Data multi-stream transmission method and system

Abstract

The invention relates to the technical field of mobile communication, in particular to a data multi-stream transmission method and system. The system comprises a transmitter and a receiver, and the transmitter is used for performing channel coding, modulation, resource mapping and pre-coding processing on original signal data to generate pre-coded multi-stream superposed data. The receiver is used for constructing a multi-dimensional graph according to the received pre-coded multi-stream superposition data and the channel tensor; wherein the multi-dimensional graph comprises a plurality of vertexes and hyperedge feature data corresponding to the vertexes; generating final soft decision data according to the hyperedge feature data in combination with a preset multi-dimensional graph neural network model; and decoding the final soft decision data to obtain original signal data. The multi-dimensional graph neural network model reserves a signal physical structure, the problem of information data loss of a traditional model is solved, the balance between low calculation complexity and high signal reduction precision is finally achieved by optimizing the multi-dimensional graph neural network model, and the block error rate performance and throughput of the system are improved.

Description

Technical Field

[0001] This invention relates to the field of mobile communication technology, and in particular to a method and system for multi-stream data transmission. Background Technology

[0002] In traditional communication systems, due to the limited signal processing capabilities of the receiver, orthogonal or linear transmission methods are typically employed at the transmitter. This increases transmitter complexity in exchange for reduced receiver processing feasibility and complexity. Consequently, data symbols and pilot symbols must be orthogonally allocated in terms of time-frequency resources, leading to competition between pilot and data signals for limited resources. This increases pilot overhead, reduces data transmission spectral efficiency, and limits system throughput. When facing larger antenna arrays and high-speed mobile scenarios, the resource competition between pilot and data signals becomes even more pronounced.

[0003] In recent years, the rapid development of AI technology has brought new design ideas to communication systems. Currently, mainstream AI-based pilot overlay methods include MLP, ResNet, Transformer, and GNN. These methods fully utilize typical structures and technical modules in deep learning, but still have many shortcomings. On the one hand, methods based on MLP, Transformer, and GNN typically require flattening high-dimensional structures such as the received signal tensor, pilot tensor, and channel tensor into a two-dimensional feature input network. This flattening operation destroys the natural structural characteristics of the signal in the spatial, temporal, and frequency domains, weakening the model's ability to utilize physical properties, reducing channel estimation accuracy, and forcing the network to relearn correlations in the mixed high-dimensional features, leading to a significant increase in computational and storage overhead. On the other hand, while CNN-based schemes such as ResNet improve expressive power through deep convolutional structures and residual connections, their computational complexity is high, and the receptive field of the convolutional kernels is limited, making it difficult to capture global dependencies between distant subcarriers or symbols, thus limiting performance in complex scenarios. Summary of the Invention

[0004] This application aims to at least partially address one of the technical problems in the related art. To this end, the first aspect disclosed in this application provides a data multi-stream transmission system, the system comprising: a transmitter and a receiver; The transmitter is used to process the raw signal data through channel coding, modulation, resource mapping and precoding to generate precoded multistream overlay data; The receiver is used to construct a multidimensional graph based on the received precoded multistream superimposed data and channel tensor; wherein, the multidimensional graph includes: multiple vertices and corresponding hyperedge feature data; based on the hyperedge feature data and a preset multidimensional graph neural network model, final soft decision data is generated; the final soft decision data is decoded to obtain the original signal data.

[0005] Optionally, the transmitter is equipped with multiple first antennas, and the receiver is equipped with a number of second antennas corresponding to the number of the transmitter. The number of first antennas of the first antennas and the number of second antennas of the second antennas are used to generate precoded multistream overlay data.

[0006] Optionally, the system is further equipped with multiple subcarriers of continuous orthogonal frequency division multiplexing symbols corresponding to the number of antennas, wherein the number of the first antennas, the number of transmitter data streams, the number of orthogonal frequency division multiplexing symbols, and the number of subcarriers are used to generate a multidimensional graph.

[0007] The second aspect of this application discloses a data multi-stream transmission method, applied to a data multi-stream transmission system, the method comprising: The original signal data and multi-stream pilots are processed by channel coding, modulation, resource mapping and precoding to generate precoded multi-stream superimposed data. A multidimensional graph is constructed based on the received precoded multistream overlay data and channel tensor; wherein the multidimensional graph includes: multiple vertices and hyperedge feature data corresponding to the multiple vertices; The final soft decision data is generated by combining the hyperedge feature data with a pre-set multidimensional graph neural network model. Decode the final soft decision data to obtain the original signal data.

[0008] Optionally, constructing the multidimensional graph based on the received precoded multistream overlay data and channel tensor includes: The physical channel dimension is determined based on the channel tensor and preset rules. A multidimensional graph is constructed based on the physical channel dimension and the precoded multistream overlay data.

[0009] Optionally, the step of generating precoded multistream overlay data by processing the original signal data and multistream pilots through channel coding, modulation, resource mapping, and precoding includes: The original signal data is processed by channel coding to form bitstream data; The bitstream data and multistream pilots are modulated and resource-mapped to generate multistream pilot tensors and multistream signal data tensors. The multi-stream pilot tensor and the bitstream data tensor are superimposed according to a preset power ratio to generate multi-stream superimposed data; Pre-encode the multi-stream overlay data to generate pre-encoded multi-stream overlay data.

[0010] Optionally, the bitstream data and the multistream pilots are modulated and resource-mapped to generate multistream pilot tensors and multistream signal data tensors, including: A multi-stream pilot tensor is generated by combining the number of first antennas, the number of subcarriers, and the number of orthogonal frequency division multiplexing symbols with the multi-stream pilots. A multi-stream signal data tensor is generated by combining the number of first antennas, the number of subcarriers, the number of orthogonal frequency division multiplexing symbols, and the bitstream data.

[0011] Optionally, generating soft-decision data based on hyperedge feature data combined with a multidimensional graph neural network model includes: The hyperedge feature data is combined with a preset multidimensional graph neural network model to generate equivalent channel information data. The final soft-decision data is generated by combining the equivalent channel information data, hyperedge feature data, and multidimensional graph neural network.

[0012] Optionally, the step of combining the hyperedge feature data with a preset multidimensional graph neural network model to generate equivalent channel information data includes: The multi-stream pilot tensor and the precoded multi-stream superimposed data are converted into a four-dimensional pilot tensor and a four-dimensional precoded multi-stream superimposed data. A composite tensor is generated based on the four-dimensional pilot tensor and the four-dimensional precoded multi-stream superposition data; The integrated tensor is combined with the preset multidimensional graph neural network model to generate equivalent channel information data.

[0013] Optionally, the method further includes: Based on the equivalent channel information data and hyperedge feature data, combined with a multidimensional graph neural network, first soft decision data is generated. A loss function is generated by performing binary cross-entropy calculation based on the first soft-decision data and bitstream data. The parameters of the multidimensional graph neural network model are adjusted using this loss function to generate a preset multidimensional graph neural network model; Based on the equivalent channel information data and hyperedge feature data, combined with a preset multidimensional graph neural network, the final soft decision data is generated.

[0014] This application uses precoded multi-stream overlay data generated by the transmitter to overlay multiple data streams, reducing the computational complexity during data transmission. Furthermore, the multidimensional graph constructed in this application correlates the physical channel dimension and hyperedge feature data during data transmission, providing complete feature data for the subsequent generation of final soft-decision data using a pre-defined multidimensional graph neural network model. In this application, the multidimensional graph neural network model preserves the physical structure of the signal, solving the information loss problem of traditional models. By optimizing the multidimensional graph neural network model, a balance between low computational complexity and high signal reconstruction accuracy is ultimately achieved, improving the system's block error rate performance and throughput. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating a multi-stream data transmission method according to this application; Figure 2 This is a schematic flowchart of an embodiment of a multi-stream data transmission method according to this application; Figure 3 This is a schematic flowchart of another embodiment of a multi-stream data transmission method according to this application; Figure 4 This is a schematic flowchart of another embodiment of a multi-stream data transmission method according to this application; Figure 5 This is a schematic flowchart of another embodiment of a multi-stream data transmission method according to this application; Figure 6 This is a schematic flowchart of another embodiment of a multi-stream data transmission method according to this application; Figure 7 This is a schematic flowchart of another embodiment of a multi-stream data transmission method according to this application; Figure 8 This is a schematic diagram of the structure of a multi-stream data transmission system according to this application; Figure 9 This is a block error rate diagram of a multi-stream data transmission method according to this application; Figure 10 This is a graph showing the computational complexity and number of parameters of a multi-stream data transmission method proposed in this application. Detailed Implementation

[0016] To better understand this application, various aspects of this application will be described in more detail with reference to the accompanying drawings. It should be understood that these detailed descriptions are merely illustrative of exemplary embodiments of this application and are not intended to limit the scope of this application in any way. Throughout the specification, the same reference numerals refer to the same elements. The expression "and / or" includes any and all combinations of one or more of the associated listed items.

[0017] It should be noted that, where there is no conflict, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0018] It should be noted that in this application, OFDM stands for Orthogonal Frequency Division Multiplexing Symbol; CDL stands for Clustered Delay Line; LDPC stands for Low Density Parity Check Code; DMRS stands for Demodulation Reference Signal; QAM stands for Quadrature Amplitude Modulation; BP algorithm stands for Back Propagation; GNN stands for Multidimensional Graph Neural Network; 5G stands for Fifth Generation; 6G stands for Sixth Generation; AI stands for Artificial Intelligence; MLP stands for Multilayer Perceptron; and ResNet stands for Residual Network.

[0019] It should be noted that, in this application, the spatial domain specifically refers to the physical dimension related to spatial location within the system, and in this application, the physical channel dimension corresponding to the spatial domain is the transmitter antenna. The time domain refers to the physical dimension related to time segment division within the system, and in this application, the corresponding physical channel dimension is the OFDM symbol. The frequency domain refers to the physical dimension related to frequency resource allocation within the system, and in this application, the corresponding physical channel dimension is the subcarrier.

[0020] Traditional approaches typically flatten high-dimensional structures such as the received signal tensor and pilot tensor into two-dimensional features, which disrupts the correlation of signals in the spatial, temporal, and frequency domains (i.e., subcarriers). This weakens the model's ability to utilize the physical channel's dimensional characteristics. Furthermore, the network needs to relearn the correlations, increasing computational overhead. This application directly preserves the physical structure of the signal by connecting related vertices with multidimensional graph hyperedges, enabling subsequent multidimensional graph neural networks to directly utilize the system's physical characteristics. This improves task accuracy while reducing computational complexity.

[0021] This application will now be described in conjunction with the accompanying drawings.

[0022] like Figure 1 As shown, a data multi-stream transmission method is provided in the first aspect of this application, applied to the data multi-stream transmission system, the method comprising: S101 processes the original signal data through channel coding, modulation, resource mapping, and precoding to generate precoded multi-stream overlay data.

[0023] S102, construct a multidimensional graph based on the received precoded multistream overlay data and channel tensor; wherein, the multidimensional graph includes: multiple vertices and hyperedge feature data corresponding to the multiple vertices.

[0024] S103, based on the hyperedge feature data and the preset multidimensional graph neural network model, generates the final soft decision data.

[0025] S104, decode the final soft decision data to obtain the original signal data.

[0026] It should be noted that the data processing flow of the transmitter in this application is as follows: after the original signal data is channel-coded, bit stream data is generated; the bit stream data and multi-stream pilot data are modulated and resource-mapped to generate multi-stream data tensors and multi-stream pilot tensors; the multi-stream data tensors and multi-stream pilot tensors are superimposed to generate multi-stream superimposed data; the multi-stream superimposed data is pre-coded to finally generate pre-coded multi-stream superimposed data.

[0027] This application first pre-encodes the superimposed signal generated from multi-stream pilot and multi-stream signal data in the transmitter. This step not only integrates the data but also simplifies the raw data processed by the receiver, reducing the computational complexity during data transmission. Furthermore, the multidimensional graph constructed in this application associates the physical channel dimension with the hyperedge feature data during data transmission, providing more complete feature data for generating the final soft-decision data using a pre-defined multidimensional graph neural network model. In this application, the multidimensional graph neural network model preserves the physical structure of the signal, solving the information data loss problem of traditional models. By optimizing the multidimensional graph neural network model, a balance between low computational complexity and high signal reconstruction accuracy is ultimately achieved, improving the system's block error rate performance and throughput.

[0028] In one embodiment that can be implemented in this application, such as Figure 2 As shown, the construction of the multidimensional graph based on the received precoded multistream overlay data and channel tensor includes: S201, determine the physical channel dimension based on the channel tensor and preset rules.

[0029] S202, construct a multidimensional graph based on the physical channel dimension and precoded multistream overlay data.

[0030] It should be noted that the physical channel dimension selection described in this application follows a preset rule that the selected physical channel dimensions are correlated and do not overlap. The obtained physical channel dimensions are used as vertices to construct a multidimensional graph, where the vertices have no features. The information received by the receiver corresponds to the hyperedge features of the multidimensional graph. The vertices are interconnected to form hyperedges. The combination of vertices corresponds to the transmission scenario of multi-stream superimposed signals, and the hyperedge features map the key transmission data in the corresponding scenario.

[0031] For example, suppose the multidimensional graph has four vertices: vertex 1 is the second antenna, vertex 2 is the spatial stream number, vertex 3 is the subcarrier, and vertex 4 is the OFDM symbol. Vertex 1 and vertex 2 are connected to form hyperedge 1, representing that the spatial stream signal is transmitted through the wireless channel and received by the second antenna. Vertex 1 and vertex 3 form hyperedge 2, representing that the subcarrier-carried signal is captured by the second antenna after passing through the channel. Vertex 1 and vertex 4 form hyperedge 3, representing that the second antenna collects signals from all subcarriers within the OFDM symbol period. Vertex 2 and vertex 3 form hyperedge 4, representing that the spatial stream modulation symbol is mapped onto the subcarrier for transmission. Vertex 2 and vertex 4 form hyperedge 5, representing that the spatial stream signal is transmitted within the OFDM symbol period. Vertex 3 and vertex 4 form hyperedge 6, representing that the subcarrier and OFDM symbol constitute a time-frequency resource grid, which is the basic resource unit of OFDM.

[0032] For example, in a multi-stream data transmission system, the transmitter is equipped with The receiver is equipped with the first antenna. The second antenna. The number of data streams transmitted by the transmitter is... To allocate multi-stream transmission systems with A series of OFDM symbols Subcarriers. Definition The equivalent channel after precoding. For encoded bitstream data, For multi-stream pilot tensors, the receiver channel estimation task is viewed as constructing a mapping. The data detection task is viewed as constructing a mapping. ,in Let these represent trainable parameters, assuming they are the receiver antenna, transmitter stream, OFDM symbol, and subcarrier. The receiver constructs a multidimensional graph using these four physical dimensions as vertices, where each vertex lacks features, and the relationships between vertices are described by hyperedges, which carry the hyperedge features. Specifically, this is the multi-stream pilot tensor. It is the feature corresponding to the hyperedge connecting the three types of vertices of the transmitter stream, OFDM symbol, and subcarrier, and the received signal. It is the feature corresponding to the hyperedge connecting the three types of vertices of the receiver antenna, OFDM symbol, and subcarrier, and is the equivalent channel. It is the feature corresponding to the superedge that connects the four types of vertices: receiver antenna, transmitter stream, OFDM symbol, and subcarrier.

[0033] The multidimensional graph constructed collaboratively by steps S201 and S202 avoids the information loss and high complexity of traditional schemes, and provides feature inputs for subsequent multidimensional graph neural networks to achieve high-precision channel estimation and efficient data detection.

[0034] In one embodiment that can be implemented in this application, such as Figure 3 As shown, the process of generating precoded multistream overlay data by performing channel coding, modulation, resource mapping, and precoding on the original signal data includes: S301, the original signal data is processed by channel coding to form bit stream data; S302, the bitstream data and the multi-stream pilots are modulated and resource-mapped to generate a multi-stream pilot tensor and a multi-stream signal data tensor.

[0035] S303 superimposes the multi-stream pilot tensor and the multi-stream signal data tensor according to a preset power ratio to generate multi-stream superimposed data.

[0036] S304 precodes the multi-stream overlay data to generate precoded multi-stream overlay data.

[0037] For example, suppose in a multi-stream data transmission system, the transmitter's multi-stream pilots are... Multi-stream signal data is The preset power ratio is The expression for the multi-stream overlay data is: .

[0038] Typically, multi-stream pilots serve as reference information. The ultimate goal of data transmission is to send the original signal data to the receiver. If the power ratio of the multi-stream pilots is high, it will cause significant interference to the subsequent reconstruction of the original information data. Therefore, this application avoids interference between the multi-stream pilots and multi-stream signal data due to power imbalance by presetting a power ratio. Multi-stream superimposed data is generated according to this ratio, and then pre-encoded. This method is suitable for multi-antenna transmission scenarios, ultimately generating transmittable pre-coded multi-stream superimposed data.

[0039] In one embodiment that can be implemented in this application, such as Figure 4 As shown, the step of superimposing the multi-stream pilot tensor and the multi-stream signal data tensor according to a preset power ratio to generate multi-stream superimposed data includes: S401, based on the number of the first antennas, the number of subcarriers, and the number of orthogonal frequency division multiplexing symbols, and combined with the multi-stream pilots, a multi-stream pilot tensor is generated.

[0040] S402, generate a multi-stream signal data tensor based on the number of first antennas, the number of subcarriers, the number of orthogonal frequency division multiplexing symbols, and bitstream data.

[0041] For example, suppose in a multi-stream data transmission system, the transmitter is equipped with N t The receiver is also equipped with N antennas. t Based on the second antenna, the number of data streams from the transmitter is L, and it is configured to allocate subcarriers with T consecutive OFDM symbols for a multi-stream data transmission system, defined as follows: For multi-stream pilot tensors, For multi-stream signal data tensors, This is multi-stream overlay data, where, To preset the power ratio, precoding is performed on the multi-stream superimposed data to determine the received signal of the receiver. Specifically, it is expressed as follows: ,in, For the receiver antenna, (positive integer index) A positive integer index representing the number of multi-stream transport streams. It is the first root antenna and the first The equivalent channel corresponding to the layer, It is the first Layer emission superposition symbols and It is the first The received signal from the antenna, To receive additive white complex Gaussian noise, the variance of each element is defined as follows: .

[0042] The multi-stream signal data tensor generated in this step directly preserves the multi-dimensional structure. When the receiver constructs a multi-dimensional graph later, it can directly map the tensor to the hyperedge features connecting the transmitter stream, OFDM symbols, and subcarriers without having to reshape the physical channel dimension, thus avoiding information loss and providing input features for the receiver's multi-dimensional graph neural network processing.

[0043] In one embodiment that can be implemented in this application, such as Figure 5 As shown, the process of generating soft-decision data based on hyperedge feature data combined with a multidimensional graph neural network model includes: S501, combine the hyperedge feature data with a preset multidimensional graph neural network model to generate equivalent channel information data.

[0044] S502, based on the equivalent channel information data, hyperedge feature data, and multidimensional graph neural network, the final soft decision data is generated.

[0045] For example, for the first... The layer defines five linear transformation matrices. , , ,in The index of the defined linear transformation matrix. The hyperedge feature update equation in the application of a multidimensional graph neural network is expressed as: (This is the layer index of the multidimensional graph neural network.) ; ; ; ; ; .

[0046] in, The product of the physical channel dimensions of the tensor. Is with layer The first in Learnable weights of the transformation matrix, Aggregation along a specified physical channel dimension typically involves taking the maximum, minimum, summation, and mean. This application uses a mean function, preserving the permutation invariance of the input features. Based on the above formula, the multidimensional GNN updates the hyperedge representation through message passing and aggregation, achieving parameter sharing and cross-dimensional information fusion without disrupting the original spatiotemporal frequency structure. This design reduces the number of trainable parameters while improving the performance of channel estimation.

[0047] Specifically, this application takes a GNN (Graph Neural Network) as an example. Assuming the example neural network has a three-layer structure, the hyperedge features are updated in each layer. Assuming the feature changes are calculated using the formula described above in the first layer... Time dimension aggregation Subcarrier dimension aggregation Stream dimension aggregation Antenna dimension aggregation The final fusion result of the first layer is: On the second layer of the neural network structure Continue repeating the above hyperedge feature update steps to finally output the fusion result of the second layer. On the third layer of the neural network structure The feature transformation is performed directly, and the fusion result of the third layer is obtained based on the feature transformation result. Equivalent channel information data is obtained based on the fusion result.

[0048] The hyperedge feature data of this application directly preserves the multidimensional structure of receiver antenna, transmitter stream, OFDM symbols and subcarriers. This step takes a multidimensional graph neural network as the core and achieves high-precision channel estimation through multi-source hyperedge feature fusion. Then, based on the accurate channel estimation information, multi-stream interference is specifically canceled, and highly reliable soft decision data is output. This not only solves the problems of difficult multi-stream interference cancellation, poor robustness of channel estimation and high complexity in traditional schemes, but also forms a deep fit with the transmitter-end superimposed pilot scheme.

[0049] In one embodiment that can be implemented in this application, such as Figure 6 As shown, the step of combining the hyperedge feature data with a preset multidimensional graph neural network model to generate equivalent channel information data includes: S601, convert the multi-stream pilot tensor and the precoded multi-stream superimposed data into a four-dimensional pilot tensor and a four-dimensional precoded multi-stream superimposed data.

[0050] S602, Generate a composite tensor based on the four-dimensional pilot tensor and the four-dimensional precoded multi-stream superposition data.

[0051] S603, combine the comprehensive tensor with the preset multidimensional graph neural network model to generate equivalent channel information data.

[0052] For example, the receiver receives signals according to the system. and known pilot Two complex three-dimensional tensors and Mapping to the hyperedge features of the constructed multidimensional graph, specifically, mapping two complex-valued three-dimensional tensors. and The real and imaginary parts are split into four real-valued three-dimensional tensors. , , and And by adding a dimension to these four real-valued three-dimensional tensors, they become four real-valued four-dimensional tensors. , , and Then, the two real-valued four-dimensional tensors corresponding to the received signal are... and By piecing them together along the first dimension, we get The two real-valued four-dimensional tensors corresponding to the pilot frequency and By piecing them together along the first dimension, we get In order to perform dimension alignment, in the tensor corresponding to the received signal Insert a new dimension between the second and third dimensions, and it becomes Then perform on the extended dimension The copy operation yielded Similarly, in the tensor corresponding to the multi-channel pilot... Insert a new dimension between the first and second dimensions, and it becomes Then perform on the extended dimension The copy operation yielded Finally, and By concatenating along the first dimension, we obtain the composite tensor. Based on the permutation invariance of the channel estimation problem, the receiver inputs the synthesis tensor into a multidimensional graph neural network. By updating the hyperedge features in each layer of the multidimensional graph neural network, the channel estimate is output. The channel estimation result is the equivalent channel information data.

[0053] This step integrates key information from multiple sources by calculating a comprehensive tensor. Then, a multidimensional graph neural network is used to mine the correlation of physical dimensions to achieve equivalent channel estimation with low complexity, high accuracy, and strong scene adaptability. This provides a reliable channel foundation for subsequent multi-stream data detection and is a key preprocessing and estimation process that supports the system's high spectral efficiency and low block error rate.

[0054] In one embodiment that can be implemented in this application, such as Figure 7 As shown, the method further includes: S701, Based on the equivalent channel information data, the hyperedge feature data, and the multidimensional graph neural network, the first soft decision data is generated.

[0055] S702, perform binary cross-entropy calculation based on the first soft decision data and the bitstream data to generate a loss function.

[0056] S703 uses this loss function to adjust the parameters of the multidimensional graph neural network model and generates a preset multidimensional graph neural network model.

[0057] For example, the receiver receives signals according to the system. Known pilot frequency and channel estimation Following a similar approach to the channel estimation part, a multidimensional real-input synthesis tensor is constructed. Based on the permutation invariance of the data detection problem, the comprehensive tensor is input into a multidimensional graph neural network to obtain the final soft-decision result. Based on the final soft decision result, the corresponding decoding is performed and the restored original signal data is output. ,in Indicates bitrate.

[0058] When training a multidimensional graph neural network, the loss function is calculated as the mean square error between the true equivalent channel and the estimated channel, defined as follows: A binary cross-entropy operation is performed on the first soft-decision data and the bitstream data to generate a loss function for the data verification process. Specifically, the first soft-decision data is... and the encoded bitstream Binary cross-entropy operation between them.

[0059] It should be noted that the receiver's decoding must match the transmitter's encoding rules, and an LDPC decoder should be used.

[0060] For example, this application uses LDPC as the channel coding scheme. The decoding process is based on the LDPC code's verification rules, iteratively correcting the soft decision result, and finally outputting the original signal data. The specific logic is as follows: Before generating the multi-stream signal data tensor, the transmitter encodes the original signal data using an LDPC encoder to generate an encoded bitstream b, which is then mapped using QAM and OFDM resources to form a multi-stream superimposed data tensor. Subsequently, an iterative decoding process is performed. Specifically, after receiving the soft decision result, the LDPC decoder, based on the sparse parity-check matrix of the LDPC code, continuously corrects each signal data point through iterative message passing between variable nodes and verification nodes, for example, through a belief propagation algorithm or a backpropagation (BP) algorithm. The correction process specifically involves using the soft decision result as the initial confidence level, calculating the verification message from each verification node to the variable node, adjusting the variable node's confidence level according to the verification message, and then feeding it back to the verification node. This process is repeated until the iteration threshold is met or all verification equations are satisfied. After the iteration ends, the decoder outputs a hard decision result (0 or 1), i.e., the original signal data, based on the final bit confidence level, thus completing the decoding.

[0061] like Figure 8 As shown, a second aspect of this application provides a multi-stream data transmission system, the system comprising: a transmitter and a receiver.

[0062] The transmitter is used to process the raw signal data through channel coding, modulation, resource mapping and precoding to generate precoded multistream overlay data.

[0063] The receiver is used to construct a multidimensional graph based on the received precoded multistream superimposed data and the channel tensor; wherein, the multidimensional graph includes: multiple vertices and corresponding hyperedge feature data; the hyperedge feature data is combined with a multidimensional graph neural network to generate equivalent channel information data; soft decision data is generated based on the equivalent channel information data, the hyperedge feature data, and the multidimensional graph neural network; the soft decision data is decoded to obtain the original data.

[0064] The transmitter of this application first mixes the multi-stream pilot and multi-stream signal data according to a preset power ratio to generate a mixed signal. This step can avoid interference between the multi-stream pilot and multi-stream data due to power imbalance. Then, the mixed signal is pre-coded to generate pre-coded multi-stream superimposed data. The pre-coding operation is more adaptable to various antenna transmission scenarios.

[0065] The receiver of this application receives precoded multistream overlay data that integrates multistream pilots, multistream data, and channel information. Based on the above data information, the receiver uses a method of constructing a multidimensional graph to associate the abstract physical dimensions with the signal data features, providing a feature carrier for the subsequent generation of equivalent channel information data.

[0066] The multi-stream data transmission system of this application, through flexible configuration of the physical channel dimension, can support both small-scale multi-stream transmission and adapt to 6G massive MIMO antenna arrays without reconstructing the core framework, demonstrating strong adaptability. In addition, this system is compatible with OFDM technology and conforms to the 5G / 6G standard architecture. The system uses OFDM as the physical layer transmission basis and is compatible with the OFDM architecture of the existing 5G NR standard. At the same time, through the superposition of pilot signals and AI model optimization, it meets the requirements of 6G technology for high spectrum efficiency, low latency, and high reliability, and can smoothly transition communication technology upgrades.

[0067] In one embodiment of this application, the transmitter is equipped with multiple first antennas, and the receiver is equipped with a number of second antennas corresponding to the number of the transmitter. The number of first antennas of the first antennas and the number of second antennas of the second antennas are used to generate precoded multistream superimposed data.

[0068] In one embodiment of this application, the system is further equipped with multiple subcarriers of continuous orthogonal frequency division multiplexing symbols corresponding to the number of antennas, wherein the number of first antennas, the number of transmitter data streams, the number of orthogonal frequency division multiplexing symbols, and the number of subcarriers are used to generate a multidimensional graph.

[0069] For example, the scenario is set as a multi-stream data transmission system, which is equipped with a transmitter and a receiver, wherein the transmitter is equipped with... One antenna, one receiver equipped with One antenna. The transmitter transmits a data stream of [number]. Multi-stream data transmission system allocation has A series of OFDM symbols There are 1 subcarrier. The channel adopts the CDL channel model in 3GPP TR 38.901. The CDL channel model is widely used in link-level evaluation. Assuming the carrier frequency is 1... The carrier spacing is The number of subcarriers is OFDM symbol count is The channel coding scheme is LDPC, and the code rate is... =0.5, delay spread to In the simulation experiment, the Monte Carlo method was used to randomly generate 1000 independent CDL channels for simulation. In each simulation, the original bits were randomly generated.

[0070] Figure 9 This is a comparison chart of experimental data performance of a multi-stream data transmission method provided in an embodiment of the present invention. Specifically, the experimental performance is evaluated using the block error rate. Figure 9 The horizontal axis represents the transmitter's transmit channel ratio, which is defined as follows: The vertical axis represents the block error rate, where the LMMSE scheme is a method for LMMSE channel estimation and data detection under an orthogonal pilot DMRS in the 5G NR standard system. The covariance matrix of the LMMSE channel estimation is... The calculation is performed on channel samples. The Transformer receiver is a method for MLP channel estimation and Transformer data detection in the superimposed pilot scenario. This application has a significant improvement in block error rate performance compared to LMMSE and Transformer receiver. This performance improvement stems from the fact that the multidimensional GNN fully utilizes the multidimensional correlation of the received signal in the spatiotemporal frequency domain while retaining key structural information. In contrast, the Transformer partially compresses or reconstructs the structure, resulting in information loss and reduced estimation accuracy. It should be noted that, unlike LMMSE, which allocates additional OFDM symbols to DMRS pilots, the multidimensional GNN involved does not require additional pilot symbols, thereby achieving excellent performance while significantly improving throughput.

[0071] Figure 10 This diagram illustrates the computational complexity and parameter count of a multi-stream data transmission method provided in an embodiment of the present invention. For example, Figure 10 The three receiver schemes were compared separately for channel estimation and data detection tasks. In comparison, the computational complexity and number of model parameters of the proposed method are significantly reduced compared to other receiver schemes. This advantage stems from two key factors. First, the multidimensional GNN makes full use of the multidimensional spatial structure of the input features, achieving excellent performance while maintaining low complexity. Second, by utilizing the permutations inherent in channel estimation and data detection, cross-dimensional parameter sharing is allowed, further reducing the number of trainable parameters.

[0072] In summary, the algorithm provided in this application can achieve accurate channel estimation and efficient data detection while reducing computational complexity, thereby improving the system's block error rate performance and throughput.

[0073] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A data multi-stream transmission system, characterized by, The system comprises a transmitter and a receiver; The transmitter is configured to generate precoded multi-stream superposition data by processing original signal data through channel coding, modulation, resource mapping and precoding; The receiver is configured to construct a multi-dimensional graph according to the received precoded multi-stream superposition data and channel tensor; wherein the multi-dimensional graph comprises a plurality of vertices and hyperedge feature data corresponding to the plurality of vertices; generate final soft decision data according to the hyperedge feature data combined with a preset multi-dimensional graph neural network model; and decode the final soft decision data to obtain the original signal data.

2. A data multi-stream transmission system as claimed in claim 1, characterized in that, The transmitter is equipped with a plurality of first antennas, and the receiver is equipped with a corresponding number of second antennas corresponding to the transmitter; the first antenna quantity of the first antennas and the second antenna quantity of the second antennas are used to generate the precoded multi-stream superposition data.

3. A data multi-stream transmission system as claimed in claim 2, characterized in that, The system is also equipped with a plurality of subcarriers of continuous orthogonal frequency division multiplexing symbols corresponding to the number of antennas; the first antenna quantity, the transmitter data stream number, the orthogonal frequency division multiplexing symbol quantity and the subcarrier quantity are used to generate the multi-dimensional graph.

4. A data multi-stream transmission method applied to the data multi-stream transmission system, characterized in that, The method comprises: generating precoded multi-stream superposition data by processing original signal data through channel coding, modulation, resource mapping and precoding; constructing a multi-dimensional graph according to the received precoded multi-stream superposition data and channel tensor; wherein the multi-dimensional graph comprises a plurality of vertices and hyperedge feature data corresponding to the plurality of vertices; generating final soft decision data according to the hyperedge feature data combined with a preset multi-dimensional graph neural network model; decoding the final soft decision data to obtain the original signal data.

5. A method of data multi-stream transmission as claimed in claim 4, characterized in that, The method of constructing a multi-dimensional graph according to the received precoded multi-stream superposition data and channel tensor comprises: determining the physical channel dimension according to the channel tensor combined with a preset rule; constructing a multi-dimensional graph according to the physical channel dimension and the precoded multi-stream superposition data.

6. The method of claim 4, wherein, The method of generating precoded multi-stream superposition data by processing original signal data through channel coding, modulation, resource mapping and precoding comprises: processing the original signal data through channel coding to form bit stream data; processing the bit stream data and multi-stream pilot through modulation and resource mapping to generate a multi-stream signal data tensor and a multi-stream pilot tensor; superimposing the multi-stream pilot tensor and the multi-stream signal data tensor according to a preset power ratio to generate multi-stream superposition data; precoding the multi-stream superposition data to generate precoded multi-stream superposition data.

7. A method of data multi-stream transmission as claimed in claim 6, characterized in that, The method of superimposing the multi-stream pilot and the multi-stream signal data according to a preset power ratio to generate multi-stream superposition data comprises: generating a multi-stream pilot tensor according to the first antenna quantity, the subcarrier quantity and the orthogonal frequency division multiplexing symbol quantity combined with the multi-stream pilot; generating a multi-stream signal data tensor according to the first antenna quantity, the subcarrier quantity, the orthogonal frequency division multiplexing symbol quantity combined with the bit stream data.

8. The method of claim 4, wherein, The method of generating soft decision data according to the hyperedge feature data combined with a multi-dimensional graph neural network model comprises: generating equivalent channel information data by combining the hyperedge feature data with a preset multi-dimensional graph neural network model; generating final soft decision data according to the equivalent channel information data, the hyperedge feature data combined with a multi-dimensional graph neural network.

9. A method of data multi-stream transmission as claimed in claim 8, characterized in that, The combining the hyper-edge feature data with the preset multi-dimensional graph neural network model to generate equivalent channel information data comprises: Converting the multi-stream pilot tensor and the precoded multi-stream superposition data into a four-dimensional pilot tensor and a four-dimensional precoded multi-stream superposition data; Generating a comprehensive tensor according to the four-dimensional pilot tensor and the four-dimensional precoded multi-stream superposition data; Combining the comprehensive tensor with the preset multi-dimensional graph neural network model to generate equivalent channel information data.

10. A method of data multi-stream transmission as claimed in any of claims 4 to 9, characterized in that, The hyper-edge feature data combines the preset multi-dimensional graph neural network model to generate equivalent channel information data, and the original signal data is processed by channel coding to form bit stream data, and the method further comprises: Generating first soft decision data according to the equivalent channel information data, the hyper-edge feature data, and the multi-dimensional graph neural network; Performing binary cross-entropy operation on the first soft decision data and the bit stream data to generate a loss function; Adjusting parameters of the multi-dimensional graph neural network model using the loss function to generate a preset multi-dimensional graph neural network model.

Images