] Figure 1 It is a schematic flowchart of a precoding matrix generation method for a large-scale MIMO system provided in Embodiment 1 of the present application;

[0051] Figure 2 It is a scenario diagram of an embodiment of the present application;

[0052] Figure 3 It is a network architecture diagram of a low-complexity robust precoding deep learning design for a large-scale MIMO system in an embodiment of the present application;

[0053] Figure 4 It is a comparison diagram of the user sum rate performance of the low-complexity robust precoding design method for a large-scale MIMO system in an embodiment of the present application with the traditional SWMMSE algorithm and WMMSE algorithm under different user moving speed scenarios;

[0054] Figure 5 It is a comparison diagram of the user sum rate performance of the low-complexity robust precoding design method for a large-scale MIMO system in an embodiment of the present application with the traditional SWMMSE algorithm and WMMSE algorithm under different time blocks;

[0055] Figure 6 It is a comparison diagram of the user sum rate performance of the low-complexity robust precoding design method for a large-scale MIMO system in an embodiment of the present application with the traditional SWMMSE algorithm and WMMSE algorithm under different signal-to-noise ratios (SNR);

[0056] Figure 7 It is a comparison diagram of the computational complexity of the algorithm in an embodiment of the present application with the traditional SWMMSE algorithm and WMMSE algorithm;

[0057] Figure 8Schematic structural diagram of a precoding matrix generation device provided for an embodiment of the present application for a large-scale MIMO system. Detailed implementation manners

[0058] The embodiments of the present application will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present application and should not be construed as a limitation to the present application.

[0059] The precoding matrix generation method and device for a large-scale MIMO system according to an embodiment of the present application will be described below with reference to the accompanying drawings.

[0060] Figure 1 Schematic flow diagram of a precoding matrix generation method for a large-scale MIMO system provided for Embodiment 1 of the present application. The large-scale MIMO system includes a base station and user terminals.

[0061] As Figure 1 shown, the precoding matrix generation method for a large-scale MIMO system includes the following steps:

[0062] Step 101: Obtain base station data and user terminal data, and input them into a trained precoding matrix design network to output a precoding matrix corresponding to each user.

[0063] Among them, the precoding matrix design network is a multi-level network. After inputting the data, the calculation process of each level of the network is as follows:

[0064] Calculate the expectation of the second-order term of the channel using the base station data and user terminal data, and approximately calculate the expectation of the inverse matrix of the second-order term, where the expectation of the inverse matrix of the second-order term includes a learnable matrix.

[0065] Approximately calculate the expectation of the high-order term of the channel using the expectation of the inverse matrix of the second-order term, where the expectation of the high-order term includes a learnable matrix.

[0066] Approximately calculate the precoding matrix based on the expectation of the high-order term of the channel according to a closed-form expression.

[0067] The precoding matrix generation method for a large-scale MIMO system according to an embodiment of the present application combines the advantages of the optimization-based method and the deep learning method, and achieves a near-optimal user ergodic sum rate with less computational overhead and a faster convergence rate by using the mean and variance of the channel posterior estimation.

[0068] Optionally, in an embodiment of the present application, the low-complexity robust precoding design network for a large-scale MIMO system considers a large-scale MIMIO system including a base station and K users. Among them, the base station is equipped with a large-scale antenna array with M t antennas, and the beam domain channels of all K users in the known user set are known Posterior estimation mean and variance Each user is equipped with M r antennas. It can be assumed that the elements of H k are independent and follow a Gaussian distribution, that is where [·] r,t represents the (r, t)-th element of a matrix. The robust precoder design problem can be modeled as the problem P1 of maximizing the ergodic sum rate of all users under the total power constraint:

[0069]

[0070] where

[0071] To solve this problem, the low-complexity robust precoding design network for a large-scale MIMO system adopts a multi-level structure, and uses the channel estimation mean of each user terminal Channel estimation variance Weight Base station transmission power P max and the average noise power at each user terminal and the initialization precoding matrix as inputs

[0072] Optionally, in an embodiment of the present application, the calculation of the second-order term expectation with respect to the channel is specifically as follows: First, calculate the (i, j)-th element Q k of the matrix The calculation expression is

[0073]

[0074] where r2 = i % M t and t2 = j % M r .

[0075] For any matrix Define the operation where vec(·) represents matrix vectorization, represents vector matrixization. In particular, because the beam domain channel is sparse, that is and The elements with larger energy in the middle matrix are concentrated in the columns where B << M t column, so Q k The elements with larger energy in are concentrated in the columns where B 2 column. In this embodiment, when calculating the matrix multiplication of Q k and vec(M), the elements with larger energy in Q k are retained, and the elements with smaller energy are set to 0, which can reduce the computational complexity from to Then, for the second-order term

[0076]

[0077] In this embodiment, the expectations are calculated respectively according to the following formula

[0078]

[0079] Optionally, in an embodiment of the present application, the approximation of the expectation of the inverse of the second-order term of the channel is specifically: using

[0080] to approximate respectively, where are respectively the first-order Taylor expansions of, is a learnable residual matrix used to compensate for the error between and Define [·] + operation as replacing the diagonal elements of a matrix with the reciprocals of the elements and setting the other elements of the matrix to 0, The expression of is:

[0081]

[0082] Optionally, in an embodiment of the present application, the approximation of the expectation of the higher-order term with respect to the channel is specifically:

[0083] Higher-order expectation The expressions of are respectively:

[0084]

[0085] By discarding some cross-correlation terms with smaller energy, can be approximated as:

[0086]

[0087] where, for any matrix operator is a learnable matrix introduced to compensate for the approximation error of the high-order expectation. In particular, since the elements with larger energy in are concentrated in B << M t rows, in this embodiment, when calculating the and the matrix multiplication of vec(L), the elements with larger energy in are retained, and the elements with smaller energy are set to 0, so that the computational complexity can be reduced from to

[0088] Optionally, in an embodiment of the present application, the closed-form expression is expressed as:

[0089]

[0090] where represents the average noise power at user terminal m.

[0091] Optionally, in an embodiment of the present application, based on the expectation of the high-order terms of the channel and approximately calculating the precoding matrix according to the closed-form expression, including:

[0092] First, calculate the symmetric matrix and its inverse matrix. Since B i is sparse, its elements with larger energy are mainly concentrated in the q rows, q columns (q < M t ), and on the diagonal. In this embodiment, the elements with larger energy are retained and the elements with smaller energy are set to zero, then the complexity of (B i ) -1 can be reduced from to O(q 3 + M t ); then, calculate Finally, to meet the base station power constraint, power scaling is performed on the precoding matrix obtained above:

[0093]

[0094] where

[0095] Optionally, in an embodiment of the present application, the offline training process is specifically: each time during training, according to the distribution of a group of beam delay domain channel estimates , S groups of instantaneous channel samples are generated. Then, the negative value of the ergodic sum rate of all users is used as the loss function:

[0096]

[0097] Among them, is the output of the last layer (the L-th layer) of the network. Then, the introduced learnable matrix is trained using the stochastic gradient descent method.

[0098] The specific process of online calculation is as follows: taking the channel estimation mean, channel estimation variance, weight, base station transmission power, transmission signal-to-noise ratio, and initial precoding matrix of each user terminal as inputs, and using the trained learnable matrix to calculate the precoding matrix according to the closed-form expression.

[0099] Next, the precoding matrix generation method for a large-scale MIMO system of the present application will be described in detail through specific embodiments.

[0100] Figure 2 is a schematic diagram of the application scenario of the low-complexity robust precoding design network for a large-scale MIMO system in this embodiment. In this embodiment, the QuaDRiGa channel model is used to generate channel samples and the simulation scenario is set to "3GPP_38.901_UMa_NLOS", with a central frequency of 3.5 GHz. As Figure 3 shown, the number of base station antennas is set to 64, the number of users is 10, the base station location is (0, 0, 25 m), and 10 users with 2 antennas are randomly generated within a range of 200 m centered on the base station at the beginning of each time slot. Two scenarios where the users move eastward at speeds of 30 km / h and 120 km / h are considered. The system adopts the TDD mode, and each 1 ms duration is divided into 1 time slot, and each time slot is further divided into 7 time blocks. It is assumed that the 0th time block of each time slot is used for uplink training, and the base station can obtain accurate instantaneous channel estimates through the uplink pilots sent by the users; the remaining time blocks 1 - 6 are used for downlink transmission. However, due to channel aging, the base station only knows the posterior channel estimates:

[0101]

[0102] Among them, the subscript n represents the n-th time block, ⊙ represents element-wise multiplication, is a non-negative deterministic matrix containing channel amplitude information, the elements of are independent and follow a complex Gaussian distribution with a mean of 0 and a variance of 1, is the time correlation coefficient, and its expression is:

[0103]

[0104] In the actual scenario, the base station can obtain α through the channel sounding process k,n and In this scenario, in this embodiment, α k,n and Ω k, where the mean value of α is 0.94 in the scenario where the user's moving speed is 30 km / h, the mean value of α is 0.77 in the scenario where the user's moving speed is 120 km / h, and α corresponding to each time block in the scenario where the user's moving speed is 120 km / h 1:6 = [0.96, 0.92, 0.84, 0.75, 0.63, 0.49].

[0105] In this scenario, the posterior mean of the channel estimation for each user in each time block is The variance is For simplicity, let the weight of each user be 1. Figure 1 is a low-complexity robust precoding design network for large-scale MIMO systems proposed for this scenario. This network has a multi-level structure. The i-th level network, The specific calculation process is as follows:

[0106] 1. Calculate the expectation of the second-order term with respect to the channel: First, calculate Q according to the following expression k Each element of

[0107]

[0108] where, r2 = i % 64, t2 = j % 2. After statistics, the largest elements in the Q k matrix are generally concentrated in 100 columns. In this embodiment, the 100 columns with the largest energy in Q k are retained and other elements are set to zero. Define the operation where, vec(·) represents matrix vectorization, represents vector matrixization. Then, calculate the expectation according to the following formula

[0109]

[0110] 2. Approximately calculate the expectation of the inverse of the second-order term of the channel: First, calculate according to the following formula

[0111]

[0112] where, [·] + The operator means replacing the diagonal elements of a matrix with the reciprocal of the element and setting other elements of the matrix to 0. Then, introduce the learnable residual matrix and use to approximate

[0113] 3. Approximate calculation of the expectation of the high-order terms regarding the channel: Calculate according to the following formula

[0114]

[0115] where, for any matrix operator

[0116] 4. Specifically, approximate calculation of the precoding matrix according to the closed-form expression: First, calculate the symmetric matrix and its inverse matrix. According to statistical observations, in this scenario, the elements with larger energy in the symmetric B i matrix are mainly concentrated in the q = 30 rows / columns and the diagonal. Retain the elements with larger energy in B i and set the elements with smaller energy to zero. Then, calculate Finally, calculate and scale the precoding matrix according to the following formula:

[0117]

[0118] In this scenario, Quadriga is used to generate channel samples for 200 time slots, where 100 time slots are used for offline training and the other 100 time slots are used for online testing. During each offline training, according to the distribution of a group of beam delay-domain channel estimates generate S groups of instantaneous channel samples Then, take the negative value of the ergodic sum rate of all users as the loss function:

[0119]

[0120] where, is the output of the last layer (the L-th layer) of the network. Then, use the stochastic gradient descent method to train the introduced learnable matrix. During the online calculation process, take the channel estimation mean, channel estimation variance, weight, base station transmit power, transmit signal-to-noise ratio, and initialization precoding matrix of each user terminal as inputs, and use the trained learnable matrix to calculate the precoding matrix according to the closed-form expression.

[0121] Figure 4 Shows the convergence performance of the low-complexity robust precoding design network (PO-WMMSE Net) for large-scale MIMO systems and the traditional SWMMSE and non-robust WMMSE algorithms in scenarios with a signal-to-noise ratio of 20 dB and user moving speeds of 30 km / h and 120 km / h respectively. From Figure 4It can be seen that since the proposed PO-WMMSE Net utilizes the channel estimation error, it outperforms the non-robust WMMSE algorithm in each scenario. Additionally, since the proposed PO-WMMSE Net directly utilizes the posterior distribution information of the channel for calculation, while the SWMMSE algorithm generates samples using the channel posterior distribution for calculation, the convergence speed of the PO-WMMSE Net is significantly better than that of the SWMMSE algorithm.

[0122] Figure 5 It shows the user sum-rate performance on each time block of the low-complexity robust precoding design network (PO-WMMSE Net) for large-scale MIMO systems, the traditional SWMMSE, and the non-robust WMMSE algorithms in the scenario where the signal-to-noise ratio is 20 dB and the user moving speed is 120 km / h respectively. From Figure 5 It can be seen that the user sum-rate performance of the proposed PO-WMMSE algorithm is better than that of the other two baseline algorithms.

[0123] Figure 6 It shows the user sum-rate performance at each signal-to-noise ratio in the scenario where the user moving speeds of the low-complexity robust precoding design network (PO-WMMSE Net) for large-scale MIMO systems, the traditional SWMMSE, and the non-robust WMMSE algorithms are 120 km / h respectively. From Figure 6 It can be seen that at each signal-to-noise ratio, the user sum-rate performance of the proposed PO-WMMSE algorithm is better than that of the other two baseline algorithms. This is because at high signal-to-noise ratios, the performance of the network is mainly affected by interference, and in the case of limited iterations / fewer layers, the proposed network can more effectively mitigate interference.

[0124] Figure 7 It shows the comparison diagram of the computational complexity of the 5-layer low-complexity robust precoding design network (PO-WMMSE Net) for large-scale MIMO systems and the traditional SWMMSE and non-robust WMMSE algorithms with 5 iterations under the system configuration of the above embodiment. From Figure 7 It can be seen that the complexity of the PO-WMMSE algorithm is significantly reduced compared to the other two baseline algorithms. This is because the proposed PO-WMMSE Net greatly reduces the complexity of matrix operations by utilizing the sparsity of the beam-domain channel. Additionally, to achieve the same performance, the number of network layers required by the proposed network is generally significantly less than the number of iterations required by the baseline algorithms, which will further reduce the computational overhead.

[0125] To implement the above embodiment, the present application also proposes a precoding matrix generation device for large-scale MIMO systems.

[0126] Figure 8FIG. 0 is a schematic structural diagram of a precoding matrix generation device for a large-scale MIMO system provided by an embodiment of the present application. The large-scale MIMO system includes a base station and user terminals.

[0127] As Figure 8 shown, the precoding matrix generation device for the large-scale MIMO system includes a data acquisition module and a precoding matrix generation module. Among them,

[0128] The data acquisition module is configured to acquire base station data and user terminal data.

[0129] The precoding matrix generation module is configured to input the acquired data into a trained precoding matrix design network and output a precoding matrix corresponding to each user.

[0130] Among them, the precoding matrix design network is a multi-level network. After inputting the data, the calculation process of each level of the network is as follows:

[0131] Calculate the expectation of the second-order term about the channel using the base station data and user terminal data, and approximately calculate the expectation of the inverse matrix of the second-order term. Among them, the expectation of the inverse matrix of the second-order term includes a learnable matrix.

[0132] Approximately calculate the expectation of the high-order term about the channel using the expectation of the inverse matrix of the second-order term. Among them, the expectation of the high-order term includes a learnable matrix.

[0133] Approximately calculate the precoding matrix based on the expectation of the high-order term of the channel according to the closed-form expression..

[0134] It should be noted that the foregoing explanation of the embodiment of the precoding matrix generation method for the large-scale MIMO system also applies to the precoding matrix generation device for the large-scale MIMO system in this embodiment, and will not be repeated here.

[0135] In the description of this specification, the description with reference to terms such as "one embodiment", "some embodiments", "example", "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 this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, without conflict, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0136] In addition, 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 quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include at least one such feature. In the description of this application, "a plurality of" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0137] Any process or method description represented in a flowchart or described otherwise herein can be understood to represent a module, segment, or portion of code including one or more executable instructions for implementing a customized logical function or process. The scope of the preferred embodiments of this application includes additional implementations, where functions may be executed in a manner substantially simultaneous with or in the reverse order according to the involved functions, rather than in the order shown or discussed, which should be understood by those skilled in the technical field to which the embodiments of this application pertain.

[0138] The logic and / or steps represented in a flowchart or described otherwise herein, for example, can be considered a sequenced list of executable instructions for implementing a logical function, and can be specifically implemented in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection portion with one or more wirings (electronic device), a portable computer diskette (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or other appropriate processing as necessary, and then stored in a computer memory.

[0139] It should be understood that each part of the present application can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one of the following techniques known in the art or a combination thereof can be used: discrete logic circuits with logic gate circuits for implementing logical functions on data signals, application-specific integrated circuits with appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0140] Those of ordinary skill in the art can understand that all or part of the steps carried by the method of implementing the above embodiments can be completed by instructing relevant hardware through a program. The said program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

[0141] In addition, in each embodiment of the present application, each functional unit can be integrated in a processing module, or each unit can exist physically alone, or two or more units can be integrated in a module. The above integrated module can be implemented in the form of hardware or in the form of a software functional module. When the above integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0142] The above-mentioned storage medium can be a read-only memory, a magnetic disk, an optical disc, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present application. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present application.

Claims

1. A precoding matrix generation method for a large-scale MIMO system, characterized in that The massive MIMO system includes a base station and a user terminal, and the method includes: Obtain base station data and user terminal data, input them into the trained precoding matrix design network, and output the precoding matrix corresponding to each user. The precoding matrix design network is a multi-level network. After inputting data, the calculation process of each level of the network is: Calculate the expectation of the second-order term about the channel by using the base station data and the user terminal data, and approximately calculate the expectation of the inverse matrix of the second-order term, wherein the expectation of the inverse matrix of the second-order term includes a learnable matrix; Calculate the expectation of the high-order terms about the channel by using the expectation of the inverse matrix of the second-order terms, wherein the expectation of the high-order terms includes a learnable matrix; Approximately calculating a precoding matrix based on expectations of high-order terms of the channel according to a closed-form expression; The base station data includes base station transmit power, and the user terminal data includes channel estimation mean, channel estimation variance and average noise power; In the i-th level network, the second-order term is expressed as: Among them, H k represents the channel of user terminal k, represents H k 's conjugate transpose matrix, represents the precoding matrix of user terminal m output by the (i - 1)-th layer network, P max represents the base station transmission power, represents the average noise power at user terminal m, and I represents the identity matrix, represents the user set; Second-order term is expressed as the expectation: Among them, the operation vec(·) represents matrix vectorization, represents vector matrixization, and the matrix M t indicates that the base station is equipped with a massive MIMO array with M t antennas, and M r indicates that each user is equipped with M r antennas. The (i, j)-th element of the matrix Q k is expressed as: is expressed as: Among them, represents the mean of the channel estimation of user terminal k, ΔH k represents the variance of the channel estimation of user terminal k, and the subscript r2 = i % 64, t2 = j % 2; The expectation of the inverse matrix of the second-order term is expressed as: Among them, is the first-order Taylor expansion of used to compensate for and the error between is expressed as: Among them, [·] + The operation is to replace the diagonal elements of a matrix with the reciprocals of the elements and set the other elements of the matrix to 0; The high-order terms about the channel are expected to be expressed as: Among them, the operator is a learnable matrix introduced to compensate for the approximation error of the high-order expectation; The closed-form expression is expressed as: The method of approximately calculating the precoding matrix based on the expectation of the high-order terms of the channel and according to a closed-form expression comprises: Calculate the symmetric matrix B i and its inverse matrix, where Determine the precoding matrix according to the inverse matrix of the calculated symmetric matrix wherein The precoding matrix is power scaled to obtain a final precoding matrix, wherein the power scaling is expressed as: Among them, 2. The method according to claim 1, wherein The training process of the precoding matrix design includes: According to the distribution of a set of beam delay domain channel estimates, several groups of instantaneous channel samples are generated as training data, and the learnable matrix is trained using the stochastic gradient descent method. During training, the complex numbers of the traversals and rates of all users are used as the loss function.

3. The method according to claim 2, characterized in that, The loss function is expressed as: where S represents the number of groups of instantaneous channel samples, and ω k represents the weight of the k-th user.

4. A precoding matrix generation device for a large-scale MIMO system, characterized in that, The device implements the precoding matrix generation method for a large-scale MIMO system according to claim 1, wherein the large-scale MIMO system includes a base station and a user terminal, and the device includes a data acquisition module and a precoding matrix generation module, wherein: The data acquisition module is used to acquire base station data and user terminal data. The precoding matrix generation module is used to input the acquired data into the trained precoding matrix design network and output the precoding matrix corresponding to each user. The precoding matrix design network is a multi-level network. After inputting data, the calculation process of each level of the network is: Calculate the expectation of the second-order term about the channel by using the base station data and the user terminal data, and approximately calculate the expectation of the inverse matrix of the second-order term, wherein the expectation of the inverse matrix of the second-order term includes a learnable matrix; Calculating the expectation of the high-order terms about the channel by using the expectation of the inverse matrix of the second-order terms, wherein the expectation of the high-order terms includes a learnable matrix; The precoding matrix is approximately calculated according to a closed-form expression based on the expectation of the high-order terms of the channel.

Images