Communication optimization method, electronic device, and readable storage medium

Abstract

The application discloses a communication optimization method, an electronic device and a readable storage medium, and applies to the marketing technical field, and comprises the following steps: performing semantic coding on communication data to obtain semantic coding features; performing channel coding on the semantic coding features to obtain channel coding features; broadcasting the channel coding features to a feature subspace corresponding to the communication data, so that a receiving end pulls the channel coding features from the feature subspace, and decodes and restores the channel coding features to obtain semantic decoding features. The application solves the technical problem of high delay in multi-terminal communication.

Description

Technical Field

[0001] This application relates to the field of marketing technology, and in particular to a communication optimization method, electronic device, and readable storage medium. Background Technology

[0002] With the rapid development of science and technology, communication technology has become increasingly mature. Currently, the transmitting end and the receiving end usually communicate one-to-one through a single communication channel. When there are many transmitting ends or receiving ends that need to communicate, network congestion is likely to occur, resulting in high latency for multi-end communication. Summary of the Invention

[0003] The main objective of this application is to provide a communication optimization method, electronic device, and readable storage medium, which aims to solve the technical problem of high latency in multi-terminal communication in the prior art.

[0004] To achieve the above objectives, this application provides a communication optimization method applied at a transmitting end, the communication optimization method comprising: Semantic encoding is performed on the communication data to obtain semantic encoding features; Channel coding is performed on the semantic coding features to obtain channel coding features; The channel coding features are broadcast to the feature subspace corresponding to the communication data, so that the receiving end can retrieve the channel coding features from the feature subspace and decode and restore the channel coding features to obtain semantic decoding features.

[0005] To achieve the above objectives, this application also provides a communication optimization method applied at a receiving end, the communication optimization method comprising: Channel coding features are extracted from the feature subspace, wherein the channel coding features are obtained by channel coding of semantic coding features by the transmitter, the semantic coding features are obtained by semantic coding of communication data by the transmitter, and the channel coding features are transmitted via the light waves in the feature subspace corresponding to the communication data; The channel coding features are decoded and restored to obtain semantic decoding features.

[0006] To achieve the above objectives, this application also provides a transmitter, the transmitter comprising: The semantic encoding module is used to perform semantic encoding on communication data to obtain semantic encoded features; A channel coding module is used to perform channel coding on the semantic coding features to obtain channel coding features; The broadcast module is used to broadcast the channel coding features to the feature subspace corresponding to the communication data, so that the receiving end can pull the channel coding features from the feature subspace and decode the channel coding features to obtain semantic decoding features.

[0007] To achieve the above objectives, this application also provides a receiving end, the receiving end comprising: The pull module is used to pull channel coding features from the feature subspace, wherein the channel coding features are obtained by channel coding of semantic coding features by the transmitter, the semantic coding features are obtained by semantic coding of communication data by the transmitter, and the channel coding features are transmitted via the light waves in the feature subspace corresponding to the communication data; The decoding module is used to decode and restore the channel coding features to obtain semantic decoding features.

[0008] This application also provides an electronic device, the electronic device comprising: a memory, a processor, and a program of the communication optimization method stored in the memory and executable on the processor, wherein when the program of the communication optimization method is executed by the processor, it can implement the steps of the communication optimization method as described above.

[0009] This application also provides a computer-readable storage medium storing a program implementing a communication optimization method, wherein when the program is executed by a processor, it implements the steps of the communication optimization method as described above.

[0010] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the communication optimization method described above.

[0011] This application provides a communication optimization method, electronic device, and computer-readable storage medium applied to a transmitting end. The method involves semantically encoding communication data to obtain semantically encoded features; channel encoding the semantically encoded features to obtain channel encoded features; and broadcasting the channel encoded features to a feature subspace corresponding to the communication data, allowing the receiving end to retrieve the channel encoded features from the feature subspace. The channel encoded features are then decoded to obtain semantically decoded features. Each transmitting end can broadcast its channel encoded features through the feature subspace corresponding to its transmitted communication data, ensuring orthogonality of multi-end communication transmission in the feature domain. Therefore, multiple transmitting ends can simultaneously transmit their respective communication data's corresponding channel encoded features in the time domain without interfering with each other, thus reducing multi-end communication latency.

[0012] This application provides a communication optimization method, electronic device, and computer-readable storage medium applied at a receiving end. It extracts channel coding features from a feature subspace, wherein the channel coding features are obtained by channel coding semantic coding features at the transmitting end, and the semantic coding features are obtained by semantic coding communication data at the transmitting end. The channel coding features are then decoded and restored via the light waves in the feature subspace corresponding to the communication data to obtain semantic decoding features. Each receiving end can extract channel coding features from its respective feature subspace to decode and restore the channel coding features, thereby achieving communication between the receiving end and the transmitting end. Furthermore, using the feature subspace as the communication medium ensures the orthogonality of multi-end communication reception in the feature domain. Therefore, multiple receiving ends can simultaneously extract channel coding features from each feature subspace in the time domain without interfering with each other, thus reducing multi-end communication latency. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the hardware operating environment involved in the embodiments of the present invention; Figure 2 This is a flowchart illustrating the first embodiment of the communication optimization method of the present invention; Figure 3 A schematic diagram illustrating a process for semantic encoding of the semantic encoding model involved in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating a channel coding process for the channel coding model involved in an embodiment of the present invention. Figure 5 This is a flowchart illustrating the second embodiment of the communication optimization method of the present invention; Figure 6 This is a schematic diagram of a decoding and restoration process for the channel decoding model involved in an embodiment of the present invention; Figure 7 This is a schematic diagram of a structure for joint training of the semantic encoding model and the semantic decoding model to be trained, as described in an embodiment of the present invention. Figure 8 This refers to the accuracy of communication transmission in a training scenario with a 0dB threshold when using the communication optimization method described in this embodiment of the invention. Figure 9 This refers to the accuracy of communication transmission in a training scenario with a 5dB threshold when using the communication optimization method described in this embodiment of the invention. Figure 10 This refers to the accuracy of communication transmission in a training scenario with a 10dB threshold when using the communication optimization method described in this embodiment of the invention. Figure 11 This is a schematic diagram of the functional modules of a preferred embodiment of the transmitter of the present invention; Figure 12This is a schematic diagram of the functional modules of a preferred embodiment of the receiver of the present invention.

[0014] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0015] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0016] like Figure 1 As shown, Figure 1 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of the present invention.

[0017] It should be noted that the communication optimization device in this embodiment of the invention can be a smartphone, a personal computer, a server, or other devices, and no specific limitations are imposed here.

[0018] like Figure 1 As shown, the communication optimization device may include: a processor 1001, such as a CPU; a network interface 1004; a user interface 1003; a memory 1005; and a communication bus 1002. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be high-speed RAM or non-volatile memory, such as a disk drive. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0019] Those skilled in the art will understand that Figure 1 The device structure shown does not constitute a limitation on the communication optimization device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0020] like Figure 1 As shown, the memory 1005, as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a communication optimization program. The operating system is a program that manages and controls the device's hardware and software resources, supporting the operation of the communication optimization program and other software or programs. Figure 1In the device shown, the user interface 1003 is mainly used for data communication with the client; the network interface 1004 is mainly used for establishing a communication connection with the server; and the processor 1001 can be used to call the communication optimization program stored in the memory 1005 and perform the following operations: Semantic encoding is performed on the communication data to obtain semantic encoding features; Channel coding is performed on the semantic coding features to obtain channel coding features; The channel coding features are broadcast to the feature subspace corresponding to the communication data, so that the receiving end can retrieve the channel coding features from the feature subspace and decode and restore the channel coding features to obtain semantic decoding features.

[0021] Furthermore, the operation of semantically encoding the communication data to obtain semantically encoded features includes: Obtain a semantic coding model, wherein the semantic coding model is trained by multiple first training samples and the semantic coding model to be trained, and each first training sample consists of a first input feature data and a first real label corresponding to the first input feature data, wherein the first input feature data is training data and the first real label is the real semantic coding feature corresponding to the training data. The semantic coding model is used to semantically encode the communication data to obtain the semantic coding features.

[0022] Further, the operation of channel coding the semantic coding features to obtain channel coding features includes: A channel coding model is obtained, wherein the channel coding model is trained by multiple second training samples and a channel coding model to be trained. Each second training sample consists of a second input feature data and a second real label corresponding to the second input feature data. The second input feature data is a training semantic coding feature, and the second real label is a real channel coding feature corresponding to the training semantic coding feature. The channel coding model to be trained is jointly trained with the decoding and restoration model to be trained. The semantic coding features are channel-coded using the channel coding model to obtain the channel coding features.

[0023] Furthermore, after performing semantic encoding on the communication data to obtain semantic encoded features, the processor 1001 can also call the communication optimization program stored in the memory 1005 to perform the following operations: Obtain a preset knowledge base shared by the transmitter and the receiver; Based on the preset knowledge base, the semantic encoding features are adjusted semantically.

[0024] The processor 1001 can be used to call the communication optimization program stored in the memory 1005 and perform the following operations: Channel coding features are extracted from the feature subspace, wherein the channel coding features are obtained by channel coding of semantic coding features by the transmitter, the semantic coding features are obtained by semantic coding of communication data by the transmitter, and the channel coding features are broadcast through the feature subspace corresponding to the communication data; The channel coding features are decoded and restored to obtain semantic decoding features.

[0025] Furthermore, the operation of decoding and restoring the channel coding features to obtain semantic decoding features includes: A decoding and restoration model is obtained, wherein the decoding and restoration model is trained by multiple third training samples and the decoding and restoration model to be trained. Each third training sample consists of a third input feature data and a third real label corresponding to the third input feature data. The third input feature data is a training channel coding feature, and the third real label is a real semantic coding feature corresponding to the training channel coding feature. The decoding and restoration model to be trained is jointly trained with the channel coding model to be trained. The channel coding features are decoded and restored using the decoding and restoration model to obtain semantic decoding features.

[0026] Furthermore, the decoding and restoration model is obtained by iteratively optimizing the decoding and restoration model to be trained based on the residual corresponding to the channel coding model to be trained.

[0027] Furthermore, after decoding and restoring the channel coding features to obtain semantic decoding features, the processor 1001 can also call the communication optimization program stored in the memory 1005 to perform the following operations: Obtain a preset knowledge base shared by the transmitter and the receiver; Based on the preset knowledge base, the semantic decoding features are adjusted semantically.

[0028] Based on the above structure, various embodiments of the communication optimization method are proposed.

[0029] Reference Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of the communication optimization method of the present invention.

[0030] This invention provides an embodiment of a communication optimization method. It should be noted that although the flowchart shows a logical order, in some cases, the steps shown or described may be executed in a different order. In this embodiment, the executing entity of the communication optimization method can be a personal computer, smartphone, server, or other device; no limitation is made in this embodiment. For ease of description, the execution entity is omitted from the following description of each embodiment. In this embodiment, applied to the transmitting end, the communication optimization method includes: Step S10: Semantically encode the communication data to obtain semantically encoded features; In this embodiment, it should be noted that the communication data may include voice data corresponding to a single user or voice data corresponding to multiple users.

[0031] Optionally, in one feasible embodiment, a semantic coding model is obtained, communication data is input into the semantic coding model, and the communication data is mapped into semantic coding features through the semantic coding model.

[0032] Alternatively, in another feasible embodiment, a semantic encoding function is obtained, and the communication data is mapped to the semantic encoding features through the semantic encoding function.

[0033] Optionally, the step of mapping the communication data to the semantic coding features using the semantic coding function may specifically include:

[0034] in, The semantic encoding features are obtained by mapping the communication data corresponding to users 1 to N. For semantic encoding functions, The communication data corresponding to users 1 to N.

[0035] In step S10, the step of semantically encoding the communication data to obtain semantically encoded features includes: Step S11: Obtain a semantic coding model, wherein the semantic coding model is trained by multiple first training samples and the semantic coding model to be trained. Each first training sample consists of a first input feature data and a first real label corresponding to the first input feature data. The first input feature data is training data, and the first real label is the real semantic coding feature corresponding to the training data. Step S12: Semantically encode the communication data using the semantic coding model to obtain the semantic coding features.

[0036] Optionally, in a feasible embodiment, the semantic encoding model includes two convolutional layers, a splitting module, a self-attention module, a residual processing module, and a concatenation module. The two convolutional layers are divided into a first convolutional layer and a second convolutional layer in terms of data processing order. The second convolutional layer is connected to the first convolutional layer, and the data processing order corresponding to the first convolutional layer is before the data processing order corresponding to the second convolutional layer. The self-attention module can be a SwinT (Swin Transformer Block, multi-scale self-attention module) module, the residual processing module can be an RConv (Residual Convolutional Block, residual convolution module) module, the concatenation module can be a concat (connection) module, and the splitting module can be a Split (splitting) module.

[0037] Optionally, referring to the semantic encoding model structure described above, specifically referring to... Figure 3 , Figure 3 Including the semantic coding model (Swin-Conv(SC) Block shown in the diagram), communication data (shown in the diagram) ) and semantic encoding features (illustrated) The semantic encoding model includes a first convolutional layer (a 1×1 Conv at the top of the diagram), a splitting module (Split), a self-attention module (SwinT Block), a residual processing module (RConv Block), a concatenation module (Concat), and a second convolutional layer (a 1×1 Conv at the bottom of the diagram). The first convolutional layer performs feature mapping on the communication data to obtain mapped features, which include a first mapping feature and a second mapping feature. The splitting module splits the mapped features into the first and second mapping features. The self-attention module processes the first mapping feature to obtain a first processed feature. The residual module processes the second mapping feature to obtain a second processed feature. The concatenation module concatenates the first and second processed features to obtain a concatenated feature. The second convolutional layer performs a residual concatenation between the concatenated feature and the communication data to obtain the semantic encoding feature.

[0038] Optionally, the step of performing feature mapping on the communication data through the first convolutional layer to obtain mapped features may specifically include:

[0039] in, For mapping features, For communication data.

[0040] Optionally, the step of processing the first mapping feature through the self-attention module to obtain the first processed feature, and processing the second mapping feature through the residual module to obtain the second processed feature, may specifically include:

[0041] in, These are the first processing feature and the second processing feature. For the first mapping feature, This is the second mapping feature.

[0042] Optionally, the step of performing a residual connection between the concatenated features and the communication data through the second convolutional layer to obtain the semantic encoding features may specifically include:

[0043] in, These are semantic encoding features.

[0044] In step S10, after the step of semantically encoding the communication data to obtain semantically encoded features, the method further includes: Step A10: Obtain a preset knowledge base shared by the transmitter and the receiver; In this embodiment, it should be noted that the preset knowledge base includes a semantic set and a knowledge set.

[0045] Optionally, in one feasible embodiment, a preset configuration file is obtained, wherein the preset configuration file includes the correlation between the first end device name and the second end device name and the knowledge base; the first device name of the transmitting end is obtained, and the second device name of the receiving end is obtained; based on the first device name and the second device name, the preset configuration file is queried to obtain the preset knowledge base jointly corresponding to the transmitting end and the receiving end.

[0046] Step A20: Based on the preset knowledge base, adjust the semantic encoding features.

[0047] For example, the optimal knowledge semantics corresponding to the communication data is determined in the preset knowledge base, and the semantic encoding features are adjusted based on the optimal knowledge semantics.

[0048] Optionally, in a feasible embodiment, a likelihood function is obtained, and based on the semantic set and knowledge set corresponding to the communication data in the preset knowledge base, the optimal knowledge semantics corresponding to the communication data are selected through the likelihood function.

[0049] Optionally, the step of selecting the optimal knowledge semantics corresponding to the communication data based on the semantic set and knowledge set corresponding to the communication data in the preset knowledge base through the likelihood function may specifically include:

[0050] in, For optimal knowledge semantics. Let be the likelihood function. The knowledge set corresponding to the communication data in the preset knowledge base. This refers to the semantic set corresponding to the communication data in the preset knowledge base.

[0051] Step S20: Perform channel coding on the semantic coding features to obtain channel coding features; Optionally, in a feasible embodiment, a channel coding model is obtained, the semantic coding features are input into the channel coding model, and the semantic coding features are mapped to semantic coding features through the channel coding model.

[0052] Alternatively, in another feasible embodiment, a channel coding function is obtained, and the semantic coding features are mapped to the channel coding features through the channel coding function.

[0053] In step S20, the step of performing channel coding on the semantic coding features to obtain channel coding features includes: Step S21, obtain the channel coding model, wherein the channel coding model is trained by multiple second training samples and the channel coding model to be trained, and each second training sample consists of a second input feature data and a second real label corresponding to the second input feature data. The second input feature data is a training semantic coding feature, and the second real label is a real channel coding feature corresponding to the training semantic coding feature. The channel coding model to be trained is jointly trained with the decoding and restoration model to be trained. Step S22: Channel coding is performed on the semantic coding features using the channel coding model to obtain the channel coding features.

[0054] For example, channel state information corresponding to the communication data is obtained, and the semantic coding features are channel-coded according to the channel state information using the channel coding model to obtain the channel coding features.

[0055] Optionally, the step of performing channel coding on the semantic coding features based on the channel state information using the channel coding model to obtain the channel coding features may specifically include:

[0056] in, The channel coding features are obtained by mapping the semantic coding features corresponding to users 1 to N. The semantic encoding features corresponding to the communication data of users 1 to N, This refers to the channel state information corresponding to the communication data of users 1 through N. This is a channel coding model.

[0057] Optionally, in one feasible embodiment, the channel coding model includes three self-attention convolutional modules and three strided convolutional modules. The three self-attention convolutional modules are divided into a first self-attention convolutional module, a second self-attention convolutional module, and a third self-attention convolutional module in the order of processing data. The three strided convolutional modules are divided into a first strided convolutional module, a second strided convolutional module, and a third strided convolutional module in the order of processing data. The self-attention convolutional modules can be Swin-Conv (Swin Convolutional) modules, and the strided convolutional modules can be SConv (Strided Convolutional) modules.

[0058] Optionally, refer to Figure 4 , Figure 4 Including channel coding models and semantic coding features (illustrated) The channel coding model includes a first self-attention convolutional module (Swin-Conv at the leftmost position in the diagram), a first strided convolutional module (SConv at the leftmost position in the diagram), a second self-attention convolutional module (Swin-Conv in the middle position in the diagram), a second strided convolutional module (SConv in the middle position in the diagram), a third self-attention convolutional module (Swin-Conv at the rightmost position in the diagram), and a third strided convolutional module (SConv at the rightmost position in the diagram). The semantic coding features are input into the channel coding model.

[0059] Step S30: Broadcast the channel coding features to the feature subspace corresponding to the communication data, so that the receiving end can pull the channel coding features from the feature subspace and decode the channel coding features to obtain semantic decoding features.

[0060] For example, the channel coding features are mapped to the feature subspaces corresponding to the respective separate communication data.

[0061] Optionally, the step of mapping the channel coding features to the respective feature subspaces corresponding to the separate communication data may specifically include:

[0062] in, For users The feature subspace corresponding to the communication data.

[0063] This application provides a communication optimization method applied to a transmitting end. The method involves semantically encoding communication data to obtain semantically encoded features; channel encoding the semantically encoded features to obtain channel encoded features; and broadcasting the channel encoded features to the feature subspace corresponding to the communication data, allowing the receiving end to retrieve the channel encoded features from the feature subspace. The channel encoded features are then decoded to obtain semantically decoded features. Each transmitting end can broadcast its channel encoded features through the feature subspace corresponding to its transmitted communication data, ensuring orthogonality of multi-end communication transmission in the feature domain. Therefore, when multiple transmitting ends simultaneously transmit their respective communication data's corresponding channel encoded features in the time domain, they do not interfere with each other, thus reducing multi-end communication latency.

[0064] Furthermore, based on the first embodiment described above, a second embodiment of the communication optimization method of the present invention is proposed. In this embodiment, reference is made to... Figure 5 Applied to the receiving end, the communication optimization method includes: Step B10: Extract channel coding features from the feature subspace, wherein the channel coding features are obtained by channel coding the semantic coding features by the transmitter, the semantic coding features are obtained by semantic coding the communication data by the transmitter, and the channel coding features are broadcast through the feature subspace corresponding to the communication data; For example, channel coding features are extracted from the feature subspace corresponding to the receiving end.

[0065] It is understandable that when the transmitter directly transmits the channel coding features corresponding to the receiver to the receiver, noise interference and semantic interference from multiple users may exist during the transmission of the channel coding features. Therefore, the communication signal received by the receiver is as follows:

[0066] in, When the transmitter directly transmits the channel coding features corresponding to the receiver to the receiver, the receiver... The corresponding communication signal obtained is pulled. For the receiving end Corresponding channel coding features For the receiving end The corresponding channel state information of the channel coding features, To exclude the receiving end Other receivers Corresponding channel coding features For the receiving end The corresponding channel noise.

[0067] When the transmitter sends the channel coding features corresponding to the receiver to the feature subspace corresponding to the receiver, since each feature subspace corresponds one-to-one with the receiver, the feature subspace corresponding to the receiver does not contain the channel coding features corresponding to other ends. Therefore, semantic interference between multiple users is eliminated, ensuring accurate communication between the receiver and the transmitter.

[0068] Therefore, when the transmitter sends the channel coding features corresponding to the receiver to the feature subspace corresponding to the receiver, the communication signal received by the receiver is as follows:

[0069] in, The communication signal received by the receiver when the transmitter sends the channel coding features corresponding to the receiver to the feature subspace corresponding to the receiver.

[0070] Step B20: Decode and restore the channel coding features to obtain semantic decoding features.

[0071] Optionally, in one feasible embodiment, a channel decoding model is obtained, the channel coding features are input into the channel decoding model, and the channel coding features are mapped to semantic decoding features through the channel decoding model.

[0072] Alternatively, in another feasible embodiment, a channel decoding function is obtained, and the channel coding features are mapped to the channel decoding features through the channel decoding function.

[0073] In step B20, the step of decoding and restoring the channel coding features to obtain semantic decoding features includes: Step B21, obtain the decoding and restoration model, wherein the decoding and restoration model is trained by multiple third training samples and the decoding and restoration model to be trained. Each third training sample consists of a third input feature data and a third real label corresponding to the third input feature data. The third input feature data is a training channel coding feature, and the third real label is a real semantic coding feature corresponding to the training channel coding feature. The decoding and restoration model to be trained is jointly trained with the channel coding model to be trained. In this embodiment, it should be noted that the structure of the decoding and restoration model corresponds to the structure of the channel coding model. The correspondence is that the structure and number of modules are the same, but the order of the specific modules is reversed. For example, if the module order in the channel coding model is module A, module B, and module C, then the module order in the decoding and restoration model is module C, module B, and module A.

[0074] Step B22: The channel coding features are decoded and restored using the decoding and restoration model to obtain semantic decoding features.

[0075] For example, the channel state information and channel noise (i.e., the communication data pulled from the corresponding feature subspace) corresponding to the channel coding feature are obtained. Based on the channel state information and channel noise corresponding to the channel coding feature, the channel coding feature is decoded and restored using the decoding and restoration model to obtain the semantic decoding feature.

[0076] Optionally, the step of decoding and restoring the channel coding features based on the channel state information and channel noise corresponding to the channel coding features using the decoding and restoration model to obtain semantic decoding features may specifically include:

[0077] in, For the receiving end The corresponding semantic decoding features, This is a decoding and restoration model.

[0078] Optionally, in one feasible embodiment, the decoding and restoration model includes three self-attention convolutional modules and three striding convolutional modules. The three self-attention convolutional modules are divided into a fourth self-attention convolutional module, a fifth self-attention convolutional module, and a sixth self-attention convolutional module in the order of data processing. The three striding convolutional modules are divided into a fourth striding convolutional module, a fifth striding convolutional module, and a sixth striding convolutional module in the order of data processing. The self-attention convolutional modules can be Swin-Conv modules, and the striding convolutional modules can be SConv modules.

[0079] Optionally, refer to Figure 6 , Figure 6 This includes: the channel decoding model, and the communication data extracted from the corresponding feature subspace (as illustrated). ) and semantic decoding features (illustrated) The channel decoding model includes a fourth strided convolutional module (SConv at the leftmost position in the diagram), a fifth strided convolutional module (SConv in the middle position in the diagram), a sixth strided convolutional module (SConv at the rightmost position in the diagram), a fourth self-attention convolutional module (Swin-Conv at the leftmost position in the diagram), a fifth self-attention convolutional module (Swin-Conv in the middle position in the diagram), and a sixth self-attention convolutional module (Swin-Conv at the rightmost position in the diagram). Communication data extracted from the corresponding feature subspace is input into the channel decoding model, and the channel decoding model outputs semantic decoding features.

[0080] In step B20, after decoding and restoring the channel coding features to obtain semantic decoding features, the method further includes: Step B30: Obtain a preset knowledge base shared by the transmitter and the receiver; In this embodiment, it should be noted that the preset knowledge base includes a semantic set and a knowledge set.

[0081] Optionally, in one feasible embodiment, a preset configuration file is obtained, wherein the preset configuration file includes the correlation between the first end device name and the second end device name and the knowledge base; the first device name of the transmitting end is obtained, and the second device name of the receiving end is obtained; based on the first device name and the second device name, the preset configuration file is queried to obtain the preset knowledge base jointly corresponding to the transmitting end and the receiving end.

[0082] Step B40: Based on the preset knowledge base, adjust the semantic features of the semantic decoding features.

[0083] For example, the optimal knowledge semantics corresponding to the semantic decoding feature is determined in the preset knowledge base, and the semantic feature is adjusted based on the optimal knowledge semantics.

[0084] Optionally, the specific implementation of determining the optimal knowledge semantics corresponding to the semantic decoding feature in the preset knowledge base can refer to the specific implementation steps of step A20 above, and will not be repeated here.

[0085] Optionally, since the structure of the decoding and restoration model corresponds to the structure of the channel coding model, the semantic coding model to be trained and the semantic decoding model to be trained can be jointly trained.

[0086] Optionally, refer to Figure 7 , Figure 7It includes a semantic encoding model to be trained (the structure shown in the figure is composed of Swin-Conv, SConv, Swin-Conv, SConv, Swin-Conv, SConv connected in sequence) and a semantic decoding model to be trained (the structure shown in the figure is composed of SConv, Swin-Conv, SConv, Swin-Conv, SConv, Swin-Conv). The three self-attention convolutional modules in the semantic encoding model to be trained are respectively joined with the three sub-attention convolutional modules in the semantic decoding model to be trained. When the semantic encoding model to be trained and the semantic decoding model to be trained are jointly trained, the joint residual of the semantic encoding model to be trained and the semantic decoding model to be trained are calculated, and then the semantic encoding model to be trained and the semantic decoding model to be trained are trained separately according to the joint residual.

[0087] Optionally, refer to Figure 8 , Figure 9 and Figure 10 , Figure 8 , Figure 9 and Figure 10 These represent the accuracy of communication transmission in training scenarios at 0dB, 5dB, and 10dB, respectively, when using the communication optimization method.

[0088] This application provides a communication optimization method applied to a receiving end. It extracts channel coding features from a feature subspace, where the channel coding features are obtained by the transmitting end through channel coding of semantic coding features, and the semantic coding features are obtained by the transmitting end through semantic coding of communication data. The channel coding features are then decoded and restored via the light waves in the feature subspace corresponding to the communication data to obtain semantic decoding features. Each receiving end can extract channel coding features from its respective feature subspace to decode and restore the channel coding features, thus enabling communication between the receiving end and the transmitting end. Furthermore, using the feature subspace as the communication medium ensures the orthogonality of multi-end communication reception in the feature domain. Therefore, multiple receiving ends can simultaneously extract channel coding features from each feature subspace in the time domain without interfering with each other, thereby reducing multi-end communication latency.

[0089] Furthermore, embodiments of the present invention also propose a transmitter, referring to... Figure 11 The transmitting end includes: Semantic encoding module 10 is used to perform semantic encoding on communication data to obtain semantic encoded features; Channel coding module 20 is used to perform channel coding on the semantic coding features to obtain channel coding features; The broadcast module 30 is used to broadcast the channel coding features to the feature subspace corresponding to the communication data, so that the receiving end can pull the channel coding features from the feature subspace and decode the channel coding features to obtain semantic decoding features.

[0090] Furthermore, the semantic encoding module 10 is also used for: Obtain a semantic coding model, wherein the semantic coding model is trained by multiple first training samples and the semantic coding model to be trained, and each first training sample consists of a first input feature data and a first real label corresponding to the first input feature data, wherein the first input feature data is training data and the first real label is the real semantic coding feature corresponding to the training data. The semantic coding model is used to semantically encode the communication data to obtain the semantic coding features.

[0091] Furthermore, the channel coding module 20 is also used for: A channel coding model is obtained, wherein the channel coding model is trained by multiple second training samples and a channel coding model to be trained. Each second training sample consists of a second input feature data and a second real label corresponding to the second input feature data. The second input feature data is a training semantic coding feature, and the second real label is a real channel coding feature corresponding to the training semantic coding feature. The channel coding model to be trained is jointly trained with the decoding and restoration model to be trained. The semantic coding features are channel-coded using the channel coding model to obtain the channel coding features.

[0092] Furthermore, after the step of semantically encoding the communication data to obtain semantically encoded features, the transmitting end further includes: Obtain a preset knowledge base shared by the transmitter and the receiver; Based on the preset knowledge base, the semantic encoding features are adjusted semantically.

[0093] All embodiments of the transmitter of the present invention can refer to the various embodiments of the communication optimization method of the present invention, and will not be described again here.

[0094] Furthermore, embodiments of the present invention also propose a receiving end, referring to... Figure 12 The receiving end includes: The pull module 40 is used to pull channel coding features from the feature subspace, wherein the channel coding features are obtained by channel coding of semantic coding features by the transmitter, the semantic coding features are obtained by semantic coding of communication data by the transmitter, and the channel coding features are transmitted via the light waves of the feature subspace corresponding to the communication data; The decoding module 50 is used to decode and restore the channel coding features to obtain semantic decoding features.

[0095] Furthermore, the decoding module 50 is also used for: A decoding and restoration model is obtained, wherein the decoding and restoration model is trained by multiple third training samples and the decoding and restoration model to be trained. Each third training sample consists of a third input feature data and a third real label corresponding to the third input feature data. The third input feature data is a training channel coding feature, and the third real label is a real semantic coding feature corresponding to the training channel coding feature. The decoding and restoration model to be trained is jointly trained with the channel coding model to be trained. The channel coding features are decoded and restored using the decoding and restoration model to obtain semantic decoding features.

[0096] Furthermore, the decoding and restoration model is obtained by iteratively optimizing the decoding and restoration model to be trained based on the residual corresponding to the channel coding model to be trained.

[0097] Furthermore, after the step of decoding and restoring the channel coding features to obtain semantic decoding features, the receiving end is also used to: Obtain a preset knowledge base shared by the transmitter and the receiver; Based on the preset knowledge base, the semantic decoding features are adjusted semantically.

[0098] All embodiments of the receiver of the present invention can refer to the various embodiments of the communication optimization method of the present invention, and will not be described again here.

[0099] Furthermore, embodiments of the present invention also propose a computer-readable storage medium storing a communication optimization program, which, when executed by a processor, implements the steps of the communication optimization method described below.

[0100] The various embodiments of the communication optimization device and computer-readable storage medium of the present invention can be referred to the various embodiments of the communication optimization method of the present invention, and will not be repeated here.

[0101] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0102] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0103] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0104] The above are merely preferred embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A communication optimization method, characterized in that, Applied to the transmitting end, the communication optimization method includes: Semantic encoding is performed on the communication data to obtain semantic encoding features; Channel coding is performed on the semantic coding features to obtain channel coding features; The channel coding features are broadcast to the feature subspace corresponding to the communication data, so that the receiving end can pull the channel coding features from the feature subspace and decode the channel coding features to obtain semantic decoding features; wherein, the channel coding features are mapped to the respective feature subspaces corresponding to the communication data.

2. The communication optimization method as described in claim 1, characterized in that, The step of semantically encoding the communication data to obtain semantically encoded features includes: Obtain a semantic coding model, wherein the semantic coding model is trained by multiple first training samples and the semantic coding model to be trained, and each first training sample consists of a first input feature data and a first real label corresponding to the first input feature data, wherein the first input feature data is training data and the first real label is the real semantic coding feature corresponding to the training data. The semantic coding model is used to semantically encode the communication data to obtain the semantic coding features.

3. The communication optimization method as described in claim 1, characterized in that, The step of performing channel coding on the semantic coding features to obtain channel coding features includes: A channel coding model is obtained, wherein the channel coding model is trained by multiple second training samples and a channel coding model to be trained. Each second training sample consists of a second input feature data and a second real label corresponding to the second input feature data. The second input feature data is a training semantic coding feature, and the second real label is a real channel coding feature corresponding to the training semantic coding feature. The channel coding model to be trained is jointly trained with the decoding and restoration model to be trained. The semantic coding features are channel-coded using the channel coding model to obtain the channel coding features.

4. The communication optimization method as described in claim 1, characterized in that, After the step of semantically encoding the communication data to obtain semantically encoded features, the method further includes: Obtain a preset knowledge base shared by the transmitter and the receiver; Based on the preset knowledge base, the semantic encoding features are adjusted semantically.

5. A communication optimization method, characterized in that, Applied to the receiving end, the communication optimization method includes: Channel coding features are extracted from the feature subspace, wherein the channel coding features are obtained by channel coding semantic coding features by the transmitter, and the semantic coding features are obtained by semantic coding communication data by the transmitter. The channel coding features are broadcast via the feature subspace corresponding to the communication data; wherein the channel coding features are mapped to their respective separate feature subspaces corresponding to the communication data. The channel coding features are decoded and restored to obtain semantic decoding features.

6. The communication optimization method as described in claim 5, characterized in that, The step of decoding and restoring the channel coding features to obtain semantic decoding features includes: A decoding and restoration model is obtained, wherein the decoding and restoration model is trained by multiple third training samples and the decoding and restoration model to be trained. Each third training sample consists of a third input feature data and a third real label corresponding to the third input feature data. The third input feature data is a training channel coding feature, and the third real label is a real semantic coding feature corresponding to the training channel coding feature. The decoding and restoration model to be trained is jointly trained with the channel coding model to be trained. The channel coding features are decoded and restored using the decoding and restoration model to obtain semantic decoding features.

7. The communication optimization method as described in claim 6, characterized in that, The decoding and restoration model is obtained by iteratively optimizing the decoding and restoration model to be trained based on the residual corresponding to the channel coding model to be trained.

8. The communication optimization method as described in claim 5, characterized in that, After the step of decoding and restoring the channel coding features to obtain semantic decoding features, the method further includes: Obtain a preset knowledge base shared by the transmitter and the receiver; Based on the preset knowledge base, the semantic decoding features are adjusted semantically.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the steps of the communication optimization method according to any one of claims 1 to 4, or to perform the steps of the communication optimization method according to any one of claims 5 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program for implementing the communication optimization method, which is executed by a processor to implement the steps of the communication optimization method as described in any one of claims 1 to 4, or to implement the steps of the communication optimization method as described in any one of claims 5 to 8.

Images