Precoding gain estimation method, storage medium and system

Abstract

This invention discloses a precoding gain estimation method, device, storage medium, computer product, and system. It includes: an EDU determining a compressed channel matrix and a large-scale fading coefficient matrix; a UCDU determining an estimated value of the precoding gain based on the large-scale fading coefficient matrix; the UCDU processing and fusing the estimated values ​​of the compressed channel matrix and precoding gain using a residual-aware fusion neural network, outputting an estimated value of the residual between the precise value of the precoding gain and the estimated value of the precoding gain; and the UCDU determining the final precoding gain based on the estimated value of the precoding gain and the estimated value of the residual. This method uses a residual-aware fusion neural network that has learned the residual between the precise value of the precoding gain and the estimated value of the precoding gain to output an estimated value of the residual, thereby determining the final precoding gain, which can improve the computational efficiency of the precoding gain.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and in particular to a precoding gain estimation method, storage medium and system. Background Technology

[0002] In mobile communication networks, Link Adaptation (LA) is a key technology for ensuring transmission reliability and spectral efficiency. Its core is the dynamic selection of the Modulation and Coding Scheme (MCS) based on channel conditions. In downlink transmission of non-cellular massive MIMO systems, the ideal approach to achieve optimal performance is for the L2 layer (data link layer in the communication protocol stack) to obtain the accurate precoding matrix calculated by the L1 layer (physical layer in the communication protocol stack), then calculate the PostSINR after transmission, and finally determine the optimal Modulation and Coding Scheme (MCS).

[0003] However, due to the layered processing and timing constraints of the protocol stack, the L1 layer takes a long time to calculate the precoding matrix using existing precoding algorithms, while the L2 layer has high requirements for real-time scheduling. As a result, the L2 layer cannot wait for the L1 layer to calculate the precoding matrix after scheduling and pairing. It must determine the MCS before the L1 layer calculates the precoding matrix. This makes it impossible for the L2 layer to use the precoding matrix calculated by the L1 layer to calculate the post-precoding signal-to-interference-plus-noise ratio (PostSINR) of the user's received signal, thus affecting the link adaptive performance. Summary of the Invention

[0004] This invention provides a precoding gain estimation method, device, storage medium, computer product, and system to solve the problem that calculating the precoding matrix using existing precoding algorithms in the L1 layer takes a long time.

[0005] According to one aspect of the present invention, a precoding gain estimation method is provided, performed by an edge distributed unit (EDU) and a user-centric distributed unit (UCDU) in a downlink system of a non-cellular radio access network, comprising: The EDU determines the compressed channel matrix and the large-scale fading coefficient matrix; The UCDU determines the estimated value of the precoding gain based on the large-scale fading coefficient matrix; The UCDU processes and fuses the estimated values ​​of the compressed channel matrix and the precoding gain through the residual sensing fusion neural network, and outputs an estimated value of the residual difference between the accurate value of the precoding gain and the estimated value of the precoding gain. The UCDU determines the final precoding gain based on the estimated value of the precoding gain and the estimated value of the residual; The residual-aware fusion neural network is a neural network that measures the residual between the learned precise value of the precoding gain and the estimated value of the precoding gain.

[0006] According to another aspect of the present invention, a precoding gain estimation method is provided, performed by a distributed unit (DU) in a downlink system of a noncellular radio access network, the method comprising: The DU determines the compressed channel matrix and the large-scale fading coefficient matrix; The DU determines the estimated value of the precoding gain based on the large-scale fading coefficient matrix; The DU processes and fuses the estimated values ​​of the compressed channel matrix and the precoding gain through a residual-aware fusion neural network, and outputs an estimated value of the residual difference between the precise value of the precoding gain and the estimated value of the precoding gain. The DU determines the final precoding gain based on the estimated value of the precoding gain and the estimated value of the residual; The residual-aware fusion neural network is a neural network that measures the residual between the learned precise value of the precoding gain and the estimated value of the precoding gain.

[0007] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, which is then executed by the at least one processor to enable the at least one processor to perform the precoding gain estimation method according to any embodiment of the present invention.

[0008] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the precoding gain estimation method according to any embodiment of the present invention.

[0009] According to another aspect of the present invention, a computer program product is provided, the computer program product comprising a computer program that, when executed by a processor, implements the precoding gain estimation method according to any embodiment of the present invention.

[0010] According to another aspect of the present invention, a cellular wireless access network downlink system is provided, comprising a central unit, a user-centric distributed unit (UCDU), an edge distributed unit (EDU), and a remote radio frequency unit. The UCDU and the EDU perform the precoding gain estimation method described in one aspect of the present invention.

[0011] According to another aspect of the present invention, a non-cellular wireless access network downlink system is provided, comprising a central unit, a distributed unit (DU), and a remote radio frequency unit; The DU performs the precoding gain estimation method described in another aspect of the present invention.

[0012] The technical solution of this invention solves the problem of long computation time when calculating the precoding matrix using existing precoding algorithms by learning the residual between the accurate value of the precoding gain and the estimated value of the precoding gain through a perceptual fusion neural network, thus achieving the beneficial effect of improving the computational efficiency of the precoding gain.

[0013] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 A schematic diagram of the non-cellular wireless access network downlink system provided by the present invention; Figure 2 This is a flowchart illustrating a precoding gain estimation method provided in Embodiment 1 of the present invention; Figure 3 This is a flowchart illustrating a precoding gain estimation method provided in Embodiment 2 of the present invention; Figure 4 This is a flowchart illustrating a precoding gain estimation method provided in Embodiment 3 of the present invention; Figure 5 A flowchart illustrating a precoding gain estimation method provided in a specific embodiment of the present invention; Figure 6 This is a flowchart illustrating a precoding gain estimation method provided in Embodiment 4 of the present invention; Figure 7This is a schematic diagram of the structure of the electronic device provided in Embodiment 7 of the present invention. Detailed Implementation

[0016] To enable those skilled in the art to better understand the present invention, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention. It should be understood that the various steps described in the method embodiments of the present invention can be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0017] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0018] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0019] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0020] The names of the messages or information exchanged between the multiple devices in the embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0021] Before listing the embodiments provided by the present invention, it is necessary to introduce the system model. Figure 1This is a schematic diagram of the Cell-Free Radio Access Network (CF-RAN) downlink system provided by the present invention, as shown below. Figure 1 As shown, the system includes a central unit, which is the core control node of the network architecture. It is primarily responsible for centralized resource management and protocol processing, as well as the work of the user-centric distributed unit (UCDU), enabling global network control and scheduling. The UCDU is responsible for centralized baseband processing and scheduling, and coordinates the edge distributed units (EDUs). The EDU handles some baseband processing functions and connects to L remote radio units (RRUs), each equipped with M antennas. This system can simultaneously provide downlink transmission services to K single-antenna user equipment (UEs).

[0022] The global channel matrix is ​​a mathematical matrix that describes the channel characteristics between all RRUs and all UEs. Assume the global channel matrix is... ,in, Indicates the first UE to all The channel vector of the root antenna, i.e. .

[0023] If the system uses a linear precoding algorithm, such as the zero-forcing (ZF) precoding algorithm, its precoding matrix... It can be represented as: In the above formula, This represents the power normalization factor that guarantees the transmit power meets the constraints. The demodulated signal-to-interference-plus-noise ratio (PostSINR) of the received signal for a UE can be approximately expressed as: in, It is the first The noise power of each UE, therefore, the precoding gain Accurate estimation is crucial for calculating PostSINR and thus determining the optimal modulation and coding scheme.

[0024] Example 1 Figure 2 This is a flowchart illustrating a precoding gain estimation method provided in Embodiment 1 of the present invention. This method is applicable to downlink transmission in cellular massive MIMO systems and can be executed by the edge distributed unit (EDU) and the user-centric distributed unit (UCDU) in the downlink system of the cellular radio access network.

[0025] like Figure 2 As shown, the precoding gain estimation method provided in Embodiment 1 of the present invention includes the following steps: S110 and EDU determine the compressed channel matrix and the large-scale fading coefficient matrix.

[0026] In this embodiment, the EDU compresses the global channel matrix using a data-driven branch network to output a compressed channel matrix; the EDU then calculates the large-scale fading coefficient matrix based on the global channel matrix.

[0027] In this embodiment, after the EDU calculates the compressed channel matrix and the large-scale fading coefficient matrix, it can send the obtained compressed channel matrix and large-scale fading coefficient matrix to the UCDU through the forward link.

[0028] S120 and UCDU determine the estimated value of the precoding gain based on the large-scale fading coefficient matrix.

[0029] The estimated value of the precoding gain is based on The fast approximate calculation formula is obtained. The fast approximation calculation formula is obtained by simplifying the exact expression of the precoding gain of the linear precoding algorithm based on the channel asymptotic orthogonality principle of Massive MIMO. For example, the linear precoding algorithm may include the ZF precoding algorithm, the regularized zero-forcing (RZF) precoding algorithm, and the minimum mean square error (MMSE) precoding algorithm.

[0030] S130 and UCDU process and fuse the estimated values ​​of the compressed channel matrix and the precoding gain through the residual-aware fusion neural network, and output the estimated value of the residual difference between the accurate value of the precoding gain and the estimated value of the precoding gain.

[0031] Among them, the residual-aware fusion neural network is a neural network that measures the residual between the learned precise value of the precoded gain and the estimated value of the precoded gain.

[0032] The residual-aware fusion neural network is a dual-branch network consisting of an expert knowledge branch network and a data-driven branch network, which perform corresponding calculations in parallel. The expert knowledge branch network utilizes the asymptotic orthogonality of the channels in non-cellular massive MIMO to perform fast precoding gain estimation. Estimation results of data-driven branch network learning expert knowledge branch network residual .

[0033] The data-driven branch network includes a feature extractor and an encoder with shared weights. The feature extractor can be a convolutional neural network (CNN), a multilayer perceptron (MLP) network structure, or other network structures (such as a graph neural network (GNN)). The encoder can also be a CNN, an MLP network structure, or other network structures. The expert knowledge branch network includes an encoder, which can also be a CNN, an MLP network structure, or other network structures. Both the data-driven branch network and the expert knowledge branch network require training.

[0034] In this embodiment, complex precoding gain is used. The estimate is transformed into a more learnable residual. It is estimated that precise pre-coded gain can be learned using lightweight neural networks (i.e., residual-aware fusion neural networks). Compared with an estimate of the precoding gain that is quickly calculated based on prior communication knowledge. residuals between The final estimated precoding gain is ,in, This represents the residual estimate output by the network. In this embodiment, the residual-aware fusion neural network is learned in an end-to-end manner, and the loss function is: In the above formula, Indicates the number of training samples. Indicates the first The estimated value of each residual. Indicates the first The precise value of the precoding gain. Indicates the first An estimate of the precoding gain.

[0035] After the residual perception fusion neural network is trained, it is deployed separately on EDU and UCDU.

[0036] In this embodiment, the UCDU can perform feature encoding on the compressed channel matrix through the data-driven branch network in the residual-aware fusion neural network, and perform feature encoding on the estimated value of the precoding gain through the expert knowledge branch network in the residual-aware fusion neural network; the UCDU can perform feature fusion on the feature vectors output by the expert knowledge branch network and the data-driven branch network through the fusion network in the residual-aware fusion neural network, and perform residual estimation on the fused feature vector to obtain the residual estimate value.

[0037] S140 and UCDU determine the final precoding gain based on the estimated value of the precoding gain and the estimated value of the residual.

[0038] In this method, UCDU uses the sum of the estimated residual and the estimated precoding gain as the final estimated precoding gain, based on the fusion network in the residual-aware fusion neural network.

[0039] The present invention provides a precoding gain estimation method in the following steps: First, the EDU determines the compressed channel matrix and the large-scale fading coefficient matrix; then, the UCDU determines the estimated value of the precoding gain based on the large-scale fading coefficient matrix; subsequently, the UCDU processes and fuses the estimated value of the compressed channel matrix and the precoding gain through a residual-aware fusion neural network, and outputs an estimated value of the residual between the precise value of the precoding gain and the estimated value of the precoding gain; finally, the UCDU determines the final precoding gain based on the estimated value of the precoding gain and the estimated value of the residual; wherein, the residual-aware fusion neural network is a neural network that has learned the residual between the precise value of the precoding gain and the estimated value of the precoding gain. The above method transforms the complex precoding gain estimation problem into a more learnable residual estimation problem. This method only needs to calculate the estimated value of the precoding gain and output the estimated value of the residual through a residual-aware fusion neural network. Compared with the existing technology that uses the precoding algorithm to accurately calculate the precoding gain, it can effectively improve the computational efficiency. Compared with only calculating the estimated value of the precoding gain, the estimation accuracy of this method is significantly better than that of the estimated value of the precoding gain, achieving the best balance between computational accuracy and computational efficiency.

[0040] Example 2 Figure 3 This is a flowchart illustrating a precoding gain estimation method according to Embodiment 2 of the present invention. Embodiment 2 is an optimization based on the above embodiments. For details not covered in this embodiment, please refer to Embodiment 1.

[0041] like Figure 3 As shown in Embodiment 2 of the present invention, a precoding gain estimation method includes the following steps: S210 and EDU use the data-driven branch network in the residual sensing fusion neural network to compress the channel state information of the global channel matrix and output a compressed channel matrix.

[0042] Specifically, the EDU compresses the channel state information of the global channel matrix and outputs a compressed channel matrix through the data-driven branch network, including: the EDU separating the real and imaginary parts of the global channel matrix and reshaping them into a tensor through the data-driven branch network; and the EDU compressing the tensor using a shared-weight feature extractor through the data-driven branch network and outputting the compressed channel matrix.

[0043] The feature extractor can be a multilayer perceptron (MLP), a graph neural network (GNN), a convolutional neural network (CNN), or a Transformer layer, etc. The input is a 3D channel vector, and the output is a compressed channel matrix, which is a scalar. This operation applies to all... UE and Each RRU shares parameters and outputs a compressed channel matrix. This method transmits the compressed channel matrix to the UCDU, which, compared to existing technologies that transmit the global channel matrix to the UCDU, reduces the amount of data transmitted from [previous method]. Reduce to This greatly reduces the overhead of the fronthaul link.

[0044] S220 and EDU calculate the large-scale fading coefficient matrix based on the global channel matrix.

[0045] Specifically, the EDU calculates the large-scale fading coefficient matrix based on the global channel matrix, including: the EDU calculates the square of the modulus of the global channel vector in the global channel matrix and divides it by the number of antennas to obtain the large-scale fading coefficient matrix.

[0046] Among them, when the RRU is equipped with the number of antennas When it is large, there is , Indicates the first The user to the Channel vectors of RRUs Indicates the first The user to the The channel vectors of each RRU. Based on the global channel matrix. It can be calculated The large-scale fading matrix is ​​obtained. .

[0047] S230 and UCDU determine the estimated value of the precoding gain based on the large-scale fading coefficient matrix.

[0048] Specifically, the UCDU determines the estimated value of the precoding gain based on the large-scale fading coefficient matrix, including: the UCDU calculates the estimated value of the precoding gain according to the fast approximation calculation formula of the precoding gain, the fast approximation calculation formula of the precoding gain is determined based on the channel asymptotic orthogonality of large-scale MIMO and the accurate calculation formula of the precoding gain corresponding to different linear precoding algorithms.

[0049] The following is a detailed explanation using the zero-forcing ZF precoding algorithm as an example: ZF precoding algorithm, power normalization factor The power constraints for each RRU must be met: ,in, It is the first The precoding submatrix of each RRU can be represented as ,in express The UE to the first Substituting the channel matrix of each RRU into the power constraint formula, we obtain the precise expression for the precoding gain corresponding to the ZF precoding algorithm: The expert knowledge branch network simplifies the precise expression for the precoding gain of the ZF precoding algorithm based on the principle of asymptotic orthogonality of the channel in massive MIMO: when the number of antennas equipped on the RRU When it is large, there is ,therefore, and It can be approximated as a diagonal matrix: Substituting the above approximate diagonal matrix into the precise expression for the precoding gain corresponding to the ZF precoding algorithm, a fast approximate calculation formula for the first precoding gain can be derived: In the above formula, This represents the transmit power of the RRU, usually assumed for simplified analysis. , This represents the estimated value of the precoding gain calculated based on the simplified ZF precoding algorithm.

[0050] For the Regularized Zero-Forcing (RZF) precoding algorithm, its precoding matrix is: ,in It is the identity matrix. This represents the regularization parameter. Referring to the derivation of the ZF precoding algorithm, a fast approximate formula for calculating the second precoding gain can be obtained: In the above formula, This represents the transmit power of the RRU, usually assumed for simplified analysis. , This represents the estimated value of the precoding gain calculated based on the simplified RZF precoding algorithm.

[0051] For the minimum mean square error (MMSE) precoding algorithm, its precoding matrix is: ,in, This is the regularization parameter. Referring to the derivation of the ZF precoding algorithm, a fast approximate formula for calculating the third precoding gain can be obtained: In the above formula, This represents the transmit power of the RRU, usually assumed for simplified analysis. , This represents the estimated value of the precoding gain calculated based on the simplified MMSE precoding algorithm.

[0052] Furthermore, the linear precoding algorithm is a zero-forcing precoding algorithm. Accordingly, the UCDU calculates the estimated value of the precoding gain according to the fast approximation calculation formula of the precoding gain, including: the UCDU calculates the minimum value among the constraint ratios of each remote radio unit according to the first fast approximation calculation formula of the precoding gain to obtain the estimated value of the precoding gain. The first fast approximation calculation formula of the precoding gain is obtained by simplifying the accurate calculation formula of the precoding gain corresponding to the zero-forcing precoding algorithm based on the channel asymptotic orthogonality of massive MIMO; wherein, the constraint ratio of a remote radio unit is the ratio of the product of the number of antennas of the remote radio unit and the transmit power of the remote radio unit to the power constraint factor of the remote radio unit. The power constraint factor of the remote radio unit is calculated based on the large-scale fading coefficient matrix.

[0053] The fast approximation formula for the first precoding gain is as follows: First, calculate the total path loss value for each UE. : For each user equipment, the sum of the large-scale fading coefficients from each user to all RRUs is calculated based on the large-scale fading coefficient matrix to obtain the total path loss value for the user. .

[0054] Then, the power constraint factor for each RRU is calculated. : For each RRU, perform the following operations: 1. For all users served by RRU, calculate the ratio of the large-scale fading coefficient from the user to the RRU to the square of the total path loss value corresponding to the user. 2. Sum the above ratios calculated for all users to obtain the power constraint factor of the RRU. .

[0055] Next, the constraint ratio of each RRU is calculated. : For each RRU, calculate the constraint ratio, where the constraint ratio is the number of RRU antennas. Multiply by RRU transmit power Divide by the power constraint factor of the RRU .

[0056] Finally, the estimated value of the precoding gain is determined. : The minimum constraint ratio is obtained by selecting the minimum value among all RRU constraint ratios. The minimum constraint ratio is used as an estimate of the precoding gain. .

[0057] Furthermore, the linear precoding algorithm is a regularized zero-forcing precoding algorithm. Accordingly, the UCDU calculates the estimated value of the precoding gain according to the fast approximation calculation formula of the precoding gain, including: the UCDU calculates the minimum value among the constraint ratios of each remote radio unit according to the second fast approximation calculation formula of the precoding gain to obtain the estimated value of the precoding gain. The second fast approximation calculation formula of the precoding gain is obtained by simplifying the accurate calculation formula of the precoding gain corresponding to the regularized zero-forcing precoding algorithm based on the channel asymptotic orthogonality of large-scale MIMO. Among them, the constraint ratio of a remote radio unit is the ratio of the transmit power of the remote radio unit to the power constraint factor of the remote radio unit. The power constraint factor of the remote radio unit is calculated based on the large-scale fading coefficient matrix and the user-corrected channel strength factor. The user-corrected channel strength factor is determined based on the user's total path loss value, the number of antennas of the remote radio unit, and the regularization parameter in the regularized zero-forcing precoding algorithm.

[0058] The fast approximation formula for the second precoding gain is as follows: First, calculate the total path loss value for each UE. : For each user equipment, the sum of the large-scale fading coefficients from each user to all RRUs is calculated based on the large-scale fading coefficient matrix to obtain the total path loss value for the user. .

[0059] Secondly, a modified channel strength factor is introduced. For each user equipment, multiply the total path loss value corresponding to the user by the number of RRU antennas. Then, add a preset regularization parameter. The corrected channel strength factor is obtained. .

[0060] Then, the power constraint factor for each RRU is calculated. : For each RRU, perform the following operations: 1. For all users served by the RRU, calculate the ratio of the user's large-scale fading coefficient to the RRU to the square of the user's corrected channel strength factor. ; 2. Sum the above ratios for all users to obtain the summation result. ; 3. Multiply the summation result by the number of antennas in the RRU. The power constraint factor of the RRU is obtained. .

[0061] Next, the constraint ratio of each RRU is calculated. : For each RRU, calculate the constraint ratio. The constraint ratio of an RRU is equal to its transmit power. Divide by the power constraint factor of the RRU Finally, the estimated value of the coding gain is determined. : The minimum constraint ratio is obtained by selecting the minimum value among all RRU constraint ratios. The minimum constraint ratio is used as an estimate of the precoding gain. .

[0062] Furthermore, the linear precoding algorithm is a minimum mean square error precoding algorithm. Accordingly, the UCDU calculates the estimated value of the precoding gain according to the fast approximation calculation formula of the precoding gain, including: the UCDU calculates the minimum value among the constraint ratios of each remote radio unit according to the fast approximation calculation formula of the third precoding gain to obtain the estimated value of the precoding gain. The fast approximation calculation formula of the third precoding gain is obtained by simplifying the accurate calculation formula of the precoding gain corresponding to the minimum mean square error precoding algorithm based on the channel asymptotic orthogonality of massive MIMO. Wherein, the constraint ratio of a remote radio unit is the ratio of the transmit power of the remote radio unit to the power constraint factor of the remote radio unit. The power constraint factor of the remote radio unit is calculated based on the large-scale fading coefficient matrix and the user-corrected channel strength factor. The user-corrected channel strength factor is determined based on the user's total path loss value, the number of antennas of the remote radio unit, and the regularization parameter in the minimum mean square error precoding algorithm.

[0063] The fast approximation formula for the third precoding gain is as follows: First, calculate the total path loss value for each UE. For each user equipment, the sum of the large-scale fading coefficients from each user to all RRUs is calculated based on the large-scale fading coefficient matrix to obtain the total path loss value for the user.

[0064] Secondly, calculate the regularization parameter of MMSE. : Based on system noise power Total number of user devices and RRU transmit power Calculate the regularization parameters corresponding to the MMSE precoding. .

[0065] Secondly, a modified channel strength factor is introduced. : For each user equipment, multiply its corresponding total path loss value by the number of antennas. Then, add regularization parameters. The corrected channel strength factor is obtained. .

[0066] Then, the power constraint factor for each RRU is calculated. : For each RRU, perform the following operations: 1. For all users served by the RRU, calculate the ratio of the user's large-scale fading coefficient to the RRU to the square of the user's corrected channel strength factor. ; 2. Sum the above ratios for all users to obtain the summation result. ; 3. Multiply the summation result by the number of antennas in the RRU. The power constraint factor of the RRU is obtained.

[0067] Next, the constraint ratio of each RRU is calculated. : Calculate the constraint ratio for each RRU, which is the transmit power of that RRU. Divide by the power constraint factor of the RRU.

[0068] Finally, the estimated value of the coding gain is determined. : The minimum constraint ratio is obtained by selecting the minimum value among all RRU constraint ratios. The minimum constraint ratio is used as an estimate of the precoding gain. .

[0069] It should be noted that steps S210 and S220 can be executed synchronously and in parallel.

[0070] S240 and UCDU process and fuse the estimated values ​​of the compressed channel matrix and the precoding gain through the residual sensing fusion neural network, and output the estimated value of the residual difference between the accurate value of the precoding gain and the estimated value of the precoding gain.

[0071] Among them, the residual-aware fusion neural network is a neural network that measures the residual between the learned precise value of the precoded gain and the estimated value of the precoded gain.

[0072] S250 and UCDU determine the final precoding gain based on the estimated value of the precoding gain and the estimated value of the residual.

[0073] The second embodiment of the present invention provides a precoding gain estimation method. In this method, the expert knowledge branch network provides the estimation result of precoding gain based on the large-scale channel fading coefficient, which can reduce the learning difficulty and effectively compensate for system errors.

[0074] Example 3 Figure 4 This is a flowchart illustrating a precoding gain estimation method according to Embodiment 3 of the present invention. Embodiment 3 is an optimization based on the above embodiments. For details not covered in this embodiment, please refer to the above embodiments.

[0075] like Figure 4 As shown, the precoding gain estimation method provided in Embodiment 3 of the present invention includes the following steps: S310 and EDU use the data-driven branch network in the residual sensing fusion neural network to compress the channel state information of the global channel matrix and output a compressed channel matrix.

[0076] S320 and EDU calculate the large-scale fading coefficient matrix based on the global channel matrix.

[0077] S330 and UCDU determine the estimated value of the precoding gain based on the large-scale fading coefficient matrix.

[0078] S340 and UCDU use the encoders of the data-driven branch network and the expert knowledge branch network in the residual sensing fusion neural network to perform feature encoding on the estimated values ​​of the compressed channel matrix and the precoding gain, respectively, and output the first feature vector and the second feature vector.

[0079] Specifically, the UCDU uses the encoders of the data-driven branch network and the expert knowledge branch network in the residual sensing fusion neural network to perform feature encoding on the compressed channel matrix and the estimated value of the precoding gain, respectively, and outputs a first feature vector and a second feature vector. This includes: the UCDU converts the compressed channel matrix into a one-dimensional vector, inputs the one-dimensional vector into the first encoder of the data-driven branch network, and outputs a first feature vector; the UCDU inputs the estimated value of the precoding gain into the second encoder of the expert knowledge branch network and outputs a second feature vector.

[0080] Among them, UCDU will compress the channel matrix. After being converted into a one-dimensional vector, it is input into the first encoder in the data-driven branch network, and the output is a high-level feature vector (i.e., the first feature vector). , This represents the dimension of the first eigenvector; UCDU estimates the precoding gain. The second encoder in the expert knowledge branch network is input, and the output is a feature vector (i.e., the second feature vector). , This represents the dimension of the second feature vector.

[0081] It should be noted that the first encoder and the second encoder can be multilayer perceptrons, or convolutional neural networks, graph neural networks, or Transformer layers.

[0082] S350 and UCDU perform feature fusion on the first feature vector and the second feature vector through the fusion network in the residual perception fusion neural network to obtain a fused feature vector.

[0083] Among them, UCDU uses the fusion network in the residual-aware fusion neural network to integrate the first feature vector. Second eigenvector The features are concatenated to output a fused feature vector. .

[0084] S360 and UCDU input the fused feature vector into the residual estimator of the fused network and output the estimated value of the residual.

[0085] The residual estimator can be a multilayer perceptron, a convolutional neural network, a graph neural network, or a Transformer layer. The residual estimator can be used for residual estimation.

[0086] S370 and UCDU output the sum of the estimated residual and the estimated precoding gain as the final precoding gain.

[0087] The third embodiment of the present invention provides a precoding gain estimation method that can meet the real-time scheduling requirements of L2 layer in terms of computation time and can approximate the accurate value in terms of estimation accuracy.

[0088] Based on the technical solutions of the above embodiments, this invention provides a specific implementation method.

[0089] As one specific implementation method of this embodiment. Figure 5 A flowchart illustrating a precoding gain estimation method provided in a specific embodiment of the present invention is shown below. Figure 5 As shown, the process includes the following: S01: EDU Processing EDU receives global channel matrix The initial computations for both branches are performed using a residual-aware fusion neural network: Branch 1: Feature extraction via data-driven branch network: a) Global channel state information compression: The real and imaginary parts are separated and reshaped into a tensor. The tensor is compressed using a multilayer perceptron with shared weights, and the compressed channel matrix is ​​output. .

[0090] Branch 2: Calculate the large-scale fading coefficient: based on the global channel matrix. Calculate the large-scale fading coefficient The large-scale fading coefficient matrix is ​​obtained. .

[0091] S02: Fronthaul transmission EDU compresses the intermediate result—the channel matrix. and large-scale fading coefficient matrix It is sent to UCDU via the fronthaul link.

[0092] S03: UCDU Processing and Fusion Reasoning 1) Baseline gain calculation: The expert baseline estimate (i.e., the estimate of the precoding gain) is calculated based on the fast approximation formula for precoding gain. .

[0093] 2) Feature encoding: Branch-one encoder (i.e., the first encoder of the data-driven branch network): will Flattening (corresponding to Flatten in the figure) and passing it through a multilayer perceptron (MLP) (corresponding to MLP encoder in the figure) outputs a high-level feature vector. ; Branch-two encoder (i.e., the second encoder of the expert branch network): will Output feature vector through a small MLP .

[0094] 3) Feature fusion: and By concatenating the features, a fused feature vector is obtained. .

[0095] 4) Residual estimation: Input a residual estimator and output the residual estimate. .

[0096] 5) Final Output: Calculate the final precoding gain: .

[0097] S04: Application L2 layer uses Calculate PostSINR ( This allows us to determine the optimal MCS.

[0098] A comparative analysis was conducted on the MSE and MAE of the precoding gain estimation using the LS channel coefficients based on the asymptotic orthogonality principle, the direct use of deep neural networks (DNN), and graph neural networks (GNN), and the proposed scheme, Residual-Aware Fusion Neural Network (RAFNN), to estimate the precoding gain error. The results are shown in Table 1, which compares the number of parameters and performance. As can be seen from Table 1, RAFNN has a significant improvement in estimation accuracy.

[0099] Table 1 Example 4 Figure 6 This is a flowchart illustrating a precoding gain estimation method provided in Embodiment 4 of the present invention. This method is applicable to downlink transmission in cellular massive MIMO systems and can be executed by a distributed unit (DU) in the downlink system of a cellular radio access network.

[0100] like Figure 6 As shown, the precoding gain estimation method provided in Embodiment 4 of the present invention includes the following steps: S410 and DU determine the compressed channel matrix and the large-scale fading coefficient matrix.

[0101] S420 and DU determine the estimated value of the precoding gain based on the large-scale fading coefficient matrix.

[0102] S430 and DU process and fuse the estimated values ​​of the compressed channel matrix and the precoding gain through a residual-aware fusion neural network, and output an estimated value of the residual difference between the accurate value of the precoding gain and the estimated value of the precoding gain.

[0103] The residual-aware fusion neural network is a neural network that learns the residual between the precise value of the precoded gain and the estimated value of the precoded gain. After the residual-aware fusion neural network has been learned, it is deployed on DU.

[0104] S440 and DU determine the final precoding gain based on the estimated value of the precoding gain and the estimated value of the residual.

[0105] The relevant content and explanation of the precoding gain estimation method provided in Embodiment 4 of the present invention can be found in Embodiments 1 to 3, and will not be repeated here.

[0106] This invention provides a precoding gain estimation method in Embodiment 4. First, the differential input (DU) determines the compressed channel matrix and the large-scale fading coefficient matrix. Then, the DU determines the estimated value of the precoding gain based on the large-scale fading coefficient matrix. Next, the DU processes and fuses the estimated value of the compressed channel matrix and the precoding gain through the residual-aware fusion neural network, outputting an estimated value of the residual between the precise value of the precoding gain and the estimated value of the precoding gain. Finally, the DU determines the final precoding gain based on the estimated value of the precoding gain and the estimated value of the residual. The residual-aware fusion neural network is a neural network that has learned the residual between the precise value of the precoding gain and the estimated value of the precoding gain. This method transforms the complex precoding gain estimation problem into a more easily learned residual estimation problem. This method only needs to calculate the estimated value of the precoding gain and outputs the estimated value of the residual through the residual-aware fusion neural network. Compared to existing technologies that use precoding algorithms to accurately calculate the precoding gain, this method can effectively improve computational efficiency. Compared to simply calculating the estimated value of the precoding gain, the estimation accuracy of this method is significantly better than that of the precoding gain estimation, achieving an optimal balance between accuracy and computational efficiency.

[0107] Example 5 Embodiment 5 of the present invention provides a downlink system for a non-cellular wireless access network. The system includes a central unit, a user-centric distributed unit (UCDU), an edge distributed unit (EDU), and a remote radio unit. The UCDU and EDU are capable of executing the precoding gain estimation methods provided in Embodiments 1 to 3 of the present invention.

[0108] The structural diagram of the downlink system of the non-cellular wireless access network can be found in the following reference. Figure 1 .

[0109] The fifth embodiment of the present invention provides a downlink system for a non-cellular wireless access network that can improve the calculation efficiency of precoding gain.

[0110] Example 6 Embodiment 6 of the present invention provides a downlink system for a non-cellular wireless access network, the system including a central unit, a distributed unit (DU), and a remote radio frequency unit (EDU); the EDU is capable of executing the precoding gain estimation method provided in Embodiment 4 of the present invention.

[0111] The downlink system of a non-cellular wireless access network provided in Embodiment 6 of the present invention can improve the calculation efficiency of precoding gain.

[0112] Example 7 Figure 7 A schematic diagram of an electronic device 10 that can be used to implement embodiments of the present invention is shown. The electronic device can be an edge distributed unit (EDU) in a non-cellular wireless access network downlink system. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the invention described and / or claimed herein.

[0113] like Figure 7 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0114] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0115] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as precoding gain estimation methods.

[0116] In some embodiments, the precoding gain estimation method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or mounted on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the precoding gain estimation method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the precoding gain estimation method by any other suitable means (e.g., by means of firmware).

[0117] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0118] In some embodiments, the precoding gain estimation method may be implemented as a computer program, which is implicitly included in a computer program product. When executed by a processor, the computer program implements the precoding gain estimation method of the present invention. The computer program product can be understood as a software product that primarily implements its solution through a computer program. The computer program used to implement the method of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer program causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The computer program may be executed entirely on a machine, partially on a machine, partially on a remote machine as a standalone software package, or entirely on a remote machine or server.

[0119] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0120] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0121] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0122] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0123] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0124] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A precoding gain estimation method, characterized in that, The method, performed by the edge distributed unit (EDU) and the user-centric distributed unit (UCDU) in a non-cellular radio access network downlink system, includes: The EDU determines the compressed channel matrix and the large-scale fading coefficient matrix; The UCDU determines the estimated value of the precoding gain based on the large-scale fading coefficient matrix; The UCDU processes and fuses the estimated values ​​of the compressed channel matrix and the precoding gain through a residual-aware fusion neural network, and outputs an estimated value of the residual difference between the precise value of the precoding gain and the estimated value of the precoding gain. The UCDU determines the final precoding gain based on the estimated value of the precoding gain and the estimated value of the residual; The residual-aware fusion neural network is a neural network that measures the residual between the learned precise value of the precoding gain and the estimated value of the precoding gain.

2. The method according to claim 1, characterized in that, The residual-aware fusion neural network includes an expert knowledge branch network and a data-driven branch network, which perform corresponding calculations in parallel. The EDU determines the compressed channel matrix and the large-scale fading coefficient matrix, including: The EDU compresses the channel state information of the global channel matrix and outputs a compressed channel matrix through the data-driven branch network. The EDU calculates the large-scale fading coefficient matrix based on the global channel matrix.

3. The method according to claim 2, characterized in that, The EDU, through the data-driven branch network, compresses the channel state information of the global channel matrix and outputs a compressed channel matrix, including: The EDU, through the data-driven branch network, separates the real and imaginary parts of the global channel matrix and reshapes it into a tensor; The EDU compresses the tensor using a shared-weight feature extractor through the data-driven branch network, and outputs a compressed channel matrix.

4. The method according to claim 2, characterized in that, The EDU calculates the large-scale fading coefficient matrix based on the global channel matrix, including: The EDU calculates the square of the modulus of the global channel vector in the global channel matrix and divides it by the number of antennas to obtain the large-scale fading coefficient matrix.

5. The method according to claim 1, characterized in that, The UCDU determines the estimated value of the precoding gain based on the large-scale fading coefficient matrix, including: The UCDU calculates the estimated value of the precoding gain according to the fast approximation calculation formula of the precoding gain. The fast approximation calculation formula of the precoding gain is determined based on the channel asymptotic orthogonality of large-scale MIMO and the accurate calculation formula of the precoding gain corresponding to different linear precoding algorithms.

6. The method according to claim 5, characterized in that, The linear precoding algorithm is a zero-forcing precoding algorithm, and correspondingly... The UCDU calculates an estimate of the precoding gain based on a fast approximation formula for precoding gain, including: The UCDU calculates the minimum value among the constraint ratios of each remote radio unit according to the fast approximation calculation formula of the first precoding gain, and obtains the estimated value of the precoding gain. The fast approximation calculation formula of the first precoding gain is obtained by simplifying the accurate calculation formula of the precoding gain corresponding to the zero-forcing precoding algorithm based on the channel asymptotic orthogonality of large-scale MIMO. The constraint ratio of a remote radio frequency unit is the ratio of the product of the number of antennas of the remote radio frequency unit and the transmit power of the remote radio frequency unit to the power constraint factor of the remote radio frequency unit. The power constraint factor of the remote radio frequency unit is calculated based on the large-scale fading coefficient matrix.

7. The method according to claim 1, 2, or 5, characterized in that, The UCDU processes and fuses the estimated values ​​of the compressed channel matrix and the precoding gain through a residual-aware fusion neural network, outputting an estimate of the residual between the precise value of the precoding gain and the estimated value of the precoding gain, including: The UCDU uses the encoders of the data-driven branch network and the expert knowledge branch network in the residual sensing fusion neural network to perform feature encoding on the estimated values ​​of the compressed channel matrix and the precoding gain, respectively, and outputs a first feature vector and a second feature vector. The UCDU performs feature fusion on the first feature vector and the second feature vector through the fusion network in the residual perception fusion neural network to obtain a fused feature vector; The UCDU inputs the fused feature vector into the residual estimator in the fusion network and outputs the estimated value of the residual.

8. The method according to claim 7, characterized in that, The UCDU uses the encoders of the data-driven branch network and the expert knowledge branch network in the residual-aware fusion neural network to perform feature encoding on the estimated values ​​of the compressed channel matrix and the precoding gain, respectively, and outputs a first feature vector and a second feature vector, including: The UCDU converts the compressed channel matrix into a one-dimensional vector, inputs the one-dimensional vector into the first encoder of the data-driven branch network, and outputs a first feature vector. The UCDU inputs the estimated value of the precoding gain into the second encoder of the expert knowledge branch network and outputs a second feature vector.

9. A precoding gain estimation method, characterized in that, The method, performed by a distributed unit (DU) in a downlink system of a noncellular radio access network, includes: The DU determines the compressed channel matrix and the large-scale fading coefficient matrix; The DU determines the estimated value of the precoding gain based on the large-scale fading coefficient matrix; The DU processes and fuses the estimated values ​​of the compressed channel matrix and the precoding gain through a residual-aware fusion neural network, and outputs an estimated value of the residual difference between the precise value of the precoding gain and the estimated value of the precoding gain. The DU determines the final precoding gain based on the estimated value of the precoding gain and the estimated value of the residual; The residual-aware fusion neural network is a neural network that measures the residual between the learned precise value of the precoding gain and the estimated value of the precoding gain.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the precoding gain estimation method of claim 11.

11. A downlink system for a non-cellular wireless access network, characterized in that, The system includes a central unit, a user-centric distributed unit (UCDU), an edge distributed unit (EDU), and a remote radio frequency unit. The UCDU and the EDU perform the precoding gain estimation method as described in any one of claims 1-8.

12. A downlink system for a non-cellular wireless access network, characterized in that, The system includes a central unit, a distributed unit (DU), and a remote radio frequency unit. The DU performs the precoding gain estimation method as described in claim 9.

Images