阅读说明：本技术 一种基于深度学习的功率分配方法及分配装置 (Power distribution method and device based on deep learning ) 是由 董超 牛凯 王伟 汪诗雨 于 2019-09-25 设计创作，主要内容包括：本发明实施例提供了一种基于深度学习的功率分配方法及分配装置,其中方法包括：获取用户的信道矩阵；对信道矩阵进行奇异值分解处理,得到等效信道特征值以及左酉矩阵；基于等效信道特征值以及用户的预设功率限制参数,生成用户的信道特征信息；将信道特征信息输入预先训练好的全连接神经网络模型中,得到用户的数据流数；基于用户的数据流数,等效信道特征值,以及左酉矩阵,构建用户的功率分配协方差矩阵；基于用户的功率分配协方差矩阵,为用户分配传输功率。本发明实施例能够提高分配传输功率的效率。(The embodiment of the invention provides a power distribution method and a power distribution device based on deep learning, wherein the method comprises the following steps: acquiring a channel matrix of a user; performing singular value decomposition processing on the channel matrix to obtain an equivalent channel characteristic value and a left unitary matrix; generating channel characteristic information of the user based on the equivalent channel characteristic value and a preset power limiting parameter of the user; inputting the channel characteristic information into a pre-trained fully-connected neural network model to obtain the data flow number of a user; constructing a power distribution covariance matrix of the user based on the data stream number of the user, the equivalent channel characteristic value and the left unitary matrix; and allocating transmission power for the users based on the power allocation covariance matrix of the users. The embodiment of the invention can improve the efficiency of distributing the transmission power.)

1. A method for deep learning based power allocation, the method comprising:

acquiring a channel matrix of a user;

performing singular value decomposition processing on the channel matrix to obtain an equivalent channel eigenvalue and a left unitary matrix, wherein the equivalent channel eigenvalue is a singular value corresponding to each of a plurality of orthogonal channels included in an equivalent channel corresponding to the channel matrix;

generating channel characteristic information of the user based on the equivalent channel characteristic value and a preset power limiting parameter of the user;

inputting the channel characteristic information into a pre-trained fully-connected neural network model to obtain the data flow number of the user, wherein the data flow number is used for expressing the data flow size distributed to the user;

constructing a power distribution covariance matrix of the user based on the data stream number of the user, the equivalent channel characteristic value and the left unitary matrix;

and allocating transmission power for the users based on the power allocation covariance matrix of the users.

2. The method of claim 1, wherein when the number of the users is one, the user is in a MIMO channel scenario, and the step of calculating the power allocation covariance matrix of the user based on the equivalent channel eigenvalue, the left unitary matrix, and the number of data streams of the user comprises:

calculating a power distribution covariance matrix of the single user based on the equivalent channel characteristic value of the single user, the left unitary matrix and the data stream number of the single user;

the step of allocating transmission power to the user based on the power allocation covariance matrix of the user comprises:

and allocating transmission power for the single user based on the power allocation covariance matrix of the single user.

3. The method of claim 1, wherein when the number of the users is multiple, the users are in a multiple access channel MIMO-MAC channel scenario, and the step of calculating the power allocation covariance matrix of the users based on the equivalent channel eigenvalues, the left unitary matrix, and the number of data streams of the users comprises:

for each user in the plurality of users, calculating a power distribution covariance matrix of each user based on an equivalent channel characteristic value corresponding to each user, the left unitary matrix and the data stream number;

the step of allocating transmission power to the user based on the power allocation covariance matrix of the user comprises:

and distributing transmission power for each user based on the power distribution covariance matrix corresponding to each user.

4. The method according to claim 1, wherein the step of performing singular value decomposition on the channel matrix to obtain an equivalent channel eigenvalue and a left unitary matrix comprises:

performing singular value decomposition processing on the channel matrix by using a first preset expression, wherein the first preset expression is as follows:

SVD(H)＝UΛV^H

in the formula, H represents a channel matrixSVD (H) represents that H is subjected to singular value decomposition, U represents a left unitary matrix, Λ represents a diagonal matrix of H, V represents a right unitary matrix of H, and V represents^HRepresenting the conjugate transpose of V.

5. The method of any one of claims 1-4, wherein the training process of the fully-connected neural network model comprises:

constructing an initial fully-connected neural network model, wherein the initial fully-connected neural network model at least comprises a 1-layer input layer, a 3-layer hidden layer and a 1-layer output layer;

acquiring sample channel characteristic information of a sample channel matrix and a data stream number label corresponding to each sample channel characteristic information;

and inputting the characteristic information of each sample channel and the data stream number label corresponding to the characteristic information of each sample channel into the initial fully-connected neural network model, and training to obtain the fully-connected neural network model.

6. An apparatus for deep learning based power distribution, the apparatus comprising:

the first acquisition module is used for acquiring a channel matrix of a user;

the processing module is used for performing singular value decomposition processing on the channel matrix to obtain an equivalent channel characteristic value and a left unitary matrix, wherein the equivalent channel characteristic value is a singular value corresponding to each of a plurality of orthogonal channels included in an equivalent channel corresponding to the channel matrix;

a generating module, configured to generate channel characteristic information of the user based on the equivalent channel characteristic value and a preset power limitation parameter of the user;

the input module is used for inputting the channel characteristic information into a pre-trained fully-connected neural network model to obtain the data flow number of the user, wherein the data flow number is used for expressing the data flow size distributed to the user;

a first constructing module, configured to construct a power allocation covariance matrix of the user based on the number of data streams of the user, the equivalent channel eigenvalue, and the left unitary matrix;

and the distribution module is used for distributing the transmission power for the user based on the power distribution covariance matrix of the user.

7. The apparatus according to claim 6, wherein when the number of the users is one, the user is in a MIMO channel scenario, and the first constructing module is specifically configured to:

calculating a power distribution covariance matrix of the single user based on the equivalent channel characteristic value of the single user, the left unitary matrix and the data stream number of the single user;

the allocation module has means for:

and allocating transmission power for the single user based on the power allocation covariance matrix of the single user.

8. The apparatus according to claim 6, wherein when the number of the users is multiple, the users are in a multiple access channel MIMO-MAC channel scenario, and the first constructing module is specifically configured to:

for each user in the plurality of users, calculating a power distribution covariance matrix of each user based on an equivalent channel characteristic value corresponding to each user, the left unitary matrix and the data stream number;

the allocation module has means for:

and distributing transmission power for each user based on the power distribution covariance matrix corresponding to each user.

9. An electronic device, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete communication with each other through the communication bus;

the memory is used for storing a computer program;

the processor, when executing the program stored in the memory, implementing the method steps of any of claims 1-5.

10. A computer-readable storage medium, characterized in that a computer program is stored in the computer-readable storage medium, which computer program, when being executed by a processor, carries out the method steps of any one of claims 1-5.

Technical Field

The present invention relates to the field of wireless communication technologies, and in particular, to a power allocation method and an allocation apparatus based on deep learning.

Background

MIMO (Multiple-Input Multiple-Output) technology is a technology capable of independently transmitting signals by using Multiple antennas at a transmitting end, and simultaneously receiving and recovering original information by using Multiple antennas at a receiving end, and has become one of key technologies of a wireless communication system as communication technology is continuously developed.

The MIMO technology may be applied to a single-user MIMO Channel scenario or a MIMO-MAC (Multiple Access Channel) Channel scenario, where the single-user MIMO Channel scenario is for a user of a single Access system, and the MIMO-MAC Channel scenario is for users of Multiple Access systems, and the single-user MIMO Channel scenario may be considered as a special case of the MIMO-MAC scenario. For both single-user MIMO channel scenarios and MIMO-MAC scenarios, the power allocation covariance matrix of the user needs to be calculated, so that the transmission power is allocated to the user according to the power allocation covariance matrix.

Disclosure of Invention

Embodiments of the present invention provide a power allocation method and an allocation apparatus based on deep learning, so as to reduce the computation complexity in the process of computing a power allocation covariance matrix, thereby improving the efficiency of allocating transmission power. The specific technical scheme is as follows:

in a first aspect, an embodiment of the present invention provides a power allocation method based on deep learning, where the method includes:

acquiring a channel matrix of a user;

performing singular value decomposition processing on the channel matrix to obtain an equivalent channel eigenvalue and a left unitary matrix, wherein the equivalent channel eigenvalue is a singular value corresponding to each of a plurality of orthogonal channels included in an equivalent channel corresponding to the channel matrix;

generating channel characteristic information of the user based on the equivalent channel characteristic value and a preset power limiting parameter of the user;

inputting the channel characteristic information into a pre-trained fully-connected neural network model to obtain the data flow number of the user, wherein the data flow number is used for expressing the data flow size distributed to the user;

constructing a power distribution covariance matrix of the user based on the data stream number of the user, the equivalent channel characteristic value and the left unitary matrix;

and allocating transmission power for the users based on the power allocation covariance matrix of the users.

Optionally, when the number of the users is one, the user is in a MIMO channel scenario, and the step of calculating the power allocation covariance matrix of the user based on the equivalent channel eigenvalue, the left unitary matrix, and the number of data streams of the user includes:

calculating a power distribution covariance matrix of the single user based on the equivalent channel characteristic value of the single user, the left unitary matrix and the data stream number of the single user;

the step of allocating transmission power to the user based on the power allocation covariance matrix of the user comprises:

and allocating transmission power for the single user based on the power allocation covariance matrix of the single user.

Optionally, when the number of the users is multiple, the users are in a multiple access channel MIMO-MAC channel scenario, and the step of calculating the power allocation covariance matrix of the users based on the equivalent channel eigenvalue, the left unitary matrix, and the number of data streams of the users includes:

for each user in the plurality of users, calculating a power distribution covariance matrix of each user based on an equivalent channel characteristic value corresponding to each user, the left unitary matrix and the data stream number;

the step of allocating transmission power to the user based on the power allocation covariance matrix of the user comprises:

and distributing transmission power for each user based on the power distribution covariance matrix corresponding to each user.

Optionally, the step of performing singular value decomposition processing on the channel matrix to obtain an equivalent channel eigenvalue and a left unitary matrix includes:

performing singular value decomposition processing on the channel matrix by using a first preset expression, wherein the first preset expression is as follows:

SVD(H)＝UΛV^H

wherein H represents a channel matrix, SVD (H) represents singular value decomposition processing on H, U represents a left unitary matrix, Λ represents a diagonal matrix, V represents a right unitary matrix of H, and V represents^HRepresenting the conjugate transpose of V.

Optionally, the training process of the fully-connected neural network model includes:

constructing an initial fully-connected neural network model, wherein the initial fully-connected neural network model at least comprises a 1-layer input layer, a 3-layer hidden layer and a 1-layer output layer;

acquiring sample channel characteristic information of a sample channel matrix and a data stream number label corresponding to each sample channel characteristic information;

and inputting the characteristic information of each sample channel and the data stream number label corresponding to the characteristic information of each sample channel into the initial fully-connected neural network model, and training to obtain the fully-connected neural network model.

In a second aspect, an embodiment of the present invention provides a deep learning-based power distribution apparatus, where the apparatus includes:

the first acquisition module is used for acquiring a channel matrix of a user;

the processing module is used for performing singular value decomposition processing on the channel matrix to obtain an equivalent channel characteristic value and a left unitary matrix, wherein the equivalent channel characteristic value is a singular value corresponding to each of a plurality of orthogonal channels included in an equivalent channel corresponding to the channel matrix;

a generating module, configured to generate channel characteristic information of the user based on the equivalent channel characteristic value and a preset power limitation parameter of the user;

the input module is used for inputting the channel characteristic information into a pre-trained fully-connected neural network model to obtain the data flow number of the user, wherein the data flow number is used for expressing the data flow size distributed to the user;

a first constructing module, configured to construct a power allocation covariance matrix of the user based on the number of data streams of the user, the equivalent channel eigenvalue, and the left unitary matrix;

and the distribution module is used for distributing the transmission power for the user based on the power distribution covariance matrix of the user.

Optionally, when the number of the users is one, the user is in a MIMO channel scenario, and the first building module is specifically configured to:

calculating a power distribution covariance matrix of the single user based on the equivalent channel characteristic value of the single user, the left unitary matrix and the data stream number of the single user;

the allocation module has means for:

and allocating transmission power for the single user based on the power allocation covariance matrix of the single user.

Optionally, when the number of the users is multiple, the users are in a multiple access channel MIMO-MAC channel scenario, and the first constructing module is specifically configured to:

for each user in the plurality of users, calculating a power distribution covariance matrix of each user based on an equivalent channel characteristic value corresponding to each user, the left unitary matrix and the data stream number;

the allocation module has means for:

and distributing transmission power for each user based on the power distribution covariance matrix corresponding to each user.

Optionally, the processing module is specifically configured to:

performing singular value decomposition processing on the channel matrix by using a first preset expression, wherein the first preset expression is as follows:

SVD(H)＝UΛV^H

wherein H represents a channel matrix, SVD (H) represents that H is subjected to singular value decomposition, U represents a left unitary matrix, Λ represents a diagonal matrix, V represents a right unitary matrix, and V represents^HRepresenting the conjugate transpose of V.

Optionally, the power distribution apparatus based on deep learning according to the embodiment of the present invention further includes:

the second construction module is used for constructing an initial fully-connected neural network model, and the initial fully-connected neural network model at least comprises 1 input layer, 3 hidden layers and 1 output layer;

the second acquisition module is used for acquiring the sample channel characteristic information of the sample channel matrix and the data stream number labels corresponding to the sample channel characteristic information;

and the training module is used for inputting the characteristic information of each sample channel and the data stream number label corresponding to the characteristic information of each sample channel into the initial fully-connected neural network model and training to obtain the fully-connected neural network model.

In a third aspect, an embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, where the processor, the communication interface, and the memory complete mutual communication through the communication bus; the machine-readable storage medium stores machine-executable instructions executable by the processor, the processor being caused by the machine-executable instructions to: the method steps of the power allocation method based on deep learning provided by the first aspect of the embodiment of the invention are realized.

In a fourth aspect, the present invention provides a computer-readable storage medium, in which a computer program is stored, where the computer program is executed by a processor to perform the method steps of the deep learning based power allocation method provided in the first aspect of the present invention.

After a channel matrix of a user is obtained, singular value decomposition processing is carried out on the channel matrix through a preset singular value decomposition algorithm to obtain an equivalent channel characteristic value and a left unitary matrix, then channel characteristic information of the user is generated based on the equivalent channel characteristic value and preset power limiting parameters of the user, then the channel characteristic information is input into a pre-trained fully-connected neural network model to obtain the data flow number of the user, further, a power distribution covariance matrix of the user is constructed based on the data flow number, the equivalent channel characteristic value and the left unitary matrix of the user, then transmission power is distributed to the user based on the power distribution covariance matrix of the user, and therefore, the data flow number of the user can be obtained through the fully-connected neural network model, and then, a power distribution covariance matrix of the user is constructed based on the data flow number, and the power distribution covariance matrix of the user is not required to be calculated by using the existing water injection algorithm, so that the problems of large calculation amount and high complexity caused by repeated iterative calculation of the water injection algorithm are solved, and the efficiency of distributing transmission power is improved. Of course, it is not necessary for any product or method of practicing the invention to achieve all of the above-described advantages at the same time.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.

Fig. 1 is a schematic flowchart of a power allocation method based on deep learning according to an embodiment of the present invention;

fig. 2 is another schematic flow chart of a deep learning-based power allocation method according to an embodiment of the present invention;

FIG. 3 is a schematic structural diagram of a fully-connected neural network model according to an embodiment of the present invention;

fig. 4 is a schematic structural diagram of a deep learning-based power distribution apparatus according to an embodiment of the present invention;

fig. 5 is a schematic structural diagram of a deep learning-based power distribution apparatus according to an embodiment of the present invention;

FIG. 6 is a system rate comparison graph of a deep learning based power allocation method and a water filling algorithm in accordance with an embodiment of the present invention;

FIG. 7 is a graph illustrating a comparison of accuracy in determining the number of data streams for different antenna numbers according to an embodiment of the present invention;

FIG. 8 is a graph comparing the power allocation method of the present invention with the rate cumulative distribution function of the water-filling algorithm of the prior art;

fig. 9 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

As shown in fig. 1, an embodiment of the present invention provides a deep learning-based power allocation method, which may include the following steps:

s101, obtaining a channel matrix of a user.

The power distribution method of the embodiment of the invention can be applied to a single-user MIMO channel scene or an MIMO-MAC channel scene. The channel matrix of the user can be obtained according to the channel estimation of the user connected with the signal transmitting terminal. The user in the embodiment of the present invention may refer to a user terminal, for example, a mobile phone, a tablet computer, and the like.

It should be noted that, when the MIMO channel scenario is adopted, only the channel matrix of the single user may be obtained; when the channel is a MIMO-MAC channel scenario, for multiple users accessing the channel, the channel matrix of each user can be obtained respectively.

Referring to fig. 1, S102 performs singular value decomposition on the channel matrix to obtain an equivalent channel eigenvalue and a left unitary matrix.

The singular value decomposition is a common channel processing method, the embodiment of the invention can carry out singular value decomposition on a channel matrix, and after decomposition, an equivalent channel characteristic value and a left unitary matrix of the channel matrix can be obtained. The equivalent channel is an equivalent channel obtained by performing singular value decomposition on a channel matrix, for example, if the original channel matrix is a 4 × 4 matrix, the original channel matrix can be decomposed into 4 × 1 orthogonal channels, the 4 orthogonal channels form a group of equivalent channels, each orthogonal channel corresponds to a channel characteristic value, or is called a singular value, and a user represents a channel response of the orthogonal channel.

It should be noted that, when the MIMO channel scenario is adopted, singular value decomposition processing may be performed only on the channel matrix of the single user; when the channel is a MIMO-MAC channel scenario, for multiple users accessing the channel, singular value decomposition processing may be performed on the channel matrix of each user.

For a MIMO-MAC system with K number of users, the maximized system capacity is expressed as:

wherein I is additive white Gaussian noise, and the power limiting parameter of the user I belongs to {1,2,3_i，j∈{1,2,3,...,K}，H_iIndicating the channel response, S, of the ith user_iRepresenting powerDistribution matrix, G_iRepresenting the channel response when multiple users are equivalent to a single user, with the dimension of the number n of receiving and transmitting antennas_R×n_T。

Multi-user waterflooding can be converted into single-user waterflooding by the above formula, and then the equivalent channel of user i is represented as:

in the formula, H_jA channel response matrix, S, representing user j_jRepresenting the power allocation matrix for user j.

By performing single-user water injection on the equivalent channels of the users, the computational complexity of the traditional water injection algorithm can be simplified.

As an optional implementation manner of the embodiment of the present invention, singular value decomposition processing may be performed on a channel matrix by using a first preset expression, where the first preset expression is:

SVD(H)＝UΛV^H

wherein H represents a channel matrix, SVD (H) represents singular value decomposition processing on H, U represents a left unitary matrix, Λ represents a diagonal matrix of H, V represents a right unitary matrix of H, and V represents^HRepresenting the conjugate transpose of V. Wherein the channel characteristic value can be expressed as

Wherein N is_RIs the number of receive antennas.

Referring to fig. 1, S103 generates channel characteristic information of a user based on the equivalent channel characteristic value and a preset power limiting parameter of the user.

The embodiment of the invention can utilize the effective channel characteristic value obtained by calculation and the preset power limiting parameter of the user to jointly construct the channel characteristic information of the user. For example, the channel characteristic information may be expressed as:wherein λ is_iThe ith eigenvalue of the channel matrix is represented,

denotes λ_iIs squared, i e {1,2,3_RP denotes a power limitation parameter. Of course, different power limiting parameters described above may be set for different users.

It should be noted that, when the MIMO channel scenario is adopted, only the channel characteristic information of the single user may be generated; when the channel is a MIMO-MAC channel scenario, for multiple users accessing the channel, corresponding channel characteristic information may be generated for each user.

And S104, inputting the channel characteristic information into a pre-trained fully-connected neural network model to obtain the data flow number of the user.

After the channel characteristic information of the user is obtained, the channel characteristic information of the user can be input into a pre-trained fully-connected neural network model, so that the data flow number of the user is obtained. The number of data streams is used to represent the size of the data traffic allocated to the user, and it can be understood that, since the number of data streams can be used to represent the size of the data traffic allocated to the user, and the data traffic is positively correlated with the transmission power, the embodiment of the present invention can allocate the transmission power to the user based on the number of data streams.

It should be noted that, when the MIMO channel scenario is adopted, only the channel characteristic information of the single user may be input into the model, so as to obtain the channel characteristic information of the single user; when the scene is an MIMO-MAC channel scene, for a plurality of accessed users, the channel characteristic information of each user can be respectively input into the model, so that the data stream number corresponding to each user is respectively obtained.

Referring to fig. 1, S105, a power allocation covariance matrix of a user is constructed based on the number of data streams of the user, an equivalent channel eigenvalue, and a left unitary matrix.

In the embodiment of the present invention, after the data stream number of the user is obtained in step S104, the power allocation covariance matrix of the user may be jointly constructed based on the data stream number, the equivalent channel eigenvalue of the user obtained in step S101, and the left unitary matrix.

For example, the power allocation vector of the user may be calculated first through the number of data streams and the equivalent channel eigenvalue, and then the power allocation covariance matrix of the user may be calculated based on the power allocation vector and the left unitary matrix. The specific process can be as follows:

the calculated number of data streams of the user is expressed asThe elements in the power allocation vector can be represented as:

in the formula, p_jRepresents the size of the power allocation on the equivalent channel j, P_iDenotes the power limit, λ, of user i_jRepresenting the channel characteristic value, b representing an intermediate variable, a representing the second of the equivalent channels

The square of the individual eigenvalues. When making a comparison, to avoid when

When the number of the data streams is larger than that of the data streams obtained by the existing water injection algorithm, the generated power distribution vector has negative elements

Can be reduced

The value is that no negative element exists until the power distribution vector, and it should be noted that, because the number of data streams output by the neural network has high accuracy, when the number of database streams output by the neural network is greater than the number of data streams obtained by the existing water filling algorithm, that is, the number of data streams output by the neural network is greater than the number of data streams obtained by the existing water filling algorithm

Can be converted into correct data stream number

Calculating the elements in each power allocation vector by the above formula, and then obtaining the power allocation vector represented as

Then, according to the obtained power allocation vector and the left unitary matrix, a power allocation covariance matrix of the user can be constructed, which is expressed as:

wherein S represents a power distribution covariance matrix, U represents a left unitary matrix, and U represents a left unitary matrix^HA conjugate transpose matrix representing the left unitary matrix.

As an optional implementation manner of the embodiment of the present invention, when the number of users is one, and the user is in a MIMO channel scene at this time, the power allocation covariance matrix of the single user may be calculated based on the equivalent channel eigenvalue of the single user, the left unitary matrix, and the number of data streams of the single user.

As an optional implementation manner of the embodiment of the present invention, when the number of users is multiple, and the users are in a multiple access channel MIMO-MAC channel scenario, a power allocation covariance matrix of each user may be calculated for each of the multiple users based on an equivalent channel eigenvalue, a left unitary matrix, and a data stream number corresponding to each user.

Referring to fig. 1, transmission power is allocated to users based on their power allocation covariance matrix S106.

After the power distribution covariance matrix of the user is determined, the power distribution covariance matrix carries the power distribution information of the user, so that the transmission power can be distributed to the user based on the power distribution covariance matrix of the user.

As an optional implementation manner of the embodiment of the present invention, when the number of users is one, and the users are in a MIMO channel scenario at this time, the transmission power may be allocated to the single user based on the power allocation covariance matrix of the single user.

As an optional implementation manner of the embodiment of the present invention, when the number of users is multiple, and the users are in a multiple access channel MIMO-MAC channel scenario, the transmission power may be allocated to each user based on the power allocation covariance matrix corresponding to each user.

As shown in fig. 2, the training process of the fully-connected neural network model according to the embodiment of the present invention includes:

s201, constructing an initial full-connection neural network model.

The initial fully-connected neural network model constructed in the embodiment of the present invention is shown in fig. 3, and at least includes 1 input layer, 3 hidden layers, and 1 output layer.

The adopted neural network is a fully-connected neural network, the circles in the graph represent nodes, and the number of input nodes of the neural network can be N_R+1, the hidden layer is 3 layers, the number of output nodes can be N_RIn which N is_RAssuming that the number of nodes in each layer is n, the activation function is expressed as:

in the formula (I), the compound is shown in the specification,

it is shown that the activation function is,

representing the output of each node of the neural network model, e representing a natural constant, j ∈ {1,2, 3.

S202, obtaining the sample channel characteristic information of the sample channel matrix and the data stream number label corresponding to each sample channel characteristic information.

The embodiment of the invention can train the initial fully-connected neural network model by using the D sample channel characteristic information and the data stream number labels corresponding to the sample channel characteristic information. The training process can refer to the existing neural network training process, and the embodiment of the invention is not repeated herein.

And S203, inputting the characteristic information of each sample channel and the data stream number label corresponding to the characteristic information of each sample channel into the initial fully-connected neural network model, and training to obtain the fully-connected neural network model.

In the training process, each iteration obtains the equivalent channel characteristic value of the user iAnd a power limiting parameter P_iAnd sample channel characteristic information constituting the neural network, expressed as:

the output label of the neural network can be the data flow number N obtained by the water filling algorithm_iAnd counting the number of data streams N_iOnehot (one-hot encoding) mapping was performed. It should be noted that, when it is a single user, the sample channel characteristic information of the user may be represented as:

the data stream number N is y after being mapped by Onehot coding_iThe loss function may be a cross-entropy function, expressed as:

in the formula, loss_iRepresents the loss function, y_iRepresenting a mapping of the number of data streams N via Onehot encoding,

representing the output of each node of the neural network model, i represents a variable, j ∈ {1,2,3 ∈ },...,n}。

In the embodiment of the present invention, the process of performing Onehot mapping on the number of data streams is represented as follows:

in the formula, N represents the number of data streams, and the position where the value 1 is located in each Onehot code represents the size of the number of data streams, for example, [ 10 … 0] represents that the number of data streams is 1, and [ 01 … 0] represents that the number of data streams is 2.

As an optional implementation manner of the embodiment of the present invention, corresponding to a MIMO-MAC channel scenario, when a channel state of one of users changes, the embodiment of the present invention may reallocate transmission power for each user.

Exemplarily, after obtaining the power distribution covariance matrix from the first user to the ith user, the equivalent channel of the (i + 1) th user needs to be calculated, and based on the expression of calculating the equivalent channel, the equivalent channel of the user mod (i +1, K) can be obtained, and is represented as:

in the equation, mod (i +1, K) ═ K indicates that the equivalent channel of the user i +1 is calculated when i ≠ K, and indicates that the equivalent channel of the user 1 is calculated when i ≠ K. And when all the users calculate the equivalent channel, namely i is equal to K, the iteration process is finished, otherwise, the channel characteristic information of the users is constructed again, and the power distribution covariance matrix of the users is recalculated. Of course, when the preset maximum iteration number is reached, the water filling power allocation process of deep learning of all users is completed.

The embodiment of the invention provides a power distribution method based on deep learning, which comprises the steps of obtaining a channel matrix of a user, performing singular value decomposition processing on the channel matrix through a preset singular value decomposition algorithm to obtain an equivalent channel characteristic value and a left unitary matrix, generating channel characteristic information of the user based on the equivalent channel characteristic value and preset power limiting parameters of the user, inputting the channel characteristic information into a pre-trained fully-connected neural network model to obtain the data stream number of the user, further constructing a power distribution covariance matrix of the user based on the data stream number, the equivalent channel characteristic value and the left unitary matrix, distributing transmission power to the user based on the power distribution covariance matrix of the user, and thus, the embodiment of the invention can obtain the data stream number of the user through the fully-connected neural network model, and further constructing the power distribution covariance matrix of the user based on the data stream number, the power distribution covariance matrix of the user is not required to be calculated by using the existing water injection algorithm, so that the problems of large calculation amount and high complexity caused by repeated iterative calculation of the water injection algorithm are avoided, and the efficiency of distributing transmission power is improved.

A specific embodiment of a deep learning-based power distribution apparatus provided in an embodiment of the present invention corresponds to the flow shown in fig. 1, and referring to fig. 4, fig. 4 is a schematic structural diagram of a deep learning-based power distribution apparatus according to an embodiment of the present invention, including:

a first obtaining module 301, configured to obtain a channel matrix of a user.

The processing module 302 is configured to perform singular value decomposition processing on the channel matrix to obtain an equivalent channel eigenvalue and a left unitary matrix, where the equivalent channel eigenvalue is a singular value corresponding to each of a plurality of orthogonal channels included in an equivalent channel corresponding to the channel matrix.

A generating module 303, configured to generate channel characteristic information of the user based on the equivalent channel characteristic value and a preset power limiting parameter of the user.

An input module 304, configured to input the channel characteristic information into a pre-trained fully-connected neural network model to obtain a data flow number of the user, where the data flow number is used to represent a data flow size allocated to the user.

A first constructing module 305, configured to construct a power allocation covariance matrix of a user based on the number of data streams of the user, an equivalent channel eigenvalue, and a left unitary matrix.

An allocating module 306, configured to allocate transmission power for the user based on the power allocation covariance matrix of the user.

As an optional implementation manner of the embodiment of the present invention, when the number of users is one, the user is in an MIMO channel scenario, and the first building module is specifically configured to:

calculating a power distribution covariance matrix of the single user based on the equivalent channel characteristic value of the single user, the left unitary matrix and the data stream number of the single user;

the dispensing module has means for:

and allocating transmission power for the single user based on the power allocation covariance matrix of the single user.

As an optional implementation manner of the embodiment of the present invention, when the number of the users is multiple, the users are in a multiple access channel MIMO-MAC channel scenario, and the first building module is specifically configured to:

aiming at each user in a plurality of users, calculating a power distribution covariance matrix of each user based on an equivalent channel characteristic value, a left unitary matrix and a data stream number corresponding to each user;

the dispensing module has means for:

and distributing transmission power for each user based on the power distribution covariance matrix corresponding to each user.

As an optional implementation manner of the embodiment of the present invention, the processing module is specifically configured to:

carrying out singular value decomposition processing on the channel matrix by using a first preset expression, wherein the first preset expression is as follows:

SVD(H)＝UΛV^H

wherein H represents a channel matrix, SVD (H) represents singular value decomposition processing on H, U represents a left unitary matrix, Λ represents a diagonal matrix of H, V represents a right unitary matrix, and V represents^HRepresenting the conjugate transpose of V.

As an optional implementation manner of the embodiment of the present invention, on the basis of the apparatus shown in fig. 4, as shown in fig. 5, the deep learning based power distribution apparatus according to the embodiment of the present invention may further include:

the second building module 401 is configured to build an initial fully-connected neural network model, where the initial fully-connected neural network model at least includes 1 input layer, 3 hidden layers, and 1 output layer.

A second obtaining module 402, configured to obtain sample channel characteristic information of the sample channel matrix and a data stream number label corresponding to each sample channel characteristic information.

The training module 403 is configured to input the sample channel characteristic information and the data stream number label corresponding to the sample channel characteristic information into the initial fully-connected neural network model, and train to obtain the fully-connected neural network model.

The power distribution device based on deep learning provided by the embodiment of the invention obtains a channel matrix of a user, carries out singular value decomposition processing on the channel matrix through a preset singular value decomposition algorithm to obtain an equivalent channel characteristic value and a left unitary matrix, then generates channel characteristic information of the user based on the equivalent channel characteristic value and a preset power limiting parameter of the user, inputs the channel characteristic information into a pre-trained fully-connected neural network model to obtain the data stream number of the user, further constructs a power distribution covariance matrix of the user based on the data stream number, the equivalent channel characteristic value and the left unitary matrix, and then distributes transmission power to the user based on the power distribution covariance matrix of the user, thus, the embodiment of the invention can obtain the data stream number of the user through the fully-connected neural network model and then constructs the power distribution covariance matrix of the user based on the data stream number, the power distribution covariance matrix of the user is not required to be calculated by using the existing water injection algorithm, so that the problems of large calculation amount and high complexity caused by repeated iterative calculation of the water injection algorithm are avoided, and the efficiency of distributing transmission power is improved.

In order to compare the power distribution method of the embodiment of the invention with the existing water injection algorithm, the embodiment of the invention carries out a simulation experiment, uses a complex Gaussian channel as a channel model, and under the condition of a single user, the variation range of the signal-to-noise ratio is as follows:

MIMO employs two configurations：n_R＝n_T∈{4,8}。

Fig. 6 is a system rate comparison diagram of the deep learning-based power allocation method and the existing water filling algorithm according to the embodiment of the present invention. As can be seen from fig. 6, compared with the existing water filling algorithm, the performance of the power allocation method based on deep learning according to the embodiment of the present invention is very close to that of the existing water filling algorithm, because: the embodiment of the invention can obtain accurate data flow number through the fully-connected neural network model, so that the user power distribution covariance matrix obtained by the embodiment of the invention is mostly the optimal power distribution matrix, and better system rate can be obtained as the optimal power distribution algorithm, namely a water injection method. Simulation results show that the power distribution method of the embodiment of the invention can be well close to a single-user water injection algorithm.

Fig. 7 is a graph comparing the accuracy of determining the number of data streams for different antenna numbers according to an embodiment of the present invention. The simulation uses a complex gaussian channel as a channel model, with signal-to-noise ratio in the case of a single user:

the number of antennas varies within a range of: n is_R＝n_TThe number of simulation samples N per number of antennas is 10000: 4: 32. As can be seen from fig. 7, in the embodiment of the present invention, when the signal-to-noise ratio is 5dB and 10dB, the correct rate of the number of data streams output by the fully-connected neural network model is all above 85%, the number of erroneous data streams output by the neural network is also close to the correct number of data streams, and when the output of the neural network is equal to or greater than the correct number of data streams, the optimal user power allocation covariance matrix can be obtained. The simulation result shows that the power distribution method of the embodiment of the invention can select the number of the user data streams with high accuracy.

Fig. 8 is a comparison graph of rate cumulative distribution functions of the power allocation method of the embodiment of the present invention and the existing water filling algorithm, where a complex gaussian channel is used as a channel model, and the number of users is: k4, the limited power of user i is (3i +2) dB, MIMO employs two configurations: n is_R＝n_TWith 8, the maximum number of iterations is set to n_max4. As can be seen from fig. 8, the power allocation method of the embodiment of the present invention has very close performance compared to the existing water-filling algorithm. The reason is that: the embodiment of the invention can well approach the optimal power distribution covariance matrix of each user in each iteration. The multi-user power allocation is to use the interference and noise effects of the other users as equivalent channels, so that the multi-user power allocation can be regarded as the power allocation to the equivalent channels. Therefore, the embodiment of the invention can determine the selection of the number of data streams through the neural network of the equivalent channel to obtain the power distribution covariance matrix of the user. Simulation results show that the power distribution method of the embodiment of the invention can better approach the water injection power distribution method of MIMO-MAC.

An embodiment of the present invention further provides an electronic device, as shown in fig. 9, which includes a processor 501, a communication interface 502, a memory 503 and a communication bus 504, where the processor 501, the communication interface 502 and the memory 503 complete mutual communication through the communication bus 504,

a memory 503 for storing a computer program;

the processor 501, when executing the program stored in the memory 503, implements the following steps:

acquiring a channel matrix of a user;

performing singular value decomposition processing on the channel matrix to obtain an equivalent channel eigenvalue and a left unitary matrix, wherein the equivalent channel eigenvalue is a singular value corresponding to each of a plurality of orthogonal channels included in an equivalent channel corresponding to the channel matrix;

generating channel characteristic information of the user based on the equivalent channel characteristic value and a preset power limiting parameter of the user;

inputting the channel characteristic information into a pre-trained fully-connected neural network model to obtain the data flow number of a user, wherein the data flow number is used for expressing the data flow size distributed for the user;

constructing a power distribution covariance matrix of the user based on the data stream number of the user, the equivalent channel characteristic value and the left unitary matrix;

and allocating transmission power for the users based on the power allocation covariance matrix of the users.

In the electronic device provided by the embodiment of the invention, after a channel matrix of a user is obtained, singular value decomposition processing is performed on the channel matrix through a preset singular value decomposition algorithm to obtain an equivalent channel characteristic value and a left unitary matrix, then channel characteristic information of the user is generated based on the equivalent channel characteristic value and a preset power limiting parameter of the user, the channel characteristic information is input into a pre-trained fully-connected neural network model to obtain a data stream number of the user, a power distribution covariance matrix of the user is constructed based on the data stream number of the user, the equivalent channel characteristic value and the left unitary matrix, then a covariance matrix is distributed for the user based on the power of the user, transmission power is distributed for the user, and thus, the embodiment of the invention can obtain the data stream number of the user through the fully-connected neural network model and then construct the power distribution covariance matrix of the user based on the data stream number, the power distribution covariance matrix of the user is not required to be calculated by using the existing water injection algorithm, so that the problems of large calculation amount and high complexity caused by repeated iterative calculation of the water injection algorithm are avoided, and the efficiency of distributing transmission power is improved.

The communication bus mentioned in the electronic device may be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The communication bus may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is shown, but this does not mean that there is only one bus or one type of bus.

The communication interface is used for communication between the electronic equipment and other equipment.

The Memory may include a Random Access Memory (RAM) or a non-volatile Memory (non-volatile Memory), such as at least one disk Memory. Optionally, the memory may also be at least one memory device located remotely from the processor.

The Processor may be a general-purpose Processor, and includes a Central Processing Unit (CPU), a Network Processor (NP), and the like; the device can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other Programmable logic device, a discrete Gate or transistor logic device, or a discrete hardware component.

An embodiment of the present invention further provides a computer-readable storage medium, in which a computer program is stored, and is configured to execute the following steps:

acquiring a channel matrix of a user;

performing singular value decomposition processing on the channel matrix to obtain an equivalent channel eigenvalue and a left unitary matrix, wherein the equivalent channel eigenvalue is a singular value corresponding to each of a plurality of orthogonal channels included in an equivalent channel corresponding to the channel matrix;

generating channel characteristic information of the user based on the equivalent channel characteristic value and a preset power limiting parameter of the user;

inputting the channel characteristic information into a pre-trained fully-connected neural network model to obtain the data flow number of a user, wherein the data flow number is used for expressing the data flow size distributed for the user;

constructing a power distribution covariance matrix of the user based on the data stream number of the user, the equivalent channel characteristic value and the left unitary matrix;

and allocating transmission power for the users based on the power allocation covariance matrix of the users.

In the computer-readable storage medium provided by the embodiment of the invention, after a channel matrix of a user is obtained, singular value decomposition processing is performed on the channel matrix through a preset singular value decomposition algorithm to obtain an equivalent channel characteristic value and a left unitary matrix, then channel characteristic information of the user is generated based on the equivalent channel characteristic value and a preset power limiting parameter of the user, then the channel characteristic information is input into a pre-trained fully-connected neural network model to obtain a data stream number of the user, a power distribution covariance matrix of the user is constructed based on the data stream number of the user, the equivalent channel characteristic value and the left unitary matrix, a covariance matrix of the power distribution of the user is constructed based on the power distribution covariance matrix of the user, and thus transmission power is distributed to the user, the embodiment of the invention can obtain the data stream number of the user through the fully-connected neural network model, and then the covariance matrix of the power distribution of the user, the power distribution covariance matrix of the user is not required to be calculated by using the existing water injection algorithm, so that the problems of large calculation amount and high complexity caused by repeated iterative calculation of the water injection algorithm are avoided, and the efficiency of distributing transmission power is improved.

For the apparatus/electronic device/storage medium embodiment, since it is substantially similar to the method embodiment, the description is relatively simple, and for the relevant points, reference may be made to part of the description of the method embodiment.

It should be noted that, the apparatus, the electronic device and the storage medium according to the embodiments of the present invention are respectively an apparatus, an electronic device and a storage medium to which the power allocation method based on deep learning is applied, and all embodiments of the power allocation method based on deep learning are applicable to the apparatus, the electronic device and the storage medium, and can achieve the same or similar beneficial effects.

It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.

All the embodiments in the present specification are described in a related manner, and the same and similar parts among the embodiments may be referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the system embodiment, since it is substantially similar to the method embodiment, the description is simple, and for the relevant points, reference may be made to the partial description of the method embodiment.

The above description is only for the preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention shall fall within the protection scope of the present invention.