RSMA beam forming prediction method and device in non-ground network based on deep learning

Abstract

The invention relates to the field of beam forming control, and discloses an RSMA beam forming prediction method and device in a non-ground network based on deep learning, and the method comprises the steps: obtaining historical CSI information of a plurality of channels; determining a precoder construction parameter and a common rate allocation parameter for performing beam forming on the plurality of channels according to the historical CSI information of the plurality of channels through a beam forming prediction model; wherein the beamforming prediction model comprises a Transform unit, a CNN unit and a generation unit, the Transform unit is used for extracting CSI time dependent features in historical CSI data of a plurality of channels, the CNN unit is used for extracting spatial correlation features among the historical CSI data of the plurality of channels, and the generation unit is used for generating the spatial correlation features of the historical CSI data of the plurality of channels. The generation unit is used for generating precoder construction parameters and common rate allocation parameters for beamforming a plurality of channels. According to the method, the spatial-temporal characteristic beam forming prediction is extracted from the historical CSI through the transformer and the convolutional neural network, and the application effect of the RSMA technology is improved.

Description

Technical Field

[0001] The present application relates to the field of beamforming control technology, and more specifically, to a method and device for predicting RSMA beamforming in a non-terrestrial network based on deep learning. Background Art

[0002] In the communication process of low earth orbit (LEO) satellites and drones, rate-splitting multiple access (RSMA) technology makes non terrestrial networks (NTNs) feasible by virtue of interference management and reliable data transmission. However, the precoder design in RSMA technology still relies on accurate channel state information (CSI) feedback and complex optimization, which faces great challenges in the actual deployment of RSMA technology.

[0003] In the communication process between low-orbit satellites and UAVs, beamforming is required to ensure the communication quality between low-orbit satellites and UAVs. In the traditional predictive beamforming design, when the low-orbit satellite performs beamforming prediction for each UAV, in the last round of communication between the low-orbit satellite and the UAV (corresponding to time t-1), each UAV first reports its own CSI to the low-orbit satellite; after collecting the CSI of all UAVs reported in the last round of communication (corresponding to time t-1), the low-orbit satellite continues to predict the channel state information (CSIT) at the transmitter corresponding to the UAV in the next round of communication (corresponding to time t+1), so that the low-orbit satellite can use the CSIT prediction results in the prediction stage to coordinate and reduce the interference between the channels of multiple UAVs through RSMA technology. However, in the process of realizing beamforming through this beamforming prediction method and reducing channel interference through RSMA technology, data needs to be transmitted in different modules, and the data processing time is long, resulting in poor application effect of reducing channel interference through RSMA technology. Summary of the invention

[0004] The purpose of this application is to provide a method and device for RSMA beamforming prediction in a non-terrestrial network based on deep learning, which solves the technical problems that data needs to be transmitted in different modules and the data processing time is long in traditional beamforming prediction, and achieves the technical effect of extracting spatiotemporal features from historical CSI through transformer and convolutional neural network for beamforming prediction, thereby improving the application effect of RSMA technology.

[0005] An embodiment of the present application provides a deep learning-based RSMA beamforming prediction method in a non-terrestrial network, the method comprising: obtaining historical CSI information of multiple channels; determining, through a beamforming prediction model, precoder construction parameters and common rate allocation parameters for beamforming the multiple channels according to the historical CSI information of the multiple channels; wherein the beamforming prediction model comprises a Transformer unit, a CNN unit and a generation unit, the Transformer unit is used to extract CSI time-dependent features from the historical CSI data of the multiple channels, the CNN unit is used to extract spatial correlation features between the historical CSI data of the multiple channels, and the generation unit is used to generate precoder construction parameters and common rate allocation parameters for beamforming the multiple channels.

[0006] In a possible implementation, the method also includes: respectively representing the real part and the imaginary part of the historical CSI information of multiple channels by real numbers, and inputting the real part and the imaginary part of the historical CSI information represented by real numbers into the Transformer unit of the beamforming prediction model; the generation unit outputs a complex-valued matrix corresponding to the precoder construction parameters for beamforming the multiple channels.

[0007] In another possible implementation, the Transformer unit includes a multi-head attention layer, a first residual connection and normalization layer, a feedforward network layer, and a second residual connection and normalization layer arranged in sequence; the CNN unit includes a convolutional layer, a pooling layer, and a flattening layer arranged in sequence; and the generation unit includes a fully connected layer and a generation layer.

[0008] In another possible implementation, the common rate allocation parameters output by the beamforming prediction model are normalized and meet the total power limit of the low-orbit satellite.

[0009] In another possible implementation, when the common rate allocation parameter output by the beamforming prediction model is normalized and satisfies the total power limit of the low-orbit satellite, it is expressed by the following formula:

[0010]

[0011] tr(O H O)=P t

[0012] in, represents the channel between the low-orbit (LEO) satellite and the kth UAV or channel k, k represents the UAV number or channel number, γ k represents the signal-to-noise ratio (SINR) of the public stream decoded by the drone or channel k, represents the total number of drones, represents the transposed channel vector between the low-orbit satellite and the kth UAV, o c represents a common linear precoder, represents the mth private linear precoder, σ k represents the variance of the distribution of the additive Gaussian white noise at the kth drone, represents the instantaneous reachability, represents the complex value matrix output by the beamforming prediction model, O represents the precoding matrix, Representation output The calculated achievable common rate, min means to find the minimum value, C k represents the public rate part of the kth UAV, represents the normalized absolute value, tr represents the trace of the solution matrix, P t Indicates the transmission power of low-orbit satellites.

[0013] In another possible implementation, a beamforming prediction model is trained on-orbit for a low-orbit satellite, and the method includes: acquiring historical CSI information of multiple channels; the low-orbit satellite trains the beamforming prediction model on-orbit according to the historical CSI information of the multiple channels; wherein the loss function of the beamforming prediction model is expressed by the following formula:

[0014]

[0015] Among them, L represents the loss function value, represents the expected value of the total downlink rate obtained by the kth UAV, u k represents the weight of the obtained rate, Φ represents the number of samples of historical CSI information, represents the instantaneous rate obtained under the corresponding sample;

[0016] The process of updating the beamforming prediction model can be expressed by the following formula:

[0017] w n+1 =w n -η n g n

[0018] η n =η 0 *λ n

[0019] Among them, η n Represents the learning rate of the nth round of training, η 0 represents the initial learning rate, λ n Represents the decay ratio of the learning rate, w n represents the weight parameter of the global model in the nth round of training, w n+1represents the weight parameter of the global model in the n+1th round of training, g n represents the local estimate of the gradient of the loss function in satellite s.

[0020] An embodiment of the present application also provides a deep learning-based RSMA beamforming prediction device in a non-terrestrial network, comprising a unit for executing any of the methods described above.

[0021] An embodiment of the present application also provides a deep learning-based RSMA beamforming prediction device in a non-terrestrial network, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the methods described above when executing the computer program.

[0022] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in any one of the above items is implemented.

[0023] An embodiment of the present application also provides a low-orbit satellite, which uses the RSMA beamforming prediction method in a non-ground network based on deep learning as described in any of the above items when performing beamforming prediction.

[0024] An embodiment of the present application also provides a low-orbit satellite, which uses the on-orbit training method of the beamforming prediction model as described above when training the beamforming prediction model.

[0025] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0026] The embodiment of the present application provides a method for predicting RSMA beamforming in a non-terrestrial network based on deep learning, and the method includes: obtaining historical CSI information of multiple channels; determining the precoder construction parameters and common rate allocation parameters for beamforming multiple channels according to the historical CSI information of multiple channels through a beamforming prediction model; wherein the beamforming prediction model includes a Transformer unit, a CNN unit and a generation unit, the Transformer unit is used to extract the CSI time-dependent features in the historical CSI data of multiple channels, the CNN unit is used to extract the spatial correlation features between the historical CSI data of multiple channels, and the generation unit is used to generate the precoder construction parameters and common rate allocation parameters for beamforming multiple channels. The method in the embodiment of the present application can accurately predict beamforming, reduce data transmission time, and improve the application effect of reducing channel interference through RSMA technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0028] Figure 1 A flowchart of a method for predicting RSMA beamforming in a non-terrestrial network based on deep learning provided in an embodiment of the present application;

[0029] Figure 2 A schematic diagram of a working scenario of a RSMA beamforming prediction method in a non-terrestrial network based on deep learning provided in an embodiment of the present application;

[0030] Figure 3 A schematic diagram of the structure of a beamforming prediction model provided in an embodiment of the present application;

[0031] Figure 4 A performance comparison diagram of a RSMA beamforming prediction method in a non-terrestrial network based on deep learning provided in an embodiment of the present application;

[0032] Figure 5 A flowchart of an on-orbit training method for a beamforming prediction model provided in an embodiment of the present application;

[0033] Figure 6 A schematic diagram of the logical structure of a deep learning-based RSMA beamforming prediction device in a non-terrestrial network provided in an embodiment of the present application;

[0034] Figure 7 A schematic diagram of the physical structure of an RSMA beamforming prediction device in a non-terrestrial network based on deep learning provided in an embodiment of the present application. DETAILED DESCRIPTION

[0035] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0036] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0037] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0038] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0039] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0040] In the prior art, in the process of realizing beamforming through beamforming prediction method and reducing channel interference through RSMA technology, data needs to be transmitted in different modules, and the data processing time is long, resulting in poor application effect of reducing channel interference through RSMA technology.

[0041] Based on the above reasons, an embodiment of the present application provides a RSMA beamforming prediction method in a non-terrestrial network based on deep learning, and the method includes: obtaining historical CSI information of multiple channels; through a beamforming prediction model, according to the historical CSI information of multiple channels, determining the precoder construction parameters and common rate allocation parameters for beamforming multiple channels; wherein the beamforming prediction model includes a Transformer unit, a CNN unit and a generation unit, the Transformer unit is used to extract the CSI time-dependent features in the historical CSI data of multiple channels, the CNN unit is used to extract the spatial correlation features between the historical CSI data of multiple channels, and the generation unit is used to generate the precoder construction parameters and common rate allocation parameters for beamforming multiple channels. The method in the embodiment of the present application can accurately predict beamforming, reduce data transmission time, and improve the application effect of reducing channel interference through RSMA technology.

[0042] In some scenarios, a deep learning-based RSMA beamforming prediction method in a non-ground network of an embodiment of the present application can be applied to the communication process supporting RSMA in a non-ground network composed of low-orbit satellites and drones. It can accurately predict the beamforming, reduce the data transmission time, and improve the application effect of reducing channel interference through RSMA technology.

[0043] In other scenarios, the RSMA in a non-terrestrial network based on deep learning in an embodiment of the present application, the beamforming prediction method can also be applied to the communication process between low-orbit satellites and ground mobile terminals that support RSMA. It can accurately predict the beamforming during the communication process between low-orbit satellites and ground mobile terminals, reduce data transmission time, and improve the application effect of reducing channel interference through RSMA technology.

[0044] The following is a detailed description of a deep learning-based RSMA beamforming prediction method in a non-terrestrial network provided in an embodiment of the present application with reference to specific examples.

[0045] Figure 1 A flowchart of a method for predicting RSMA beamforming in a non-terrestrial network based on deep learning is provided in an embodiment of the present application, such as Figure 1 As shown, the method includes S110 to S120, and S110 to S120 are described in detail below.

[0046] Figure 2 A schematic diagram of a working scenario of a RSMA beamforming prediction method in a non-terrestrial network based on deep learning provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the method in the embodiment of the present application can be applied to the communication between low-orbit satellites and multiple drones. When communicating, the low-orbit satellites and multiple drones can manage interference and ensure reliable data transmission through rate split multiple access (RSMA) technology, thereby ensuring the feasibility of non-terrestrial networks (NTNs).

[0047] S110. Obtain historical CSI information of multiple channels.

[0048] Figure 3 A schematic diagram of the structure of a beamforming prediction model provided in an embodiment of the present application is shown in FIG. Figure 3 As shown, when working, the method in the embodiment of the present application realizes beamforming prediction for multiple channels through deep learning technology, so the historical CSI information of multiple channels can be obtained first, and a historical CSI information includes channel state information (CSI) of a channel.

[0049] like Figure 2 As shown, when a low-orbit satellite and multiple drones communicate, there can be multiple channels, each channel can obtain the channel state information (CSI) corresponding to the channel, and then can obtain the historical CSI information of the channel, and then can predict the beamforming through the historical CSI information.

[0050] S120. Determine, through a beamforming prediction model, the precoder construction parameters and common rate allocation parameters for beamforming the multiple channels according to the historical CSI information of the multiple channels. The beamforming prediction model includes a Transformer unit, a CNN unit, and a generation unit. The Transformer unit is used to extract the CSI time-dependent features in the historical CSI data of the multiple channels. The CNN unit is used to extract the spatial correlation features between the historical CSI data of the multiple channels. The generation unit is used to generate the precoder construction parameters and common rate allocation parameters for beamforming the multiple channels.

[0051] like Figure 3 As shown, the beamforming prediction model can be used to directly determine the precoder construction parameters and common rate allocation parameters used to perform beamforming on multiple channels according to the historical CSI information of multiple channels. The precoder construction parameters obtained in the beamforming prediction can be directly used to adjust the signal transmission direction and strength in the RSMA technology to improve the signal transmission efficiency and reception quality. At the same time, the common rate allocation parameters are used to split the transmission rate of each user into multiple parts to adapt to different transmission conditions and requirements.

[0052] like Figure 3 As shown, the beamforming prediction model may include a Transformer unit, a CNN unit and a generation unit. When working, after inputting the CSI history information of multiple channels, the Transformer unit (Transformer module), the CNN unit (convolution module) and the generation unit (generation module) successively complete the feature extraction of the historical CSI data of multiple channels, and generate the precoder construction parameters and common rate allocation parameters for beamforming.

[0053] When working, the Transformer unit is used to extract CSI time-dependent features from historical CSI data of multiple channels. The Transformer unit can capture complex time series dependencies in multiple historical CSI information. Through the attention mechanism, the Transformer dynamically extracts features of different parts of multiple historical CSI information when making predictions.

[0054] When working, the CNN unit is used to extract the spatial correlation features between the historical CSI data of multiple channels. The convolutional neural network (CNN) can extract complex nonlinear features from the matrix input. Since the beamforming prediction also needs to deal with the interference management between different user channels, the spatial dependency between different channels during beamforming prediction can be processed by the CNN unit.

[0055] When working, the generating unit is used to generate precoder construction parameters and common rate allocation parameters for beamforming multiple channels.

[0056] It should be noted that the beamforming prediction model in the embodiment of the present application can be abbreviated as the TranCN model. TranCN integrates the Transformer and CNN architectures and provides a new method to improve the effectiveness of the predicted beamforming strategy.

[0057] Figure 4 A performance comparison diagram of a RSMA beamforming prediction method in a non-terrestrial network based on deep learning provided in an embodiment of the present application, such as Figure 4 As shown, Figure 4 The horizontal axis represents the signal-to-noise ratio (SNR), and the vertical axis represents the weighted ergodic sum rate (WESR) performance under different LEO satellite signal-to-noise ratios (SNR). The performance of the TranCN model in the embodiment of the present application is better than the performance when only the Transformer model works, only the CNN model works, and only the SCA model works.

[0058] The beneficial effect brought about by the above implementation method is that the precoder construction parameters and common rate allocation parameters for beamforming of multiple channels can be directly generated, the beamforming can be accurately predicted, the data transmission time is reduced, and the application effect of reducing channel interference through RSMA technology is improved.

[0059] The beneficial effect brought about by the above implementation method is that the integration of Transformer units and CNN units can effectively capture the spatial and temporal dependent features in CSI data, thereby improving the overall performance of beamforming prediction of RSMA in NTNs.

[0060] In some implementations, the above method further includes S111 and S121, and S111 and S121 are described in detail below.

[0061] S111. The real part and the imaginary part of the historical CSI information of the multiple channels are respectively represented by real numbers, and the real part and the imaginary part of the historical CSI information of the multiple channels represented by real numbers are input into the Transformer unit of the beamforming prediction model.

[0062] During operation, in order to accurately reflect the historical CSI information of multiple channels, the real part and the imaginary part of the historical CSI information of multiple channels can be represented by real numbers respectively, and the real part and the imaginary part of the historical CSI information represented by real numbers are input into the Transformer unit of the beamforming prediction model.

[0063] For example, Figure 3 As shown, the historical CSI information of multiple channels may be from time t-1 to time tt 0 Multiple historical CSI information, and the time t-1 to time tT 0 The multiple historical CSI information is used as the basic input of the beamforming prediction model (TranCN).

[0064] Exemplarily, multiple historical CSI information can be represented by a matrix H, where H represents the historical channel coefficients. H can be represented by The historical channel coefficient of the first UAV is Superscript represents the round of the communication cycle, The subscript 1 of represents the serial number of the UAV, and the historical channel coefficients of other UAVs are represented in a similar way. Among them, H represents the matrix corresponding to multiple historical CSI information, T 0 Indicates time, represents the historical channel coefficient of the previous communication cycle, Indicates the last round of communication cycle, represents the drone with serial number 1, t represents the time, Indicates the serial number is UAV, N t represents the number of antennas, K represents the total number of drones, Indicates size is T 0 ×KN t The plural set of Means “…is defined as…”.

[0065] When performing beamforming prediction, in order to ensure the availability of gradient information during the training process, multiple historical CSI information may be split into real and imaginary parts, and the real and imaginary parts of the multiple historical CSI information may be represented by real numbers.

[0066] When inputting the model, the real and imaginary parts of the historical CSI information of multiple channels represented by real numbers can be represented as a time series, recorded as in, represents the matrix corresponding to the historical CSI information. The real and imaginary parts of the channel coefficients in the matrix are represented by real numbers. R represents a real number set. Indicates size is T 0 ×2KN t The set of real numbers.

[0067] This conversion method can encapsulate the real part and the imaginary part of the historical CSI information of multiple channels, and retain the basic channel angle information of the historical CSI information of multiple channels input.

[0068] S121. A generating unit outputs a complex-valued matrix corresponding to precoder construction parameters for beamforming a plurality of channels.

[0069] When the generation unit is working, the generation unit can output a complex-valued matrix corresponding to the precoder construction parameters for beamforming multiple channels. When constructing the precoder, the complex-valued matrix corresponding to the precoder construction parameters can be split to obtain the real part and the imaginary part of the precoder construction parameters.

[0070] Exemplarily, the complex value matrix corresponding to the precoder construction parameters can be expressed as a complex value matrix in, represents the precoder construction parameters represented by complex numbers, Indicates size N t ×(K+1) complex numbers.

[0071] The beneficial effect brought by the above implementation method is that by splitting multiple historical CSI information into real parts and imaginary parts, the availability of gradient information is guaranteed during the training process.

[0072] The beneficial effect brought by the above implementation manner is that the generation unit can output a complex-valued matrix corresponding to the precoder construction parameters for beamforming multiple channels, which is convenient for directly constructing the precoder.

[0073] In some implementations, the Transformer unit includes a multi-head attention layer, a first residual connection and normalization layer, a feed-forward network layer, and a second residual connection and normalization layer arranged in sequence.

[0074] like Figure 3 As shown, the Transformer unit may include a multi-head attention layer, a first residual connection and normalization layer, a feedforward network layer, and a second residual connection and normalization layer arranged in sequence.

[0075] Structurally, there are D self-attention layers in the multi-head attention layer. The original input (multiple historical CSI information) is linearly transformed to produce three matrices: in, Indicates size is T 0 × 1 These transformations can be expressed by formula (1):

[0076]

[0077] In formula (3), Represents the matrix corresponding to the historical CSI information of multiple channels, the matrix The real and imaginary parts of the channel coefficients are represented by real numbers, i represents the number of heads in the multi-head attention layer, and Z represents the parameter matrix, where ι 1 represents the dimension of the feature, Indicates size is 2KN t × 1 where M represents a set of real numbers, M represents a set of multi-head attention layers, Q, V, and K represent the Q matrix, V matrix, and K matrix in the Transformer unit, respectively.

[0078] When working, the output of the self-attention layer can be expressed by formula (2):

[0079]

[0080] In formula (2), Attention(Q,V,K) represents the output of the self-attention layer, and softmax represents the normalization function.

[0081] When working, the outputs of all self-attention layers are connected in the multi-head attention layer using the Concat layer. When working, the output of the multi-head attention layer can be expressed by formula (3):

[0082] X 1 =Concat(ω 1 ,…,ω D )Z o (3)

[0083] In formula (3), X 1 represents the output of the multi-head attention layer, D represents the number of attention layers, and Concat represents the Concat layer.

[0084] Furthermore, in order to enhance the model's attention to the difference between the input (X) and the output (F(X)), the present embodiment introduces a residual connection and a normalization layer. The residual connection and the normalization layer include a normalization layer, which can be expressed by formula (4):

[0085] Add&Norm(X,F(X))=Norm(F(X))+X (4)

[0086] In formula (4), X represents the input of the same layer, F(X) represents the output of the multi-head attention layer or the feed-forward layer, and Add&Norm(X,F(X)) represents the normalization and residual connection of X and F(X).

[0087] In the embodiment of the present application, a feed-forward layer is formed by two fully connected (FC) layers, and the feed-forward layer can be expressed by formula (5):

[0088] Feed Forward(X 2 )=max(A 1 X 1 +b 1 ,0)A 2 +b 2 (5)

[0089] In formula (5), A 1 X 1 +b 1 Indicates X 1 Linear transformation, A 1 and b 1 is the linear transformation parameter, A 2 and b 2 represents the linear transformation parameter, max(A 1 X 1 +b 1 ,0) means taking the maximum value. If A 1 X 1 +b 1 >0, then output A 1 X 1 +b 1 , otherwise output 0, Indicates that X 2 Represents the output of a Transformer unit.

[0090] In some implementations, the CNN unit includes a convolutional layer, a pooling layer, and a flattening layer arranged in sequence.

[0091] like Figure 3As shown, the CNN unit includes a convolution layer, a pooling layer, and a flattening layer arranged in sequence, and CNN can be used to extract complex nonlinear features from the matrix input. The beamforming prediction model in this application can collect the spatial correlation between channels with different interferences by merging CNN modules.

[0092] like Figure 3 As shown in the figure, the initial layer of the CNN unit uses convolution operations, which can use 2×2 filters to convolve with the input data, so that basic features can be identified between different interference channels. Subsequently, in order to further simplify the generated features and reduce the spatial dimension of the features, an average pooling operation is introduced in the pooling layer, which helps to extract key information while maintaining the basic features required for subsequent processing. The last layer of the CNN unit performs a flattening operation, which can reshape the output into a one-dimensional vector This reshaping facilitates seamless integration with subsequent layers and ensures the continuity of information flow in the beamforming prediction model. The output information of the CNN unit can capture nonlinear features from mutually interfering channels.

[0093] In some implementations, the generation unit includes a fully connected layer and a generation layer.

[0094] like Figure 3 As shown, the generation unit includes a fully connected layer and a generation layer, which work together to parameterize the beamforming strategy. The fully connected layer is the first part, and the fully connected layer can help the model learn details from the extracted features. The fully connected layer completes this operation through a fully connected operation and LeakyReLU. The fully connected layer can capture nonlinear relationships in the feature space. LeakyReLU activation introduces a controllable amount of nonlinearity, enhancing the model's ability to recognize and absorb complex patterns.

[0095] After the fully connected layer, the generative layer is seamlessly integrated to reshape the output, thereby obtaining the pre-encoder construction parameters and common rate allocation parameters Fully connected layers and generative layers can refine the learned features into precise and actionable parameters.

[0096] The beneficial effect of the above implementation is that the use of multiple attention heads in the Transformer unit enhances the function of the Transformer unit, and each attention head of the Transformer unit can learn different relationships in the data, so that the model can capture a wide range of dependencies, including long-range dependencies. The Transformer unit plays a vital role in extracting the high-dimensional features and temporal correlations embedded in the CSI historical information, improving the predictive ability of beamforming prediction.

[0097] The beneficial effect brought about by the above implementation method is that in the Transformer unit, residual connections and normalization layers are introduced, which enhances the attention to the difference between the input (X) and output (F(X)) of each layer and improves the effect of beamforming prediction using multiple historical CSI information.

[0098] In some implementations, the common rate allocation parameters output by the beamforming prediction model are normalized and meet the total power limit of the low-orbit satellite.

[0099] During operation, the common rate allocation parameters output by the beamforming prediction model can be normalized first to facilitate the calculation of the sum of the common rate allocation parameters output by the beamforming prediction model, and the common rate allocation parameters output by the beamforming prediction model meet the total power limit of the low-orbit satellite, thereby enabling the common rate allocation in RSMA to be performed under the total power limit of the low-orbit satellite.

[0100] The beneficial effect brought about by the above implementation method is that the public rate allocation parameters output by the beamforming prediction model meet the total power limit of the low-orbit satellite, and thus can effectively perform public rate allocation in RSMA under the total power limit of the low-orbit satellite.

[0101] In some implementations, when the common rate allocation parameters output by the beamforming prediction model are normalized and the total power limit of the low-orbit satellite is met, it is expressed by formulas (5) to (9):

[0102]

[0103] tr(O H O)=P t (12)

[0104] In formulas (6) to (12), represents the channel between the low-orbit (LEO) satellite and the kth UAV or channel k, k represents the UAV number or channel number, γ k represents the signal-to-noise ratio (SINR) of the public stream decoded by the drone or channel k, represents the total number of drones, represents the transposed channel vector between the low-orbit satellite and the kth UAV, o c represents a common linear precoder, represents the mth private linear precoder, σ k represents the variance of the distribution of the additive Gaussian white noise at the kth drone, represents the instantaneous reachability, represents the complex value matrix output by the beamforming prediction model, O represents the precoding matrix, Representation output The calculated achievable common rate, min means to find the minimum value, C k represents the public rate part of the kth UAV, represents the normalized absolute value, tr represents the trace of the solution matrix, P t Indicates the transmission power of low-orbit satellites.

[0105] The beneficial effect brought about by the above implementation method is that it is easy to calculate the common rate allocation parameters output by the beamforming prediction model to meet the total power limit of the low-orbit satellite.

[0106] The embodiment of the present application also provides an on-orbit training method of a beamforming prediction model, which is used for on-orbit training of a beamforming prediction model for a low-orbit satellite. Figure 5 A flowchart of an on-orbit training method for a beamforming prediction model provided in an embodiment of the present application is shown in FIG. Figure 5 As shown, the method includes S210 to S220, and S210 to S220 are described in detail below.

[0107] S210: Obtain historical CSI information of multiple channels.

[0108] S220, the low-orbit satellite trains the beamforming prediction model on-orbit according to the historical CSI information of multiple channels. The low-orbit satellite has on-orbit computing capability, and the loss function of the beamforming prediction model is expressed by formula (13) and formula (14):

[0109]

[0110] In formula (13) and formula (14), L represents the loss function value, represents the expected value of the total downlink rate obtained by the kth UAV, u k represents the weight of the obtained rate, Φ represents the number of samples of historical CSI information, Represents the instantaneous rate obtained under the corresponding sample.

[0111] Through the above formulas (13) and (14), the original optimization problem P0 is converted into a trainable loss function, which enables the model to learn and adjust its parameters according to the desired objectives.

[0112] In some implementations, the process of updating the beamforming prediction model can be expressed by formula (15) and formula (16):

[0113] w n+1 =w n -η n g n (15)

[0114] η n =η 0 *λ n (16)

[0115] In formula (15) and formula (16), η n Represents the learning rate of the nth round of training, η 0 represents the initial learning rate, λ n Represents the decay ratio of the learning rate, w n represents the weight parameter of the global model in the nth round of training, w n+1 represents the weight parameter of the global model in the n+1th round of training, g n represents the local estimate of the gradient of the loss function in satellite s.

[0116] The beneficial effect brought about by the above implementation method is that the above training method is continued until the model converges, thereby ensuring the training effect of the model.

[0117] An embodiment of the present application also provides a deep learning-based RSMA beamforming prediction device in a non-terrestrial network, comprising a unit for executing any of the methods described above.

[0118] Figure 6 A schematic diagram of the logical structure of a RSMA beamforming prediction device in a non-terrestrial network based on deep learning is provided in an embodiment of the present application, such as Figure 6 As shown, the device 1 of this embodiment includes a processing unit 11, a storage unit 12 and a transceiver unit 13. The processing unit 11 is used to process data, the storage unit 12 is used to store data, and the transceiver unit 13 is used to send and receive data. The processing unit 11, the storage unit 12 and the transceiver unit 13 cooperate with each other to implement the above method. The beneficial effects brought by the embodiment of the present application have been described in the above method and will not be repeated here.

[0119] An embodiment of the present application also provides a deep learning-based RSMA beamforming prediction device in a non-terrestrial network, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the methods described above when executing the computer program.

[0120] Figure 7 A schematic diagram of the physical structure of a RSMA beamforming prediction device in a non-terrestrial network based on deep learning is provided in an embodiment of the present application, such as Figure 7 As shown, the device 2 of this embodiment includes: at least one processor 20 ( Figure 7Only one processor 20 is shown in the figure), a memory 21, and a computer program 22 stored in the memory 21 and executable on the at least one processor 20. When the processor 20 executes the computer program 22, the steps in any of the above-mentioned method embodiments are implemented. The beneficial effects brought about by the embodiments of the present application have been described in the above-mentioned methods and will not be repeated here.

[0121] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0122] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0123] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in any one of the above items is implemented.

[0124] An embodiment of the present application also provides a low-orbit satellite, which uses the RSMA beamforming prediction method in a non-ground network based on deep learning as described in any of the above items when performing beamforming prediction.

[0125] An embodiment of the present application also provides a low-orbit satellite, which uses the on-orbit training method of the beamforming prediction model as described above when training the beamforming prediction model.

[0126] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can at least include: any entity or device that can carry the computer program code to the camera / terminal device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, RandomAccess Memory), electric carrier signal, telecommunication signal and software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk or an optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0127] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0128] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0129] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0130] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0131] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A deep learning-based RSMA beamforming prediction method in non-terrestrial networks, It is characterized in that The method comprises: Obtain historical CSI information of multiple channels; Through a beamforming prediction model, according to historical CSI information of multiple channels, precoder construction parameters and common rate allocation parameters for beamforming multiple channels are determined; wherein the beamforming prediction model includes a Transformer unit, a CNN unit and a generation unit, the Transformer unit is used to extract CSI time-dependent features in historical CSI data of multiple channels, the CNN unit is used to extract spatial correlation features between historical CSI data of multiple channels, and the generation unit is used to generate precoder construction parameters and common rate allocation parameters for beamforming multiple channels.

2. The method according to claim 1, It is characterized in that The method further comprises: The real part and the imaginary part of the historical CSI information of the multiple channels are respectively represented by real numbers, and the real part and the imaginary part of the historical CSI information represented by real numbers are input into the Transformer unit of the beamforming prediction model; The generating unit outputs a complex value matrix corresponding to the precoder construction parameters for beamforming the multiple channels.

3. The method according to claim 2, It is characterized in that The Transformer unit includes a multi-head attention layer, a first residual connection and normalization layer, a feedforward network layer, and a second residual connection and normalization layer arranged in sequence; The CNN unit includes a convolutional layer, a pooling layer, and a flattening layer arranged in sequence; The generation unit includes a fully connected layer and a generation layer.

4. The method according to claim 3, It is characterized in that The common rate allocation parameters output by the beamforming prediction model are normalized and meet the total power limit of the low-orbit satellite.

5. The method according to claim 4, It is characterized in that When the common rate allocation parameters output by the beamforming prediction model are normalized and meet the total power limit of the low-orbit satellite, it is expressed by the following formula: ash H O)=P t in, represents the channel between the low-orbit (LEO) satellite and the kth UAV or channel k, k represents the UAV number or channel number, γ k represents the signal-to-noise ratio (SINR) of the public stream decoded by the drone or channel k, represents the total number of drones, represents the transposed channel vector between the low-orbit satellite and the kth UAV, o c represents a common linear precoder, represents the mth private linear precoder, σ k represents the variance of the distribution of the additive Gaussian white noise at the kth drone, represents the instantaneous reachability, represents the complex value matrix output by the beamforming prediction model, O represents the precoding matrix, Representation output The calculated achievable common rate, min means to find the minimum value, C k represents the public rate part of the kth UAV, represents the normalized absolute value, tr represents the trace of the solution matrix, P t Indicates the transmission power of low-orbit satellites.

6. An on-orbit training method for a beamforming prediction model, It is characterized in that For training a beamforming prediction model for a low-orbit satellite on-orbit, the method comprises: Obtain historical CSI information of multiple channels; The low-orbit satellite trains the beamforming prediction model on-orbit based on the historical CSI information of multiple channels. The low-orbit satellite has on-orbit computing capabilities, and the loss function of the beamforming prediction model is expressed by the following formula: Among them, L represents the loss function value, represents the expected value of the total downlink rate obtained by the kth UAV, u k represents the weight of the obtained rate, Φ represents the number of samples of historical CSI information, represents the instantaneous rate obtained under the corresponding sample; The process of updating the beamforming prediction model can be expressed by the following formula: w n+1 =w n -or n g n or n =the 0 *l n Among them, η n Represents the learning rate of the nth round of training, η 0 represents the initial learning rate, λ n Represents the decay ratio of the learning rate, w n represents the weight parameter of the global model in the nth round of training, w n+1 represents the weight parameter of the global model in the n+1th round of training, g n represents the local estimate of the gradient of the loss function in satellite s.

7. A deep learning-based RSMA beamforming prediction device in non-terrestrial networks, It is characterized in that The method comprises means for performing the method according to any one of claims 1 to 5.

8. A deep learning-based RSMA beamforming prediction device in a non-terrestrial network, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the computer program, the method according to any one of claims 1 to 5 is implemented.

9. A computer-readable storage medium storing a computer program. It is characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

10. A low-orbit satellite, It is characterized in that When performing beamforming prediction, the RSMA beamforming prediction method in a non-ground network based on deep learning described in any one of claims 1 to 5 is adopted, or when training a beamforming prediction model, the on-orbit training method of the beamforming prediction model described in claim 6 is adopted.

Images