Multi-beam satellite precoding method and device

Abstract

The invention provides a multi-beam satellite pre-coding method and device, and the method comprises the following steps: carrying out the coding based on the signal-to-noise ratio data of a current receiving end, and obtaining a first coding vector; obtaining a modulation mode of the sending end to the current receiving end, mapping the modulation mode to obtain a corresponding mapping value, and encoding the mapping value to obtain a second encoding vector; constructing an input vector based on the first coding vector and the second coding vector, inputting the input vector to a preset ResMLP layer, the ResMLP layer being provided with a parameter generation network, a dynamic affine transformation layer, a linear layer and an adaptive gating residual connection layer, the parameter generation network outputting a scaling weight vector and a shift weight vector, and the shift weight vector outputting a shift weight vector; calculating an output embedded vector based on the input vector, the scaled weight vector, and the shifted weight vector; and inputting the output embedded vector into an output layer, and outputting a modulo operation frame size parameter through an activation function layer of the output layer.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of satellite communication, and in particular to a multi-beam satellite precoding method and device. BACKGROUND

[0002] In a multi-user MIMO system, in order to effectively solve the problem of inter-beam interference, a nonlinear precoding algorithm is usually used for transmission optimization. In this technology, the sending end calculates the precoding matrix according to the channel state information fed back by the receiving end, and performs nonlinear modulo operation processing on the sending symbol vector through a specific algorithm, thereby significantly suppressing inter-beam interference. The modulo operation block is related to the modulation mode and is fixed throughout the process. Compared with the traditional linear precoding algorithm, this algorithm introduces nonlinear modulo operation in the precoding process, which can effectively constrain the power amplification of the sending signal, reduce the influence of noise in the receiving end, and thus improve the system bit error rate performance.

[0003] However, in the traditional nonlinear precoding algorithm, the size of the modulo operation block is usually fixed, but this will result in significant modulo loss at low signal-to-noise ratio and serious power loss at high signal-to-noise ratio, both of which affect the system bit error rate performance. SUMMARY

[0004] In view of this, the embodiments of the present application provide a multi-beam satellite precoding method to eliminate or improve one or more defects in the prior art.

[0005] One aspect of the present application provides a multi-beam satellite precoding method, the method comprising the following steps: Encoding based on the signal-to-noise ratio data of the current receiving end to obtain a first encoding vector; Obtaining the modulation mode of the sending end for the current receiving end, mapping the modulation mode to obtain a corresponding mapping value, encoding the mapping value to obtain a second encoding vector; Based on the first encoding vector and the second encoding vector, an input vector is constructed, and the input vector is input into a preset ResMLP layer, the ResMLP layer being provided with a parameter generation network, a dynamic affine transformation layer, a linear layer and an adaptive gated residual connection layer, the parameter generation network outputting a scaling weight vector and a shift weight vector, and an output embedding vector being calculated based on the input vector, the scaling weight vector and the shift weight vector; The output embedding vector is input into an output layer, and a modulo operation block size parameter is output through an activation function layer of the output layer.

[0006] By adopting the above scheme, the network architecture of this scheme achieves an end-to-end deep customization for predicting the parameter values ​​of the dynamic optimal modulus calculation operation box through dedicated preprocessing of communication data in the input layer, the introduction of a dynamic adaptive mechanism for communication feature awareness in the ResMLP layer, and the imposition of strict physical constraints in the output layer. This scheme enables the model to not only have a strong fitting ability, but also to deeply integrate prior knowledge of the communication scenario, thereby significantly improving the performance of nonlinear precoding in complex channel environments.

[0007] In some embodiments of the present invention, the method further includes feeding back the modulo operation frame size parameter to the sending end, so that the sending end performs nonlinear precoding based on the modulo operation frame size parameter and sends the encoded data to the receiving end.

[0008] In some embodiments of the present invention, in the steps of acquiring the signal-to-noise ratio (SNR) data of each receiving end, encoding the SNR data, and obtaining a first encoding vector, the SNR data of each receiving end is subjected to logarithmic transformation and processed through a linear layer to obtain the first encoding vector.

[0009] In some embodiments of the present invention, in the step of encoding the mapping value to obtain a second encoding vector, the mapping value is first encoded by discrete encoding and then processed by an embedding layer to obtain the second encoding vector.

[0010] In some embodiments of the present invention, in the step of the parameter generation network outputting the scaling weight vector and the shift weight vector, the parameter generation network is provided with two channels, each channel having at least one linear layer, the input vector is input into the parameter generation network, and the two channels of the parameter generation network output the scaling weight vector and the shift weight vector respectively.

[0011] In some embodiments of the present invention, the network structure of the ResMLP layer includes a parameter generation network and sequentially connected dynamic affine transformation layer, linear layer, activation function layer, linear layer, dynamic affine transformation layer and adaptive gated residual connection layer. In the step of calculating the output embedding vector based on the input vector, scaling weight vector and shift weight vector, the input vector, scaling weight vector and shift weight vector are input into the first dynamic affine transformation layer, and the output embedding vector is obtained through the adaptive gated residual connection layer.

[0012] In some embodiments of the present invention, the step of inputting the input vector, scaling weight vector, and shift weight vector into the first dynamic affine transformation layer and obtaining the output embedding vector through an adaptive gated residual connection layer is expressed by the following formula: in, This indicates the output embedding vector. Represents the input vector. This represents the preset gate vector. This indicates the processing of the first dynamic affine transformation layer. This indicates the processing of the second dynamic affine transformation layer. This represents the parameter matrix of the first linear layer. This represents the parameter matrix of the second linear layer.

[0013] In some embodiments of the present invention, in the step of outputting the modulo operation box size parameter through the activation function layer of the output layer, the output layer includes a linear layer and an activation function layer arranged in sequence. The process parameter is obtained through the linear layer of the output layer, and the process parameter is input to the activation function layer. The activation function layer outputs the modulo operation box size parameter.

[0014] In some embodiments of the present invention, in the step of inputting the process parameters into the activation function layer, and the activation function layer outputting the modulo operation box size parameter, the activation function layer calculates the modulo operation box size parameter based on the following formula: in, This indicates the size parameter of the modulo operation box. Indicates process parameters.

[0015] A second aspect of the present invention also provides a multi-beam satellite precoding apparatus, the apparatus including a computer device including a processor and a memory, the memory storing computer instructions, the processor executing the computer instructions stored in the memory, and the apparatus performing the steps of the method described above when the computer instructions are executed by the processor.

[0016] A third aspect of the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the aforementioned multi-beam satellite precoding method.

[0017] Additional advantages, objects, and features of the invention will be set forth in part in the description which follows, and will also become apparent in part to those skilled in the art upon studying the text, or may be learned by practice of the invention. The objects and other advantages of the invention will become apparent from the description and the accompanying drawings.

[0018] Those skilled in the art will understand that the objectives and advantages achievable with the present invention are not limited to those specifically described above, and that the above and other objectives achievable with the present invention will become clearer from the following detailed description. Attached Figure Description

[0019] The accompanying drawings, which are provided to further illustrate the invention and form part of this application, are not intended to limit the scope of the invention.

[0020] Figure 1 This is a schematic diagram of one embodiment of the multi-beam satellite precoding method of the present invention; Figure 2 This is a schematic diagram of another embodiment of the multi-beam satellite precoding method of the present invention; Figure 3 This is a schematic diagram illustrating the overall logic of this solution; Figure 4 This is a schematic diagram of the principle architecture of this solution; Figure 5 This is a flowchart of a non-linear precoding process; Figure 6 A schematic diagram of the modulo operation block; Figure 7 This is a schematic diagram showing the optimal size of the modulo operation box; Figure 8 This is a schematic diagram of the combined structure of the input layer, ResMLP layer, and output layer. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and descriptions of this invention are used to explain the invention, but are not intended to limit the invention.

[0022] It should also be noted that, in order to avoid obscuring the invention with unnecessary details, only the structures and / or processing steps closely related to the solution according to the invention are shown in the accompanying drawings, while other details that are not closely related to the invention are omitted.

[0023] Introduction to existing technologies: Prior art 1 proposes a nonlinear precoding method for MIMO systems. At the transmitter, a precoding matrix is ​​constructed based on the channel correlation matrix to better utilize the statistical characteristics of the channel. Subsequently, modulo operations are performed on the data symbols, and a perturbation vector is superimposed to constrain the transmit power, with the size of the modulo operation frame remaining constant throughout. At the receiver, the perturbation term coefficients are normalized by equalizing only the coefficients before the data symbols, thereby completely eliminating the perturbation vector during demodulation through modulo operations. This effectively suppresses inter-beam interference and improves the system's bit error rate performance.

[0024] Existing technology 2 proposes a dual-mode nonlinear precoding method for 5G / NR (New Radio) systems, comprising two operating modes: Mode 1 is based on explicit beamforming CSI (Channel State Information), and Mode 2 is based on explicit complete downlink CSI. The frame size remains constant throughout the precoding process at the transmitter in both modes. The base station dynamically selects the mode based on the UE (User Equipment) capabilities and channel quality, and notifies the UE via DCI (Downlink Control Information) signaling. The receiver differentiates its DMRS processing based on the mode indication: Mode 1 requires combined DMRS (Demodulation Reference Signal) 1 and DMRS 2, while Mode 2 only requires DMRS 2. The estimated CSI is then used to detect the received data. This method improves system throughput performance while maintaining compatibility.

[0025] The drawbacks of existing technologies 1 and 2 are as follows: Traditional nonlinear precoding techniques in existing technologies 1 and 2 fail to fully consider the spatial geometric relationship between satellite beam positions and users when modeling the channel matrix, leading to a mismatch between the precoding matrix and the actual channel characteristics, and a weakened inter-beam interference cancellation effect. Furthermore, the size of the modulus calculation frame is fixed throughout the precoding process, making it unable to adapt to dynamically changing satellite channel environments. In poor channel environments, this may result in significant modulus loss; in good channel environments, it may cause unnecessary power loss, leading to a decrease in system bit error rate performance.

[0026] Existing technology 3 is based on a dynamically adjusted adaptive scheme. By monitoring the modulation scheme and channel state information in real time, it intelligently selects the optimal size of the modulus calculation frame. Simultaneously, it employs a predefined feedback index table to achieve efficient information exchange between the transmitting and receiving ends. This technology not only effectively improves the performance of nonlinear precoding but also significantly saves valuable feedback resources. In one scheme of existing technology 3, the system adjusts the size of the modulo operation frame according to the transport layer identifier: when the transport layer identifier increases, the size of the modulo operation frame is reduced to suppress the increasing interference; while when the transport layer identifier is below a preset threshold, an infinitely large modulo operation frame is used to avoid unnecessary signal processing. The transmitting end efficiently instructs the UE with the optimized modulo operation frame parameters through DCI control signaling combined with a predefined parameter mapping table, and the receiving end performs demodulation operation based on these parameters. This method not only improves the system's bit error rate performance but also significantly reduces signaling overhead through the predefined mapping table. In another scheme of existing technology 3, the base station adaptively selects the modulus calculation frame size based on channel quality and CSI accuracy: increasing the frame size when the signal-to-noise ratio is low and decreasing it when the signal-to-noise ratio is high. A predefined candidate frame size set table and scaling factor table are used, and the optimal frame size is dynamically calculated by combining the modulation scheme with the scaling factor. The base station sends the modulation scheme and frame size index to the UE via DCI signaling, achieving efficient parameter interaction.

[0027] The drawbacks of existing technology 3 are as follows: The traditional adaptive modulo operation frame adjustment algorithm in existing technology 3 dynamically adjusts the size of the modulo operation frame based on factors such as channel quality and interference. However, the adjustment mechanism only employs a simple linear mapping relationship: increasing the size of the modulo operation frame when the channel quality deteriorates and decreasing it when it improves. This coarse-grained adjustment cannot accurately match subtle changes in the channel state. Furthermore, the large spacing between adjacent modulo operation frame sizes in the traditional feedback index table, along with this coarse-grained quantification, prevents the receiver from accurately feeding back the optimal modulo operation frame size, leading to additional performance loss in bit error rate.

[0028] In nonlinear precoding, the size of the modulus calculation frame is typically determined by the modulation scheme and remains unchanged. However, in low signal-to-noise ratio (SNR) scenarios with high noise levels, a small frame size can lead to significant modulus loss and severe bit errors. Conversely, in high SNR scenarios, a large frame size can result in substantial power loss and poor precoding performance. Therefore, it is necessary to dynamically adjust the frame size based on the actual channel conditions to balance the impact of modulus loss and power loss.

[0029] like Figure 1 As shown, this invention proposes a multi-beam satellite precoding method, which includes the following steps: Step S110: Encode the signal-to-noise ratio data at the current receiver to obtain the first encoding vector; Step S120: Obtain the modulation scheme of the transmitter for the current receiver, map the modulation scheme to obtain the corresponding mapping value, and encode the mapping value to obtain the second coding vector; Step S200: Construct an input vector based on the first encoding vector and the second encoding vector, and input the input vector into a preset ResMLP layer. The ResMLP layer is configured with a parameter generation network, a dynamic affine transformation layer, a linear layer and an adaptive gated residual connection layer. The parameter generation network outputs a scaling weight vector and a shift weight vector, and calculates an output embedding vector based on the input vector, the scaling weight vector and the shift weight vector. Specifically, the ResMLP layer in this solution can be multiple.

[0030] Step S300: The output embedding vector is input into the output layer, and the modulo operation box size parameter is output through the activation function layer of the output layer.

[0031] like Figure 8 As shown, this scheme constructs a network architecture based on a residual multilayer perceptron, mainly consisting of an input layer, a ResMLP layer, and an output layer. The core function of this network is to intelligently predict the optimal modulus calculation frame size under the given channel conditions, using the receiver's signal-to-noise ratio and modulation scheme as input features. During model training, T receiver SNR data points and T corresponding modulation scheme data points are used as input data vectors, which are then processed by the input layer to obtain T data points. The T embedding vectors are then fed into a series of ResMLP (Residual Multi-Layer Perceptron) layers, ultimately outputting T vectors. The embedding vector is dimensional. Finally, through output layer processing, the optimal T predictions are obtained. Data. This architecture specifically comprises three core components: (1) Input layer: The continuous SNR data from the receiving end first undergoes a logarithmic transformation, i.e. To transform its exponential distribution characteristics into an approximately linear distribution more conducive to neural network learning, it is then processed through a linear layer to generate T values. The embedding vector is dimensional. Meanwhile, discrete modulation scheme data (e.g., QPSK, 16QAM, 64QAM, etc., mapped to integer values ​​0, 1, 2, etc.) are processed through an embedding layer to generate T embedding vectors. An embedding vector of dimension. The embedding layer maps each modulation scheme to a dense, continuous vector. In a dimensional vector space, it is used to automatically capture the intrinsic relationships between different modulation schemes, as well as the optimal modulation scheme. The influence pattern of the values. Finally, the two embedding vectors are fused and concatenated into T values. An embedding vector of dimension, where The aforementioned input layer processing collectively enhances the neural network model's ability to achieve dynamic optimization. The value predicts the adaptability of this specific communication scenario.

[0032] (2) ResMLP layer: The ResMLP layer contains multiple sub-layers with the same structure. In each sub-layer, the normalization operation is implemented through feature-aware dynamic affine transformation, and its mathematical expression can be represented as: in, and It is dynamically generated. dimensional weight vector, Corresponding scaling, Corresponding shift, by dimensional input embedding vector This is generated in real time through a parameter-based generation network. This mechanism utilizes... The receiver SNR and modulation scheme feature information encoded in the ResMLP layer are used to generate affine transformation parameters that are highly matched with them, enabling the ResMLP layer to extract key features more quickly and accurately, and achieve faster convergence during training.

[0033] Secondly, each sub-layer contains an adaptive gated residual connection, which no longer uses a fixed equal-weight fusion method; its output is determined by... dimensional gated vector Input embedding vectors of sub-layers The output embedding vectors, obtained after internal transformations within the sublayers, are weighted and fused during model training to obtain T vectors. dimensional output embedding vector , represented as: During model training, Process each column vector in, and This is the main weight matrix of the ResMLP layer, with dimensions as follows: and .

[0034] This adaptive gated residual connection introduces a gate vector dynamically generated from feature information. This enables the network to autonomously adjust the input embedding vector based on real-time SNR data and modulation scheme data from the receiver. The ratio between the input embedding vector and the transformed output embedding vector. As described above, adaptive gated residual connections enable the network to dynamically determine, for different combinations of receiver SNR and modulation schemes, whether to rely more on the input embedding vector to maintain the stability of the sublayer or more on the transformed output embedding vector to pursue higher accuracy. This significantly enhances the model's adaptability and robustness to complex communication environments. linear layer (3) Output layer: T units After the dimensional output embedding vector passes through a linear layer, T output data are obtained. .Then, After applying the Softplus activation function, the T optimal predictions are finally obtained. Data, i.e. : The Softplus activation function is introduced to impose a nonnegativity constraint on the output prediction data, ensuring that it strictly conforms to... Physical definition of parameters.

[0035] This network architecture achieves a dynamic optimization-oriented approach through dedicated preprocessing of communication data in the input layer, a dynamic adaptive mechanism that is aware of communication features introduced in the ResMLP layer, and strict physical constraints imposed in the output layer. End-to-end deep customization of value prediction, such as Figure 8 As shown in the figure. This design not only enables the model to have a strong fitting ability, but also deeply integrates prior knowledge of the communication scenario, thereby significantly improving the performance of nonlinear precoding in complex channel environments.

[0036] By adopting the above scheme, the network architecture of this scheme achieves an end-to-end deep customization for predicting the parameter values ​​of the dynamic optimal modulus calculation operation box through dedicated preprocessing of communication data in the input layer, the introduction of a dynamic adaptive mechanism for communication feature awareness in the ResMLP layer, and the imposition of strict physical constraints in the output layer. This scheme enables the model to not only have a strong fitting ability, but also to deeply integrate prior knowledge of the communication scenario, thereby significantly improving the performance of nonlinear precoding in complex channel environments.

[0037] like Figure 2 As shown, in some embodiments of the present invention, the method further includes step S400, feeding back the modulo operation frame size parameter to the sending end, so that the sending end performs nonlinear precoding based on the modulo operation frame size parameter and sends the encoded data to the receiving end.

[0038] In the specific implementation process, in the step of feeding back the modulo operation box size parameter to the sending end, the corresponding feedback bit is determined based on the preset feedback index table, and the feedback bit is sent to the sending end. The sending end compares the feedback bit with the feedback index table to determine the actual application modulo operation box size parameter.

[0039] In the specific implementation process, in the step of determining the actual application modulo operation frame size parameter by comparing the feedback bits with the feedback index table, the transmitting end calculates the average value of the modulo operation frame size parameter of each receiving end as the actual application modulo operation frame size parameter.

[0040] Using the above scheme, the sending and receiving ends share a feedback index table. During the transmission of the modulo operation box size parameter, it is not necessary to actually transmit the modulo operation box size parameter, thus preventing leakage of the modulo operation box size parameter during data transmission. Specifically, the feedback index table can be as shown in Table 1 below, where... The QAM modulation order is... Distance between adjacent constellation points in the constellation diagram: Table 1 This scheme employs a high-precision feedback index table. The quantization range adaptively adjusts according to the modulation scheme. Specifically, the lower bound of the quantization range is set as the base value related to the modulation scheme minus an offset, and the upper bound is set as the base value plus another offset. Uniform quantization is performed with a fixed step size to ensure sufficient coverage of the predicted value by the quantized value under different modulation schemes. The receiver selects the closest quantized value from the feedback index table based on the predicted optimal modulus calculation frame size and feeds the index value back to the transmitter. Only a small number of feedback bits are needed to accurately feed back the optimal modulus calculation frame size to the transmitter while ensuring low overhead.

[0041] In some embodiments of the present invention, in the steps of acquiring the signal-to-noise ratio (SNR) data of each receiving end, encoding the SNR data, and obtaining a first encoding vector, the SNR data of each receiving end is subjected to logarithmic transformation and processed through a linear layer to obtain the first encoding vector.

[0042] In some embodiments of the present invention, in the step of encoding the mapping value to obtain a second encoding vector, the mapping value is first encoded by discrete encoding and then processed by an embedding layer to obtain the second encoding vector.

[0043] In some embodiments of the present invention, in the step of the parameter generation network outputting the scaling weight vector and the shift weight vector, the parameter generation network is provided with two channels, each channel having at least one linear layer, the input vector is input into the parameter generation network, and the two channels of the parameter generation network output the scaling weight vector and the shift weight vector respectively.

[0044] like Figure 8 As shown, in some embodiments of the present invention, the network structure of the ResMLP layer includes a parameter generation network and sequentially connected dynamic affine transformation layer, linear layer, activation function layer, linear layer, dynamic affine transformation layer and adaptive gated residual connection layer. In the step of calculating the output embedding vector based on the input vector, scaling weight vector and shift weight vector, the input vector, scaling weight vector and shift weight vector are input into the first dynamic affine transformation layer, and the output embedding vector is obtained through the adaptive gated residual connection layer.

[0045] In some embodiments of the present invention, the step of inputting the input vector, scaling weight vector, and shift weight vector into the first dynamic affine transformation layer and obtaining the output embedding vector through an adaptive gated residual connection layer is expressed by the following formula: in, This indicates the output embedding vector. Represents the input vector. This represents the preset gate vector. This indicates the processing of the first dynamic affine transformation layer. This indicates the processing of the second dynamic affine transformation layer. This represents the parameter matrix of the first linear layer. This represents the parameter matrix of the second linear layer.

[0046] In some embodiments of the present invention, in the step of outputting the modulo operation box size parameter through the activation function layer of the output layer, the output layer includes a linear layer and an activation function layer arranged in sequence. The process parameter is obtained through the linear layer of the output layer, and the process parameter is input to the activation function layer. The activation function layer outputs the modulo operation box size parameter.

[0047] In some embodiments of the present invention, in the step of inputting the process parameters into the activation function layer, and the activation function layer outputting the modulo operation box size parameter, the activation function layer calculates the modulo operation box size parameter based on the following formula: in, This indicates the size parameter of the modulo operation box. Indicates process parameters.

[0048] This scheme establishes a channel model suitable for satellite communication scenarios, improving the performance of nonlinear precoding and enhancing the feedback accuracy of the modulus calculation frame under limited feedback resources. In multi-beam satellite communication scenarios, complex spatial correlations and interference characteristics exist between different beams, and are influenced by factors such as user distribution, making it difficult for traditional channel modeling methods to accurately reflect the multi-beam satellite channel state. In nonlinear precoding, the modulus calculation frame parameters are usually fixed values, related to the modulation scheme. This static setting cannot adapt to the highly dynamic characteristics of satellite channels, significantly limiting precoding performance. Furthermore, in satellite communication scenarios with limited feedback resources, it is necessary to feed back the most accurate optimal modulus calculation frame size to the transmitter based on the known feedback index table at both the transmitter and receiver. Therefore, how to quickly and accurately select the optimal modulus calculation frame size based on time-varying modulation schemes, channel quality, and other multi-dimensional information, and achieve high-precision, low-overhead feedback of this parameter under limited feedback resource constraints, is the core key to overcoming existing performance bottlenecks and further improving the performance of nonlinear precoding in multi-beam satellite communication systems. The key issues are as follows: (1) Design of a dynamic adjustment algorithm for the modulus calculation frame in nonlinear precoding based on multi-beam satellite communication scenarios: The fixed size of the modulus calculation frame in current nonlinear precoding is difficult to adapt to the highly dynamic satellite channel environment. The core is to design a dynamic adjustment algorithm for the modulus calculation frame, which can accurately and quickly calculate the optimal size of the modulus calculation frame based on real-time changes in modulation scheme, channel quality and other information, so as to maximize the inter-beam interference cancellation performance of nonlinear precoding in multi-beam satellite communication scenarios.

[0049] (2) Design of an efficient feedback index table for the optimal modulo operation frame: Currently, the feedback index table has shortcomings such as coarse-grained quantization and a single mapping relationship, which makes it impossible to accurately reflect the size of the optimal modulo operation frame under limited feedback resources. The core is to design an efficient feedback index table that performs high-precision quantization and intelligent mapping of the size of the optimal modulo operation frame, converting it into a low-bit discrete index, so as to achieve efficient feedback of the size of the optimal modulo operation frame while ensuring feedback accuracy.

[0050] In the specific implementation process, the overall operating logic of this solution is as follows: Figure 3 and 4 As shown, the processing steps of the present invention are as follows: Step 3: Step 1: Construct a satellite channel model. Based on the off-axis angle between different users and each beam, calculate the transmit antenna gain of each beam in different user directions. Combine this with the receive antenna gain, the distance between each user and the satellite, and other information to calculate the satellite channel matrix. Based on the satellite channel model, construct a signal transmission model and calculate the received signals for all users, including the signal portion required by that user, the inter-beam interference portion, and the noise portion.

[0051] In step 1, a satellite channel model is constructed. Based on the off-axis angle between different users and each beam, the transmit antenna gain of each beam in different user directions is calculated. Combined with the receive antenna gain, the distance between each user and the satellite, and other information, the satellite channel matrix is ​​calculated.

[0052] Satellite communication relies primarily on direct sunlight, meaning the signal propagation path is unobstructed by buildings or terrain, resulting in high channel stability and relatively concentrated signal strength. However, due to the long-distance propagation between the satellite and the ground terminal, free-space path loss increases significantly, and the signal also experiences atmospheric attenuation, including rain attenuation, cloud absorption, and ionospheric scintillation.

[0053] These characteristics mean that satellite channels possess both highly stable direct-light components and must consider the effects of long-distance propagation and atmospheric attenuation. Satellite Channel Matrix Specifically, it can be expressed as: in, This refers to the number of spot beams that a multi-beam satellite can generate simultaneously. This refers to the number of UEs within the coverage area of ​​a multi-beam satellite. For Hadamard product, The channel gain matrix is... The phase shift matrix introduced by factors such as ionospheric scintillation. The element middle, For the first The beam at the 1st Transmit antenna gain in the user direction, For the first Individual user's receive antenna gain For carrier wavelength, For the first The distance from each user to a multi-beam satellite, Boltzmann's constant, Equivalent noise temperature For bandwidth. The element middle, For the first The beam to the first The phase change experienced by the signal of each user, assuming .

[0054] According to the The beam and the first Off-axis angle of each user That is, satellites and the first The line connecting the center points of each beam to the satellite and the first beam center point The angle between the user connections It can be represented as: in, Category 1 The order Bessel function. Represented as: in, This is the half-power beamwidth. for The maximum transmit antenna gain at that time.

[0055] Based on the satellite channel model, a signal transmission model is constructed to calculate the received signals for all users, including the signal component required by the user, the inter-beam interference component, and the noise component.

[0056] After the transmitted signal vector passes through the satellite channel and arrives at the receiver, the received signal vector can be represented as: in, For the signal vectors received by all UEs, The satellite channel matrix calculated in (1-1) is... For the first User channel vectors for each user For the signal vectors transmitted by all beams, AWGN (Additive White Gaussian Noise) is superimposed on the received signal.

[0057] Specifically, the The signal received by a user can be represented as: in, Indicates the first The user and the first The channel coefficients between beams. In the above formula, the first part on the right is the channel coefficient between the beams. The signal required by each user. The second part is the signal emitted by other beams for the first user. Interference caused by individual users, i.e., inter-beam interference. The third part indicates the interference caused by individual users. The impact of noise on individual users.

[0058] Step 2: After obtaining the satellite channel matrix, the transmitter performs LQ decomposition on it and uses the resulting L and Q matrices to calculate the three types of filters required in the nonlinear precoding process. The calculated three filters are then used to perform nonlinear precoding on the modulation symbol vector. First, a feedback filter is used to cancel continuous interference in the modulation symbols, obtaining the transmitted symbol vector. Simultaneously, to constrain its energy, a modulus operation is performed, where the size of the modulus operation frame typically depends on the modulation scheme. Then, the vector passes through a scaling filter and a feedforward filter sequentially to obtain the transmitted signal vector. After the transmitted signal vector reaches the receiver via the satellite channel, it undergoes nonlinear precoding processing at the receiver to obtain the received symbol vector required by the user, eliminating inter-beam interference. The initial modulus operation frame size is kept constant throughout the process, and a training set containing the receiver's signal-to-noise ratio, modulation scheme, and optimal modulus operation frame size is obtained.

[0059] In step 2, after the satellite channel matrix is ​​known, the transmitting end performs LQ decomposition on it and uses the obtained L and Q matrices to calculate the three types of filters required in the nonlinear precoding process.

[0060] The transmitter has a known satellite channel matrix. In this case, utilize For modulation symbols Nonlinear precoding is performed to eliminate the effects of inter-beam interference, as follows: Figure 5 As shown.

[0061] There are three filters that play a major role in the nonlinear precoding process: feedback filter Feedforward filter and scaling filter To calculate the three types of filters mentioned above, the transmitter first needs to perform LQ decomposition on the satellite channel matrix, as follows: in, It is a lower triangular matrix. It is a unitary matrix. and The three filters required in the nonlinear precoding process can be calculated and expressed as follows: Three calculated filters are used to perform nonlinear precoding on the modulation symbol vector. First, a feedback filter is used to cancel continuous interference in the modulation symbols, obtaining the transmitted symbol vector. Simultaneously, to constrain its energy, a modulus operation is performed, where the size of the modulus operation frame typically depends on the modulation scheme. Then, the vector passes through a scaling filter and a feedforward filter sequentially to obtain the transmitted signal vector. After the transmitted signal vector reaches the receiver via the satellite channel, it undergoes nonlinear precoding processing at the receiver to obtain the received symbol vector required by the user, thus eliminating inter-beam interference.

[0062] In the nonlinear precoding process, the modulation symbols The transmitted symbol is obtained after continuous interference cancellation. , represented as: in, yes The corresponding element. Because sequentially... When performing interference cancellation, The amplitude may exceed the modulation boundary, leading to The energy increase is significant, which is detrimental to subsequent reduction of transmission power. In order to transmit symbols... Within a certain energy constraint range, modulo operations are used in the nonlinear precoding process. right The processing is represented as follows: in, The specific value of the modulo operation frame size usually depends on the selected modulation scheme. When the average symbol energy is 1, ,in The QAM modulation order is... This represents the distance between adjacent constellation points in the constellation diagram. When the modulation scheme is QPSK (Quadrature Phase Shift Keying) and 16QAM (Quadrature Amplitude Modulation), the modulus calculation flowchart is as follows: Figure 6 As shown, proceed Equivalent to in modulation symbols Add a perturbation vector to , represented as: Using feedback filters Performing continuous interference cancellation is equivalent to utilizing Processing is performed, therefore transmitting symbols It can be represented as: Then transmit symbols After filtering and After processing, the transmitted signal vector can be obtained. , represented as: in, The normalization factor in nonlinear precoding is expressed as: When sending signal vector After passing through the satellite channel, the received symbol vector arrives at the receiving end and undergoes modulo operation. Represented as: As can be seen from the above formula, after nonlinear precoding, inter-beam interference can be effectively eliminated, and the transmitted signal energy can be suppressed by modulo operation, reducing the impact of noise at the receiver.

[0063] Step 3: Dynamically adjust the size of the modulus calculation frame for the modulus calculation process performed in Step 2. Construct a network architecture based on a residual multilayer perceptron, mainly consisting of an input encoding section, a residual multilayer perceptron backbone, and a physical constraint output layer. The core function of this network is to intelligently predict the optimal size of the modulus calculation frame under the given channel conditions, using the receiver's signal-to-noise ratio and modulation scheme as input features.

[0064] Step 4: For the network architecture based on residual multilayer perceptron built in Step 3, a lightweight deployment strategy combining offline training and online inference is adopted. First, a training set containing the receiver signal-to-noise ratio, modulation scheme, and optimal modulus calculation frame size is constructed through discrete sampling. The ResMLP layer is trained offline. Then, the trained network is deployed in the UE, and online inference is performed by measuring and inputting channel condition parameters in real time, dynamically outputting predicted values ​​to adapt to the current channel conditions.

[0065] Step 4 is used to train the input layer, ResMLP layer and output layer models corresponding to steps S110 and S120; For the network architecture based on residual multilayer perceptron built in step 3, a lightweight deployment strategy combining offline training and online inference is adopted. First, a training set containing the receiver signal-to-noise ratio, modulation scheme, and optimal modulus calculation frame size is constructed through discrete sampling. The ResMLP network is trained offline. Then, the trained network is deployed in the UE, and online inference is performed by measuring and inputting channel condition parameters in real time, dynamically outputting predicted values ​​to adapt to the current channel conditions.

[0066] Considering that the computational resources required for online learning of the ResMLP network exceed the actual carrying capacity of the UE, the system adopts a combination of offline training and online inference.

[0067] First, construct a system that includes the receiver's SNR (Signal-to-Noise Ratio), modulation scheme, and optimal... The training set of values ​​is used, where the receiver SNR is sampled at a fixed step size within a given interval, and the modulation scheme is represented by discrete coding. For each combination of receiver SNR and modulation scheme, the system BER (Bit Error Rate) after nonlinear precoding in step 2 is minimized by searching at a fixed step size within a specific range. value as optimal The ResMLP network was then trained offline based on this training set.

[0068] Then, the trained ResMLP network is deployed in the UE, and the receiver SNR is measured in real time. The SNR, along with the modulation scheme, is used as the network input. After online inference, the predicted value is obtained. .at this time This is the optimal size of the modulo operation frame for the user under this channel condition.

[0069] This scheme constructs an accurate multi-beam satellite channel matrix model, combining multi-dimensional parameters such as off-axis angle, antenna gain, and inter-satellite distance to provide accurate channel state information for nonlinear precoding. A dynamic adjustment algorithm for the modulus calculation frame based on a residual multilayer perceptron is proposed. Using real-time signal-to-noise ratio and modulation scheme as input, the algorithm predicts the optimal modulus calculation frame size after passing through a pre-trained model. An offline training and online lightweight inference strategy is employed to significantly improve the inter-beam interference cancellation performance of nonlinear precoding in high-dynamic satellite environments.

[0070] Step 5: Each receiver searches for the corresponding index value in the feedback index table based on the predicted value of the modulus operation frame size obtained in Step 4 and feeds it back to the transmitter. The transmitter calculates the average value and determines a unified optimal modulus operation frame size value. It uses this value to perform the modulus operation in the nonlinear precoding at the next time step and sends it to all receivers to ensure that they can complete the corresponding precoding receiver processing.

[0071] Each receiver searches for the corresponding index value in the feedback index table based on the predicted value of the modulus operation frame size obtained in step 4 and feeds it back to the transmitter. The transmitter calculates the average value and determines a unified optimal modulus operation frame size value. It uses this value to perform the modulus operation in the nonlinear precoding at the next time step and sends it to all receivers to ensure that they can complete the corresponding precoding receiver processing.

[0072] Each receiver receives the predicted value. Then, it needs to be fed back to the transmitting end because in nonlinear precoding, the transmitting end and all receiving ends need to use the same modulo operation frame size. This ensures that the receiving end can successfully demodulate the signal.

[0073] The system establishes a feedback index table with known information from both the sender and receiver. After completing step 4, each receiver... Combined with the feedback index table, the corresponding optimal The index value is fed back to the sending end.

[0074] After receiving the indices from all UEs, the sending end determines the optimal index for each UE in the feedback index table. Calculate their average value to obtain a unified optimal value. The value is represented as: Sending end utilizes optimal The value is used for the modulus calculation in the next time step of nonlinear precoding, as follows: Figure 7 As shown, it is simultaneously transmitted to all UEs via the control channel, enabling them to complete the corresponding precoding receiver processing.

[0075] The beneficial effects of this plan include: 1. This invention innovatively proposes a dynamic adjustment scheme for the modulus calculation frame in nonlinear precoding based on neural networks, targeting the highly dynamic channel environment of multi-beam satellite communication systems. By constructing an accurate multi-beam satellite channel model and designing a dynamic adjustment algorithm for the modulus calculation frame based on a residual multilayer perceptron, the size of the modulus calculation frame can be accurately predicted in real time based on information such as modulation scheme and channel quality. This scheme effectively overcomes the limitation of traditional fixed modulus calculation frame sizes, which are difficult to adapt to rapid changes in channel conditions, and significantly improves the performance of nonlinear precoding in inter-beam interference cancellation.

[0076] 2. This invention innovatively designs a high-precision feedback index table dynamically bound to the modulation scheme. The receiver, through a "prediction-table lookup-feedback" implementation process, combined with uniform quantization step size and dynamic range adjustment strategies, achieves fine quantization and feedback of the optimal modulus calculation frame size under limited feedback resources. This scheme achieves high-precision feedback with only a small number of feedback bits, significantly improving system feedback efficiency and providing a practical solution for nonlinear precoding in multi-beam satellite communication systems.

[0077] This invention also provides a multi-beam satellite precoding apparatus, which includes a computer device, a processor, and a memory. The memory stores computer instructions, and the processor executes the computer instructions stored in the memory. When the computer instructions are executed by the processor, the apparatus implements the steps of the method described above.

[0078] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the aforementioned multi-beam satellite precoding method. The computer-readable storage medium can be a tangible storage medium, such as random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, register, floppy disk, hard disk, removable storage disk, CD-ROM, or any other form of storage medium known in the art.

[0079] Those skilled in the art will understand that the exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention. When implemented in hardware, it can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this invention are programs or code segments used to perform the desired tasks. The programs or code segments can be stored in a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried in a carrier wave.

[0080] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.

[0081] In this invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or in place of features of other embodiments.

[0082] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations of the embodiments of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method of multi-beam satellite precoding, the method comprising: The method comprises the following steps: Encoding based on the signal-to-noise ratio data of the current receiving end to obtain a first encoding vector; Obtaining the modulation mode of the sending end to the current receiving end, mapping the modulation mode to obtain a corresponding mapping value, encoding the mapping value to obtain a second encoding vector; Based on the first encoding vector and the second encoding vector, an input vector is constructed, and the input vector is input into a preset ResMLP layer, the ResMLP layer is provided with a parameter generation network, a dynamic affine transformation layer, a linear layer and an adaptive gated residual connection layer, the parameter generation network outputs a scaling weight vector and a shift weight vector, and an output embedding vector is calculated based on the input vector, the scaling weight vector and the shift weight vector; The output embedding vector is input into an output layer, and a modulus operation frame size parameter is output through an activation function layer of the output layer.

2. The multi-beam satellite precoding method of claim 1, wherein, The method further comprises feeding back the modulus operation frame size parameter to the sending end, so that the sending end performs nonlinear precoding based on the modulus operation frame size parameter and sends encoded data to the receiving end.

3. The method of claim 1, wherein, In the step of obtaining the signal-to-noise ratio data of each receiving end and encoding the signal-to-noise ratio data to obtain a first encoding vector, the signal-to-noise ratio data of each receiving end is logarithmically transformed and processed through a linear layer to obtain the first encoding vector.

4. The multi-beam satellite precoding method of claim 1, wherein, In the step of encoding the mapping value to obtain a second encoding vector, the mapping value is first encoded through discrete encoding and then processed through an embedding layer to obtain the second encoding vector.

5. The method of claim 1, wherein, In the step of the parameter generation network outputting a scaling weight vector and a shift weight vector, the parameter generation network is provided with two channels, each of which is provided with at least one linear layer, the input vector is input into the parameter generation network, and the two channels of the parameter generation network output the scaling weight vector and the shift weight vector, respectively.

6. The method of claim 1, wherein, The network structure of the ResMLP layer comprises a parameter generation network and sequentially connected dynamic affine transformation layers, linear layers, activation function layers, linear layers, dynamic affine transformation layers and adaptive gated residual connection layers, in the step of calculating an output embedding vector based on the input vector, the scaling weight vector and the shift weight vector, the input vector, the scaling weight vector and the shift weight vector are input into the first dynamic affine transformation layer, and the output embedding vector is obtained through the adaptive gated residual connection layer.

7. The multi-beam satellite precoding method of claim 6, wherein, The step of inputting the input vector, the scaling weight vector and the shift weight vector into the first dynamic affine transformation layer and obtaining the output embedding vector through the adaptive gated residual connection layer is represented by the following formula: wherein, denotes an output embedding vector, denotes an input vector, denotes a preset gating vector, denotes a processing of a first dynamic affine transformation layer, denotes a processing of a second dynamic affine transformation layer, denotes a parameter matrix of a first linear layer, denotes a parameter matrix of a second linear layer.

8. The method of claim 1, wherein, In the step of outputting a modulus operation frame size parameter through an activation function layer of the output layer, the output layer comprises sequentially arranged linear layers and activation function layers, a process parameter is obtained through the linear layer of the output layer, the process parameter is input into the activation function layer, and the activation function layer outputs the modulus operation frame size parameter.

9. The method of claim 8, wherein, In the step of inputting the process parameter into the activation function layer and outputting the modulus operation frame size parameter by the activation function layer, the activation function layer calculates the modulus operation frame size parameter based on the following formula: wherein, represents a modulo operation block size parameter, represents a process parameter.

10. A multi-beam satellite precoding apparatus, characterized by comprising: The apparatus comprises a computer device comprising a processor and a memory having stored therein computer instructions, the processor being operative to execute the computer instructions stored in the memory, the apparatus implementing the steps of the method as claimed in any one of claims 1 to 9 when the computer instructions are executed by the processor.

Images