Semantic communication method and apparatus, and device, system, storage medium and program product

Abstract

The present disclosure relates to a semantic communication method and apparatus, and a device, a system, a storage medium and a program product. The method comprises: acquiring a first sample, wherein the first sample is obtained by means of a GAN; and on the basis of the first sample, training a semantic encoding model, wherein the semantic encoding model is used for implementing semantic communication. By means of the solution of the present disclosure, the accuracy of semantic communication can be improved.

Description

Semantic communication method and apparatus, device, system, storage medium and program product TECHNICAL FIELD

[0001] The present disclosure relates to the field of wireless communication, and particularly relates to a semantic communication method and apparatus, device, system, storage medium and program product. BACKGROUND

[0002] In a communication system, semantic communication (SemCom) can be performed between network nodes. Semantic communication can enable transmission based on semantic information, thereby providing better communication services for users.

[0003] SUMMARY

[0004] The present disclosure provides a semantic communication method and apparatus, communication device, communication system, storage medium and program product.

[0005] According to a first aspect of embodiments of the present disclosure, a semantic communication method is provided. The method is performed by a first node. The method comprises: obtaining a first sample, wherein the first sample is obtained by a generative adversarial network (GAN); and training a semantic encoding model based on the first sample, wherein the semantic encoding model is used to implement semantic communication.

[0006] According to a second aspect of embodiments of the present disclosure, a semantic communication method is provided. The method is performed by a second node. The method comprises: obtaining a first sample by a GAN based on first data; and sending the first sample to a first node, wherein the first sample is used to train a semantic encoding model, and the semantic encoding model is used to implement semantic communication.

[0007] According to a third aspect of embodiments of the present disclosure, a semantic communication apparatus is provided. The apparatus is arranged in a first node. The apparatus comprises a processing module. The processing module is configured to: obtain a first sample, wherein the first sample is obtained by a GAN; and train a semantic encoding model based on the first sample, wherein the semantic encoding model is used to implement semantic communication.

[0008] According to a fourth aspect of embodiments of the present disclosure, a semantic communication apparatus is provided. The apparatus is arranged in a second node. The apparatus comprises a processing module and a transceiver module. The processing module is configured to: obtain a first sample by a GAN based on first data. The transceiver module is configured to: send the first sample to a first node, wherein the first sample is used to train a semantic encoding model, and the semantic encoding model is used to implement semantic communication.

[0009] According to a fifth aspect of the embodiments of the present disclosure, a communication device is provided. The communication device includes one or more processors, and a memory storing instructions. The instructions, when executed by the communication device, cause the communication device to implement the semantic communication method according to the first aspect or the second aspect.

[0010] According to a sixth aspect of the embodiments of the present disclosure, a communication system is provided. The communication system includes a first node and a second node. The first node is configured to implement the semantic communication method according to the first aspect. The second node is configured to implement the semantic communication method according to the second aspect.

[0011] According to a seventh aspect of the embodiments of the present disclosure, a storage medium is provided. The storage medium stores instructions. The instructions, when executed on a communication device, cause the communication device to perform the semantic communication method according to the first aspect or the second aspect.

[0012] According to an eighth aspect of the embodiments of the present disclosure, a program product is provided. The program product, when executed by a communication device, causes the communication device to perform the semantic communication method according to the first aspect or the second aspect.

[0013] According to a ninth aspect of the embodiments of the present disclosure, a computer program is provided. The computer program, when executed on a computer, causes the computer to perform the semantic communication method according to the first aspect or the second aspect.

[0014] According to a tenth aspect of the embodiments of the present disclosure, a chip or chip system is provided. The chip or chip system includes a processing circuit. The processing circuit is configured to perform the semantic communication method according to the first aspect or the second aspect.

[0015] According to the embodiments of the present disclosure, the accuracy of semantic communication can be improved.

[0016] It should be understood that the general description above and the detailed description below are only exemplary and explanatory, and do not constitute a limitation on the embodiments of the present disclosure. BRIEF DESCRIPTION OF DRAWINGS

[0017] The drawings herein are incorporated into the specification and constitute a part of the specification, show embodiments consistent with the present disclosure, and together with the specification serve to explain the principles of embodiments of the present disclosure.

[0018] FIG. 1 is an architecture schematic diagram of a communication system according to an embodiment of the present disclosure.

[0019] FIG. 2A is an interaction schematic diagram of a semantic communication method according to an embodiment of the present disclosure.

[0020] FIG. 2B is an interaction schematic diagram of a semantic communication method according to an embodiment of the present disclosure.

[0021] FIG. 2C is an interaction diagram of a semantic communication method according to an embodiment of the present disclosure.

[0022] FIG. 3 is an architecture diagram of a GAN according to an embodiment of the present disclosure.

[0023] FIG. 4A is a flow diagram of a semantic communication method according to an embodiment of the present disclosure.

[0024] FIG. 4B is a flow diagram of a semantic communication method according to an embodiment of the present disclosure.

[0025] FIG. 4C is a flow diagram of a semantic communication method according to an embodiment of the present disclosure.

[0026] FIG. 5A is a flow diagram of a semantic communication method according to an embodiment of the present disclosure.

[0027] FIG. 5B is a flow diagram of a semantic communication method according to an embodiment of the present disclosure.

[0028] FIG. 6A is a flow diagram of a semantic communication method according to an embodiment of the present disclosure.

[0029] FIG. 6B is an interaction diagram of a semantic communication method according to an embodiment of the present disclosure.

[0030] FIG. 7A is a scenario diagram of semantic communication according to an embodiment of the present disclosure.

[0031] FIG. 7B is a training scenario diagram based on a GAN according to an embodiment of the present disclosure.

[0032] FIG. 8 is a structural diagram of a semantic communication apparatus according to an embodiment of the present disclosure.

[0033] FIG. 9A is a structural diagram of a communication device according to an embodiment of the present disclosure.

[0034] FIG. 9B is a structural diagram of a chip according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0035] Embodiments of the present disclosure provide a semantic communication method and apparatus, a communication device, a communication system, a storage medium, and a program product.

[0036] In a first aspect, the embodiments of the present disclosure provide a semantic communication method. The method is performed by a first node. The above method comprises: obtaining a first sample, wherein the first sample is obtained by a GAN; and training a semantic encoding model based on the first sample, wherein the semantic encoding model is used to implement semantic communication.

[0037] According to the embodiment, the first sample generated based on the GAN is obtained, and the semantic encoding model is trained based on the first sample. In this way, a sufficient amount of first samples can be generated by the GAN, thereby ensuring sufficient training of the semantic encoding model. In addition, since the discriminator is used in the GAN to measure the difference between the generated sample and the real sample, the accuracy of the first sample generated by the trained GAN can be ensured, thereby improving the accuracy of the semantic communication implemented by the semantic encoding model.

[0038] With reference to some embodiments of the first aspect, in some embodiments, the operation of obtaining the first sample can include at least one of the following: receiving the first sample sent by the second node; obtaining the first sample based on the first data by the GAN.

[0039] With reference to some embodiments of the first aspect, in some embodiments, the GAN can include a generator; and the operation of obtaining the first sample based on the first data by the GAN can include: obtaining the first sample based on the first data by the generator.

[0040] With reference to some embodiments of the first aspect, in some embodiments, the generator can include a context adjuster and a generative artificial intelligence (GAI) model; and the operation of obtaining the first sample based on the first data by the generator can include: determining a first context by the context adjuster; and obtaining the first sample based on the first stem in the first data and the first context by the GAI model.

[0041] With reference to some embodiments of the first aspect, in some embodiments, the method can further include: training the GAN based on second data.

[0042] With reference to some embodiments of the first aspect, in some embodiments, the training of the GAN can include at least one of the following: training of a generator in the GAN; and training of a discriminator in the GAN.

[0043] With reference to some embodiments of the first aspect, in some embodiments, the training of the generator can include: obtaining a second sample based on the second data by the generator; discriminating the second sample by the discriminator; determining a first loss function value of the generator according to a discrimination result; and adjusting a parameter value of the generator in a case where the first loss function value satisfies a first condition.

[0044] In some embodiments combined with the first aspect, in some embodiments, the training of the discriminator can include: obtaining, by the generator, a second sample based on the second data; discriminating, by the discriminator, the second sample and a third sample in the second data; determining a second loss function value of the discriminator according to a discrimination result; and adjusting a parameter value of the discriminator in a case where the second loss function value satisfies a second condition.

[0045] In some embodiments combined with the first aspect, in some embodiments, the training of the generator and the training of the discriminator can be alternately performed.

[0046] In some embodiments combined with the first aspect, in some embodiments, the method can further include: sending the trained semantic encoding model to a third node, where the trained semantic encoding model is used by the third node to implement semantic communication.

[0047] In a second aspect, the embodiments of the present disclosure provide a semantic communication method. The method is performed by a second node. The method includes: obtaining, by a GAN, a first sample based on first data; and sending the first sample to a first node, where the first sample is used to train a semantic encoding model, and the semantic encoding model is used to implement semantic communication.

[0048] According to the present embodiment, the second node can generate the first sample based on the GAN. The first sample can be used to train the semantic encoding model. In this way, a sufficient amount of first samples can be generated by the GAN, thereby ensuring sufficient training of the semantic encoding model. In addition, since the discriminator is used in the GAN to measure the difference between the generated sample and the real sample, the accuracy of the first sample generated by the trained GAN can be ensured, thereby improving the accuracy of the semantic communication implemented by the semantic encoding model.

[0049] In some embodiments combined with the second aspect, in some embodiments, the GAN can include a generator; and the operation of obtaining, by the GAN, the first sample based on the first data can include: obtaining, by the generator, the first sample based on the first data.

[0050] In some embodiments combined with the second aspect, in some embodiments, the generator can include a context adjuster and a GAI model; and the operation of obtaining, by the generator, the first sample based on the first data can include: determining a first context by the context adjuster; and obtaining, by the GAI model, the first sample based on a first stem in the first data and the first context.

[0051] In some embodiments combined with the second aspect, in some embodiments, the method can further include: training the GAN based on second data.

[0052] In some embodiments of the second aspect, the training for the GAN can include at least one of: training for a generator in the GAN; training for a discriminator in the GAN.

[0053] In some embodiments of the second aspect, the training for the generator can include: obtaining, by the generator, a second sample based on the second data; discriminating, by the discriminator, the second sample; determining a first loss function value of the generator according to a result of the discriminating; and adjusting a parameter value of the generator in a case where the first loss function value satisfies a first condition.

[0054] In some embodiments of the second aspect, the training for the discriminator can include: obtaining, by the generator, a second sample based on the second data; discriminating, by the discriminator, the second sample and a third sample in the second data; determining a second loss function value of the discriminator according to a result of the discriminating; and adjusting a parameter value of the discriminator in a case where the second loss function value satisfies a second condition.

[0055] In some embodiments of the second aspect, the training for the generator and the training for the discriminator can be alternately performed.

[0056] In a third aspect, the embodiments of the present disclosure provide a semantic communication apparatus. The apparatus is arranged at a first node. The apparatus includes a processing module. The processing module is configured to: obtain a first sample, wherein the first sample is obtained by a GAN; and train a semantic encoding model based on the first sample, wherein the semantic encoding model is used to implement semantic communication.

[0057] In some embodiments of the third aspect, the processing module can be configured to perform at least one of: receiving, by a receiving module, the first sample sent by a second node; and obtaining, by the GAN, the first sample based on first data.

[0058] In some embodiments of the third aspect, the GAN can include a generator; and the processing module can be configured to: obtain, by the generator, the first sample based on the first data.

[0059] In some embodiments of the third aspect, the generator can include a context adjuster and a GAI model; and the processing module can be configured to: determine, by the context adjuster, a first context; and obtain, by the GAI model, the first sample based on a first stem in the first data and the first context.

[0060] In some embodiments of the third aspect, the processing module can be further configured to: train the GAN based on second data. In some embodiments of the third aspect, the processing module can be further configured to: train the GAN based on second data.

[0061] With some embodiments of the third aspect, in some embodiments, the training for the GAN can include at least one of: training for a generator in the GAN; training for a discriminator in the GAN.

[0062] With some embodiments of the third aspect, in some embodiments, the training for the generator can include: obtaining, by the generator, a second sample based on the second data; discriminating, by the discriminator, the second sample; determining a first loss function value of the generator according to a result of the discriminating; and adjusting a parameter value of the generator in a case where the first loss function value satisfies a first condition.

[0063] With some embodiments of the third aspect, in some embodiments, the training for the discriminator can include: obtaining, by the generator, a second sample based on the second data; discriminating, by the discriminator, the second sample and a third sample in the second data; determining a second loss function value of the discriminator according to a result of the discriminating; and adjusting a parameter value of the discriminator in a case where the second loss function value satisfies a second condition.

[0064] With some embodiments of the third aspect, in some embodiments, the training for the generator and the training for the discriminator can be alternately performed.

[0065] With some embodiments of the third aspect, in some embodiments, the apparatus can further include a transceiver module; the transceiver module can be configured to: send the trained semantic encoding model to a third node, wherein the trained semantic encoding model is used by the third node to implement semantic communication.

[0066] In a fourth aspect, the embodiments of the present disclosure provide a semantic communication apparatus. The apparatus is arranged at a second node. The apparatus includes a processing module and a transceiver module. The processing module is configured to: obtain, by a GAN, a first sample based on first data. The transceiver module is configured to: send the first sample to a first node, wherein the first sample is used to train a semantic encoding model, and the semantic encoding model is used to implement semantic communication.

[0067] With some embodiments of the fourth aspect, in some embodiments, the GAN can include a generator; wherein the processing module can be configured to: obtain, by the generator, the first sample based on the first data.

[0068] With some embodiments of the fourth aspect, in some embodiments, the generator can include a context adjuster and a GAI model; wherein the processing module can be configured to: determine, by the context adjuster, a first context; and obtain, by the GAI model, the first sample based on a first stem in the first data and the first context.

[0069] In some embodiments combining with the fourth aspect, in some embodiments, the processing module can be further configured to train the GAN based on the second data.

[0070] In some embodiments combining with the fourth aspect, in some embodiments, the training for the GAN can include at least one of: training for a generator in the GAN; training for a discriminator in the GAN.

[0071] In some embodiments combining with the fourth aspect, in some embodiments, the training for the generator can include: obtaining, by the generator, a second sample based on the second data; discriminating, by the discriminator, the second sample; determining a first loss function value of the generator according to a result of the discriminating; and adjusting a parameter value of the generator in a case where the first loss function value satisfies a first condition.

[0072] In some embodiments combining with the fourth aspect, in some embodiments, the training for the discriminator can include: obtaining, by the generator, a second sample based on the second data; discriminating, by the discriminator, the second sample and a third sample in the second data; determining a second loss function value of the discriminator according to a result of the discriminating; and adjusting a parameter value of the discriminator in a case where the second loss function value satisfies a second condition.

[0073] In some embodiments combining with the fourth aspect, in some embodiments, the training for the generator and the training for the discriminator can be alternately performed.

[0074] In a fifth aspect, the embodiments of the present disclosure provide a communication device. The communication device includes one or more processors, and a memory storing instructions. The instructions, when executed by the communication device, cause the communication device to implement the semantic communication method according to any one of the first aspect and possible implementation manners thereof.

[0075] In a sixth aspect, the embodiments of the present disclosure provide a communication device. The communication device includes one or more processors, and a memory storing instructions. The instructions, when executed by the communication device, cause the communication device to implement the semantic communication method according to any one of the second aspect and possible implementation manners thereof.

[0076] In a seventh aspect, the embodiments of the present disclosure provide a communication system. The communication system includes a first node and a second node. The first node is configured to implement the semantic communication method according to any one of the first aspect and possible implementation manners thereof. The second node is configured to implement the semantic communication method according to any one of the second aspect and possible implementation manners thereof.

[0077] In an eighth aspect, an embodiment of the present disclosure provides a storage medium. The storage medium stores instructions. The instructions, when executed on a communication device, cause the communication device to perform the semantic communication method according to any one of the first aspect, the second aspect, and possible implementation manners thereof.

[0078] In a ninth aspect, an embodiment of the present disclosure provides a program product. The program product, when executed by a communication device, causes the communication device to perform the semantic communication method according to any one of the first aspect, the second aspect, and possible implementation manners thereof.

[0079] In a tenth aspect, an embodiment of the present disclosure provides a computer program. The computer program, when executed on a computer, causes the computer to perform the semantic communication method according to any one of the first aspect, the second aspect, and possible implementation manners thereof.

[0080] In an eleventh aspect, an embodiment of the present disclosure provides a chip or chip system. The chip or chip system includes processing circuitry. The processing circuitry is configured to perform the semantic communication method according to any one of the first aspect, the second aspect, and possible implementation manners thereof.

[0081] It can be understood that the above semantic communication apparatus, communication device, communication system, storage medium, program product, computer program, chip, and chip system are all used to perform the method provided by the embodiments of the present disclosure. Therefore, the beneficial effects that can be achieved are referred to the beneficial effects in the corresponding method, which will not be described here.

[0082] Embodiments of the present disclosure provide a semantic communication method and apparatus, a communication device, a communication system, a storage medium, and a program product. In some embodiments, the terms of semantic communication method, information processing method, and information transmission method can be replaced with each other, and the terms of semantic communication apparatus, communication device, network device, network function, and network entity can be replaced with each other, and the terms of communication system and information processing system can be replaced with each other.

[0083] Embodiments of the present disclosure are not exhaustive, but only illustrate some embodiments, and are not specific limitations on the protection scope of the present disclosure. In the case of no contradiction, each step in an embodiment can be implemented as an independent embodiment, and the steps can be combined arbitrarily, for example, the scheme after removing some steps in an embodiment can also be implemented as an independent embodiment, and the order of the steps in an embodiment can be exchanged arbitrarily, in addition, the optional implementation manners in an embodiment can be combined arbitrarily; in addition, the embodiments can be combined arbitrarily, for example, the steps of different embodiments or all steps of different embodiments can be combined arbitrarily, an embodiment can be combined with the optional implementation manners of other embodiments.

[0084] In the embodiments of the present disclosure, the terms and / or descriptions among the embodiments are consistent and can be referred to each other if there is no special description and logical conflict, and the technical features in different embodiments can be combined to form new embodiments according to their inherent logical relationship.

[0085] The terms used in the embodiments of the present disclosure are only for the purpose of describing particular embodiments and are not used as limitations of the present disclosure.

[0086] In the embodiments of the present disclosure, unless otherwise specified, the elements expressed in singular form, such as "one", "a", "an", "the", "above", "said", "preceding", "this", etc., can represent "one and only one", or can represent "one or more", "at least one", etc. For example, in the case of using articles such as "a", "an", "the" in English, the noun after the article can be understood as singular expression, or can be understood as plural expression.

[0087] In the embodiments of the present disclosure, "plurality" means two or more.

[0088] In some embodiments, the terms "at least one", "one or more", etc. can be replaced with each other.

[0089] In some embodiments, the writing methods such as "at least one of A, B", "A and / or B", "A in one case, B in another case", "in response to a case A, in response to another case B", etc. can include the following technical solutions according to the case: in some embodiments, A (A is executed regardless of B); in some embodiments, B (B is executed regardless of A); in some embodiments, A and B are selected to be executed (A and B are selectively executed); in some embodiments, A and B (A and B are both executed). When there are more branches such as A, B, C, etc., it is similar to the above.

[0090] In some embodiments, the writing methods such as "A or B", etc. can include the following technical solutions according to the case: in some embodiments, A (A is executed regardless of B); in some embodiments, B (B is executed regardless of A); in some embodiments, A and B are selected to be executed (A and B are selectively executed). When there are more branches such as A, B, C, etc., it is similar to the above.

[0091] The prefix words of "first", "second" and the like in the embodiments of the present disclosure are merely used to distinguish different description objects, and do not constitute limitation on the position, order, priority, quantity or content of the description objects. The description objects are described in the claims or embodiments, and should not be construed as redundant limitation because of the use of the prefix words. For example, the description object is "field", and the ordinal words before "field" in "first field" and "second field" do not limit the position or order between "fields", and "first" and "second" do not limit whether the "fields" modified thereby are in the same message or not, nor limit the order of "first field" and "second field". For another example, the description object is "level", and the ordinal words before "level" in "first level" and "second level" do not limit the priority between "levels". For another example, the quantity of the description object is not limited by the ordinal words, and can be one or more. For example, "first device", wherein the quantity of "device" can be one or more. In addition, the objects modified by different prefix words can be the same or different, for example, the description object is "device", and "first device" and "second device" can be the same device or different devices, and the types thereof can be the same or different. For another example, the description object is "information", and "second information" and "first information" can be the same information or different information, and the contents thereof can be the same or different.

[0092] In some embodiments, "including A", "containing A", "for indicating A", "carrying A" can be interpreted as directly carrying A, or indirectly indicating A.

[0093] In some embodiments, the terms of "in response to", "in response to determining", "in the case of", "when", "when", "if", "if" and the like can be replaced with each other.

[0094] In some embodiments, the terms of "greater than", "greater than or equal to", "not less than", "more than", "more than or equal to", "not less than", "higher than", "higher than or equal to", "not lower than", "above" and the like can be replaced with each other, and the terms of "less than", "less than or equal to", "not greater than", "less than", "less than or equal to", "not more than", "lower than", "lower than or equal to", "not higher than", "below" and the like can be replaced with each other.

[0095] In some embodiments, an apparatus or the like can be interpreted as an entity, and can also be interpreted as virtual, and the name thereof is not limited to the name described in the embodiments. The terms "apparatus", "equipment", "device", "circuit", "network element", "node", "function", "unit", "section", "system", "network", "chip", "chip system", "entity", "subject" and the like can be replaced with each other.

[0096] In some embodiments, a "network" can be interpreted as an apparatus (for example, an access network device, a core network device, and the like) included in the network.

[0097] In some embodiments, the terms "access network device (AN device)", "radio access network device (RAN device)", "base station (BS)", "radio base station", "fixed station", "node", "access point", "transmission point (TP)", "reception point (RP)", "transmission / reception point (TRP)", "panel", "antenna panel", "antenna array", "cell", "macro cell", "small cell", "femto cell", "pico cell", "sector", "cell group", "serving cell", "carrier", "component carrier", "bandwidth part (BWP)" and the like can be replaced with each other.

[0098] In some embodiments, the terms "terminal," "terminal device," "user equipment (UE)," "user terminal," "mobile station (MS)," "mobile terminal (MT)," "subscriber station," "mobile unit," "subscriber unit," "wireless unit," "remote unit," "mobile device," "wireless device," "wireless communication device," "remote device," "mobile subscriber station," "access terminal," "mobile terminal," "wireless terminal," "remote terminal," "handset," "user agent," "mobile client," "client," and so on can be replaced with each other.

[0099] In some embodiments, the access network device, the core network device, or the network device can be replaced with a terminal. For example, the embodiments of the present disclosure can also be applied to a structure in which communication between the access network device, the core network device, or the network device and the terminal is replaced with communication between a plurality of terminals (e.g., device-to-device (D2D), vehicle-to-everything (V2X), etc.). In this case, the terminal can also be configured to have all or part of the functions of the access network device. In addition, the terms "uplink," "downlink," and the like can also be replaced with terms corresponding to the inter-terminal communication (e.g., "side"). For example, the uplink channel, the downlink channel, and the like can be replaced with the side channel, and the uplink, the downlink, and the like can be replaced with the sidelink.

[0100] In some embodiments, the terminal can be replaced with the access network device, the core network device, or the network device. In this case, the access network device, the core network device, or the network device can also be configured to have all or part of the functions of the terminal.

[0101] In some embodiments, the data, information, etc. can be obtained in compliance with the laws and regulations of the country where the location is situated.

[0102] In some embodiments, the data, information, etc. can be obtained after obtaining the consent of the user.

[0103] In addition, each element, each row, or each column in the table of the embodiments of the present disclosure can be implemented as an independent embodiment, and any combination of any element, any row, or any column can also be implemented as an independent embodiment.

[0104] FIG. 1 is a schematic diagram of an architecture of a communication system according to an embodiment of the present disclosure. As shown in FIG. 1, the communication system 100 includes a first node 101, a second node 102, and a third node 103.

[0105] In some embodiments, the first node 101 can be configured to train and / or infer a GAN, and train a semantic encoding model.

[0106] In some embodiments, the second node 102 can be configured to train and / or infer a GAN.

[0107] In some embodiments, the third node 103 can be configured to implement semantic communication based on a semantic encoding model.

[0108] In some embodiments, the first node 101 can be a network device. For example, the first node 101 can be a core network device, a server, etc.

[0109] In some embodiments, the second node 102 can be a network device. For example, the first node 101 can be a core network device, a server, etc.

[0110] In some embodiments, the third node 103 can be a terminal device and / or or a network device. For example, the first node 101 can be a terminal, an access network device, a core network device, a server, etc.

[0111] In some embodiments, the server includes at least one of an application function (AF), an application server (AS), an operator server, but is not limited thereto.

[0112] In some embodiments, the terminal includes at least one of, for example, a mobile phone, a wearable device, an Internet of Things device, a communication-capable automobile, a smart automobile, a Pad, a wireless-transceiving-capable computer, a virtual reality (VR) terminal device, an augmented reality (AR) terminal device, a wireless terminal device in industrial control, a wireless terminal device in self-driving, a wireless terminal device in remote medical surgery, a wireless terminal device in a smart grid, a wireless terminal device in transportation safety, a wireless terminal device in a smart city, a wireless terminal device in a smart home, but is not limited thereto.

[0113] In some embodiments, the network device includes at least one of, for example, an access network device, a core network element.

[0114] In some embodiments, the access network device is, for example, a node or device that accesses a terminal to a wireless network. In some embodiments, the access network device can include at least one of an evolved NodeB (eNB) in a 5G communication system, a next generation eNB (ng-eNB), a next generation NodeB (gNB), a node B (NB), a home node B (HNB), a home evolved node B (HeNB), a wireless backhaul device, a radio network controller (RNC), a base station controller (BSC), a base transceiver station (BTS), a base band unit (BBU), a mobile switching center, a base station in a 6G communication system, an Open RAN, a Cloud RAN, a base station in other communication systems, an access node in a Wi-Fi system, but is not limited thereto.

[0115] In some embodiments, the technical solutions of the present disclosure can be applied to an Open RAN architecture, at which time, the interfaces between or within the access network devices involved in the embodiments of the present disclosure can become internal interfaces of the Open RAN, and the processes and information interactions between these internal interfaces can be implemented through software or programs.

[0116] In some embodiments, the access network device can be composed of a central unit (CU) and a distributed unit (DU), where the CU can also be referred to as a control unit. The CU-DU structure can split the protocol layers of the access network device, with some protocol layer functions being controlled by the CU, and the remaining or all protocol layer functions being distributed in the DU and controlled by the CU. However, the present disclosure is not limited thereto.

[0117] In some embodiments, the core network element can be one device, or a plurality of devices or device groups. The network element can be virtual or physical. The core network includes at least one of an evolved packet core (EPC), a 5G core network (5GCN), and a next generation core (NGC), for example.

[0118] In some embodiments, the communication system 100 described above can be a 4G communication system, a 5G communication system, or a 6G communication system. It should be noted that the communication system 100 can also be other communication systems, and the embodiments of the present disclosure do not make specific limitations thereto.

[0119] With the development of communication technology, the amount of data transmitted in the communication network has experienced explosive growth. In particular, the number of physical entities and digital entities deployed in the communication network is increasing. In contrast, the lack of infrastructure deployed in the communication network leads to a shortage of communication resources, which makes data transmission more and more difficult to meet.

[0120] In some embodiments, data transmission can be implemented using semantic communication. Semantic communication is a task-oriented communication method, the core of which is to understand and transmit the meaning of information. In some embodiments, in semantic communication, the sender can extract, compress, and transmit features of the original signal to obtain semantic-level information, and transmit the semantic-level information to the receiver.

[0121] Therefore, how to improve the accuracy of semantic communication is a problem to be solved.

[0122] FIG. 2A is an interaction diagram of a semantic communication method according to an embodiment of the present disclosure. The semantic communication method according to an embodiment of the present disclosure can be applied to the communication system 100. As shown in FIG. 2A, the semantic communication method according to an embodiment of the present disclosure includes steps S2101 to S2106.

[0123] In step S2101, the first node 101 obtains second data.

[0124] In some embodiments, the second data obtained by the first node 101 can be used to implement training of a GAN. In some embodiments, the second data can be referred to as GAN training data, GAN training samples, a GAN training dataset, etc.

[0125] In some embodiments, the GAN can be used to implement generation of data. In some embodiments, the GAN can be used to generate training data for a semantic encoding model.

[0126] In some embodiments, the second data can include training data for the GAN.

[0127] In some embodiments, the second data can include at least one of a second stem, a third context, a third sample.

[0128] In some embodiments, the second stem can include key information. For example, the second stem can include information input by a user.

[0129] In some embodiments, the data type of the second stem can include at least one of text, image, voice, but is not limited thereto.

[0130] In some embodiments, the third context can represent background information related to the key information.

[0131] In some embodiments, the data type of the third context can include at least one of text, image, voice, but is not limited thereto.

[0132] In some embodiments, the third sample can be real information obtained for the key information and the related background information.

[0133] In some embodiments, the data type of the third sample can include at least one of text, image, voice, but is not limited thereto.

[0134] In some embodiments, the second stem can be a description text of a question, the third context can be context information of the description text, and the third sample can be a description text of a real answer to the description text and the related context. In some embodiments, the second stem can be an image of a question, the third context can be context information of the image, and the third sample can be a description image or a description text of a real answer to the description image and the related context.

[0135] In some embodiments, the second data can include the second stem and the third sample.

[0136] In some embodiments, the second data can include the second stem, the third context, and the third sample.

[0137] In some embodiments, the number of the second stems included in the second data can be one or more. In some embodiments, the number of the third contexts included in the second data can be one or more. In some embodiments, the number of the third samples included in the second data can be one or more. In some embodiments, the second stems and the third contexts in the second data can be combined in any manner. In an example, in the second data, the combination between one second stem and any third context can correspond to one third sample. In some embodiments, one second stem in the second data can correspond to one or more third contexts. In an example, in the second data, the combination between one second stem and any third context of the one or more third contexts corresponding to the second stem can correspond to one third sample.

[0138] In some embodiments, the second data can be stored in the first node 101. For example, the second data can be stored in a database deployed in the first node 101, and then the first node 101 can obtain the second data from the local database.

[0139] In some embodiments, the second data can be received by the first node 101 from one or more other nodes. For example, the second data can be stored in other nodes, and then the first node 101 can receive the second data from the other nodes.

[0140] In some embodiments, the second data can be historical data. In an example, the second data can include historical data obtained from a database or a server. For example, the first node 101 can be a GAI server, and the GAI server can save historical data. At this time, the GAI server can obtain the historical data from the local and use it as the second data. For example, the first node 101 can be different from the GAI server, and the GAI server can save historical data. At this time, the first node 101 can obtain the historical data from the GAI server and use it as the second data.

[0141] In step S2102, the first node 101 trains the GAN.

[0142] In some embodiments, the first node 101 can train the GAN based on the second data.

[0143] FIG. 3 is a schematic diagram of an architecture of a GAN according to an embodiment of the present disclosure. As shown in FIG. 3, in some embodiments, the GAN can include a generator 301 and a discriminator 302. The generator 301 can be configured to generate data. The discriminator 302 can be configured to determine whether input data is real data or generated data.

[0144] In some embodiments, the objective of the generator 301 is to make the discriminator 302 unable to determine that the input data is generated by the generator 301, and the objective of the discriminator 302 is to more accurately determine whether the input data is generated data or real data. Therefore, an antagonistic relationship is formed between the generator 301 and the discriminator 302.

[0145] In some embodiments, step S2102 can include: obtaining, by the generator 301, the second sample; determining, by the discriminator 302, the second sample and / or the third sample; and training the GAN according to the determination result.

[0146] In some embodiments, the training of the GAN can include at least one of the following: training of the generator 301 in the GAN, and training of the discriminator 302 in the GAN.

[0147] In some embodiments, the operation of obtaining, by the generator 301, the second sample can include: obtaining, by the generator 301, the second sample based on the second data. In some embodiments, the second data can include a second stem and / or a third context. The second stem and / or the third context can be input into the generator 301 to generate the second sample.

[0148] In some embodiments, the second data can include a second stem. In this case, the second stem can be input into the generator 301 to generate the second sample. In some embodiments, one or more second samples can be generated for one second stem.

[0149] In some embodiments, the second data can include a second stem and a third context. In this case, the second stem and the third context can be input into the generator 301 to generate the second sample. In some embodiments, one second sample can be generated for one second stem and one third context. In some embodiments, one or more second samples can be generated for one second stem and one third context.

[0150] In some embodiments, the second stem and / or the third context input into the generator 301 can be raw data. In some embodiments, the second stem and / or the third context input into the generator 301 can be embeddings generated based on the raw data.

[0151] In some embodiments, as shown in FIG. 3, the generator 301 can include a context adjuster 3011 and a GAI model 3012. In some embodiments, the context adjuster 3011 can be used to determine a fourth context. The fourth context is used as input into the GAI model 3012. In some embodiments, the GAI model 3012 can be used to determine a second sample based on the input fourth context and the second stem.

[0152] In some embodiments, in a case where the second data includes a third context, the context adjuster 3011 can determine a fourth context based on the third context. For example, the context adjuster 3011 can obtain the fourth context based on the third context through an adjustment operation. In some embodiments, based on one third context, the context adjuster 3011 can obtain one or more fourth contexts.

[0153] In some embodiments, the fourth context can belong to the third context. In an example, the context adjuster 3011 can select one or more third contexts as the fourth context for one second stem through an adjustment operation. In some embodiments, the fourth context can be a new context. In an example, the context adjuster 3011 can generate one or more new contexts as the fourth context based on one third context through an adjustment operation.

[0154] In some embodiments, in a case where the second data does not include a third context, the context adjuster 3011 can autonomously generate the third context. For example, the context adjuster 3011 can obtain the fourth context through an adjustment operation. In some embodiments, based on one third context, the context adjuster 3011 can obtain one or more fourth contexts.

[0155] In some embodiments, after the second stem and the fourth context are input into the GAI model 3012, the GAI model 3012 can generate the second sample.

[0156] In some embodiments, the generator 301 can first generate a feature vector, and then determine the second sample according to the feature vector. In this way, the feature vector can be considered as a probability distribution of the second sample. In some embodiments, the feature vector can be used to represent a probability distribution of information obtained for the second stem and the fourth context. For example, the second stem can be a description text of a question, and the fourth context can be context information of the description text, and the second sample can be a probability distribution of the description text generated for the description text and the related context. According to the probability distribution of the description text, the generator 301 can obtain the description text. For example, the second stem can be a description image of a question, and the fourth context can be context information of the description image, and the second sample can be a probability distribution of the description image or the description text generated for the description image and the related context. According to the probability distribution of the description image, the generator 301 can obtain the description image.

[0157] In some embodiments, the GAI model 3012 can be deployed in the first node 101. In some embodiments, the second stem and the fourth context can be input into the GAI model 3012 deployed locally in the first node 101 to obtain the second sample.

[0158] In some embodiments, the GAI model 3012 can be deployed in another node different from the first node 101. For example, the GAI model 3012 can be deployed in a GAI server. The GAI server can be used to provide inference services based on the GAI model. In some embodiments, the first node 101 can invoke the GAI model 3012 in the GAI server to obtain the second sample based on the second stem and the fourth context. For example, the first node 101 can send the second stem and the fourth context to the GAI server, and receive the second sample returned by the GAI server.

[0159] In some embodiments, after the generator 301 outputs the second sample, the discriminator 302 can discriminate the second sample.

[0160] In some embodiments, the discriminator 302 can discriminate the third sample in the second data.

[0161] In some embodiments, the discriminator 302 can be used to discriminate whether the data is true. In some embodiments, the discriminator 302 can be used to discriminate whether the second sample input into the discriminator 302 is true. For example, the discriminator 302 can be used to discriminate whether the second sample is a real sample or a generated sample. In some embodiments, the discriminator 302 can be used to discriminate whether the real input into the discriminator 302 is true. For example, the discriminator 302 can be used to discriminate whether the third sample is a real sample or a generated sample.

[0162] In some embodiments, the discrimination by the discriminator 302 can result in a discrimination result. The discrimination result can include true or false. The discrimination result being true indicates that the input sample is discriminated as a real sample. The discrimination result being false (or not true) indicates that the input sample is discriminated as a generated sample.

[0163] In some embodiments, in the case of training the generator 301, the discriminator 302 can discriminate only the second sample.

[0164] In some embodiments, in the case of training the discriminator 302, the discriminator 302 can discriminate the second sample and the third sample.

[0165] In some embodiments, the discrimination result of the discrimination of the second sample by the discriminator 302 can include: determining the second sample as a real sample, determining the second sample as a generated sample.

[0166] In some embodiments, the discrimination result of the discrimination of the third sample by the discriminator 302 can include: determining the third sample as a real sample, determining the third sample as a generated sample.

[0167] In some embodiments, after obtaining the discrimination result, the loss function can be determined based on the discrimination result; and the parameter value of the GAN can be adjusted in the case that the loss function meets a preset condition.

[0168] In some embodiments, the training of the generator 301 in the GAN and the training of the generator 301 in the GAN can be alternately performed. In an example, the parameter value of the generator 301 can be adjusted based on at least part of the second data first, then the parameter value of the discriminator 302 can be adjusted based on at least part of the second data, then the parameter value of the generator 301 can be adjusted based on at least part of the second data, and so on. In an example, the parameter value of the discriminator 302 can be adjusted based on at least part of the second data first, then the parameter value of the generator 301 can be adjusted based on at least part of the second data, then the parameter value of the discriminator 302 can be adjusted based on at least part of the second data, and so on.

[0169] In some embodiments, in the training of the generator 301, the first loss function can be determined based on the discrimination result of the discriminator 302; and the parameter value of the generator 301 can be adjusted in the case that the first loss function meets a first condition. In an example, the first loss function can be determined based on the discrimination result of the second sample.

[0170] In some embodiments, the first loss function can be used to represent the similarity between the second sample and the real data. In an example, the discriminator 302 can generate a feature vector for the second sample. The feature vector represents a probability distribution of the discrimination result of the second sample. Based on the feature vector, the first loss function can be obtained.

[0171] In some embodiments, the first loss function can be represented as: L1 = H(l, x); where L1 represents the first loss function, H represents the cross-entropy, 1 represents that the sample is the real sample, x represents the probability distribution of the discrimination result of the second sample, and H(l, x) represents the distance between the discrimination result of the second sample and 1. Here, the smaller the distance between the discrimination result of the second sample and 1, the closer the second sample is to the real sample, and the better the performance of the generator 301.

[0172] In some embodiments, the parameter values of the generator 301 can be adjusted when the first loss function satisfies a first condition. In an example, the parameter values of the context regulator 3011 in the generator 301 can be adjusted when the first loss function satisfies the first condition. In an example, the parameter values of the GAI model 3012 in the generator 301 can be adjusted when the first loss function satisfies the first condition. In some embodiments, the first condition can be a preset threshold. In an example, the parameter values of the generator 301 can be adjusted when the first loss function is greater than the preset threshold. In an example, the parameter values of the generator 301 can be adjusted when the first loss function is greater than or equal to the preset threshold.

[0173] In some embodiments, the adjustment of the parameter values of the generator 301 can be stopped when the first loss function does not satisfy the first condition.

[0174] In some embodiments, in the training of the discriminator 302, a second loss function can be determined based on the discrimination result of the discriminator 302, and the parameter values of the discriminator 302 can be adjusted when the second loss function satisfies a second condition. In an example, the second loss function can be determined based on the discrimination result of the second sample and the discrimination result of a third sample.

[0175] In some embodiments, the second loss function can be used to represent the similarity between the second sample and the generated data, and the similarity between the third sample and the real data. In an example, the discriminator 302 can generate a feature vector for the second sample and a feature vector for the third sample, respectively. The feature vectors respectively represent a probability distribution of the discrimination result of the second sample and a probability distribution of the discrimination result of the third sample. Based on the feature vectors, the second loss function can be obtained.

[0176] In some embodiments, the second loss function can be represented as: L2=H(1,y)+H(0,x); where L2 represents the second loss function, H represents cross-entropy, 1 represents that the sample is a real sample, 0 represents that the sample is a generated sample, x represents the probability distribution of the discrimination result of the second sample, y represents the third sample, H(1,y) represents the distance between the third sample and 1, and H(0,x) represents the distance between the discrimination result of the second sample and 0. Here, the smaller the distance between the discrimination result of the second sample and 0 and the smaller the distance between the third sample and 1, the closer the second sample is to the generated sample and the closer the third sample is to the real sample, and the better the performance of the discriminator 302.

[0177] In some embodiments, the parameter value of the discriminator can be adjusted when the second loss function satisfies a second condition. In some embodiments, the second condition can be a preset threshold. In an example, the parameter value of the discriminator 302 can be adjusted when the second loss function is greater than the preset threshold. In an example, the parameter value of the discriminator 302 can be adjusted when the second loss function is greater than or equal to the preset threshold.

[0178] In some embodiments, the adjustment of the parameter value of the discriminator 302 can be stopped when the second loss function does not satisfy the second condition.

[0179] In some embodiments, the probability distribution of all second samples can be obtained based on the probability distribution of the second sample. In some embodiments, the probability distribution of all third samples can be obtained based on the probability distribution of the third sample. In some embodiments, when the probability distribution of all second samples is close to the probability distribution of all third samples, the discrimination result obtained by the discriminator 302 is more likely to be a real sample. In some embodiments, when the probability distribution of all second samples is close to the probability distribution of all third samples, the discrimination result obtained by the discriminator 302 is more likely to be a generated sample.

[0180] In some embodiments, the training target of the GAN includes that the generator 301 accurately generates the second sample and the discriminator 302 accurately distinguishes the real sample and the generated sample.

[0181] In some embodiments, the training target of the GAN can be represented as:

[0182] where G represents the generator 301; D represents the discriminator 302; represents that L(G,D) is minimized for the generator 301; represents that L(G,D) is maximized for the discriminator 302; H represents the set of second samples; Q represents the set of second stems; a qrepresenting the second sample; representing the probability distribution of all third samples; representing the probability distribution of all third samples.

[0183] In some embodiments, may be used to represent the probability that the third sample is determined to be a real sample; Q(a q ) represents the probability that the second sample is determined to be a real sample.

[0184] In some embodiments, the first node 101 can implement the training of the GAN through step S2102. After one or more times of training of the generator 301 and one or more times of training of the discriminator 302, the trained GAN can be obtained.

[0185] In step S2103, the first node 101 obtains first data.

[0186] In some embodiments, the first data obtained by the first node 101 can be used to implement the inference based on the GAN. In some embodiments, the first data can be referred to as GAN inference data, GAN inference sample, GAN inference dataset, etc.

[0187] In some embodiments, the first data can be used for the GAN to generate training data for the semantic encoding model.

[0188] In some embodiments, the first data can include at least one of the following: a first stem, a second context.

[0189] In some embodiments, the first stem can include key information. For example, the first stem can include information input by a user.

[0190] In some embodiments, the data type of the first stem can include at least one of text, image, voice, but is not limited thereto.

[0191] In some embodiments, the second context can represent background information related to the key information.

[0192] In some embodiments, the data type of the second context can include at least one of text, image, voice, but is not limited thereto.

[0193] In some embodiments, the first stem can be a description text of a question, and the second context can be context information of the description text. In some embodiments, the first stem can be an image of a question, and the second context can be context information of the image.

[0194] In some embodiments, the first data can include the first stem.

[0195] In some embodiments, the first data can include a first stem, a second context.

[0196] In some embodiments, the number of the first stems included in the first data can be one or more. In some embodiments, the number of the second contexts included in the first data can be one or more. In some embodiments, the first stems and the second contexts in the first data can be combined in any manner.

[0197] In some embodiments, the first data can be stored in the first node 101. For example, the first data can be stored in a database deployed in the first node 101, and then the first node 101 can obtain the first data from the local database.

[0198] In some embodiments, the first data can be received by the first node 101 from one or more other nodes. For example, the first data can be stored in other nodes, and then the first node 101 can receive the first data from the other nodes.

[0199] In some embodiments, the first data can be historical data. In an example, the first data can include historical data obtained from a database or a server. For example, the first node 101 can be a GAI server, and the GAI server can save historical data. At this time, the GAI server can obtain the historical data locally and use it as the first data. For example, the first node 101 can be different from the GAI server, and the GAI server can save historical data. At this time, the first node 101 can obtain the historical data from the GAI server and use it as the first data.

[0200] In some embodiments, the first data can be real-time data. In an example, the first data can include real-time collected data. For example, the first node 101 can be a GAI server, and then the GAI server can determine the local real-time data as the first data. For example, the first node 101 can be different from the GAI server, and then the first node 101 can obtain the first data from the GAI server in real time.

[0201] In some embodiments, the first data can be at least part of the second data, or can be completely different from the second data, and the disclosure embodiments do not make specific limitations on comparison. In an example, the first stem can include at least part of the second stem, or can include other stems different from the second stem. In an example, the second context can include at least part of the third context, or can include at least part different from the third context.

[0202] In step S2104, the first node 101 determines the first sample.

[0203] In some embodiments, after obtaining the first data, the first node 101 can obtain the first sample by the trained GNA based on the obtained first data.

[0204] In some embodiments, the step S2104 can include obtaining the first sample by the generator 301 based on the first data.

[0205] In some embodiments, in a case where the first data includes the first stem, the first stem can be input to the generator 301. The generator 301 can generate the first sample based on the input first stem.

[0206] In some embodiments, in a case where the first data includes the first stem and the second context, the first stem and the second context can be input to the generator 301. The generator 301 can generate the first sample based on the input first stem and the second context.

[0207] In some embodiments, the first stem and / or the second context input to the generator 301 can be raw data. In some embodiments, the first stem and / or the second context input to the generator 301 can be embeddings generated based on the raw data.

[0208] In some embodiments, the process of generating the first sample by the generator 301 can include determining the first context by the context adjuster 3011, and obtaining the first sample by the GAI model 3012 based on the first stem and the first context.

[0209] In some embodiments, in a case where the first data includes the second context, the context adjuster 3011 can determine the first context based on the second context. For example, the context adjuster 3011 can obtain the first context by an adjustment operation based on the second context. In some embodiments, based on one second context, the context adjuster 3011 can obtain one or more fourth contexts.

[0210] In some embodiments, the first context can belong to the second context. In an example, the context adjuster 3011 can select one or more second contexts as the first context for one first stem by an adjustment operation. In some embodiments, the first context can be a new context. In an example, the context adjuster 3011 can generate one or more new contexts as the first context based on one second context by an adjustment operation.

[0211] In some embodiments, if the first data does not include a second context, the context adjuster 3011 can generate a first context autonomously. For example, the context adjuster 3011 can obtain the first context through an adjustment operation. In some embodiments, based on a second context, the context adjuster 3011 can obtain one or more first contexts.

[0212] In some embodiments, after the first stem and the first context are input into the GAI model 3012, the GAI model 3012 can generate a first sample.

[0213] In some embodiments, generator 301 may first generate a feature vector and then determine a first sample based on the feature vector. Thus, the feature vector can be considered as the probability distribution of the first sample. In some embodiments, the feature vector can be used to represent the probability distribution of information obtained for a first stem and a first context. For example, the first stem may be the descriptive text of the question, and the first context may be the contextual information of the descriptive text; then the first sample may be the probability distribution of the descriptive text generated for the descriptive text and its related context. Based on the probability distribution of the descriptive text, generator 301 can obtain the descriptive text. For example, the first stem may be the descriptive image of the question, and the first context may be the contextual information of the descriptive image; then the first sample may be the probability distribution of the descriptive image or descriptive text generated for the descriptive image and its related context. Based on the probability distribution of the descriptive image, generator 301 can obtain the descriptive image.

[0214] In some embodiments, the GAI model 3012 can be deployed in the first node 101. In some embodiments, the first stem and the first context can be input into the local GAI model 3012 by the first node 101 to obtain the first sample.

[0215] In some embodiments, the GAI model 3012 can be deployed on a node different from the first node 101. For example, the GAI model 3012 can be deployed on a GAI server. The GAI server can be used to provide inference services based on the GAI model. In some embodiments, the first node 101 can invoke the GAI model 3012 in the GAI server to obtain a first sample based on a first stem and a first context. For example, the first node 101 can send the first stem and the first context to the GAI server and receive the first sample returned by the GAI server.

[0216] In step S2105, the first node 101 trains a semantic coding model.

[0217] In some embodiments, the first node 101 may train the semantic coding model based on the first sample.

[0218] In some embodiments, the semantic encoding model can comprise: a semantic encoding model for a sender, a semantic encoding model for a receiver.

[0219] In some embodiments, the semantic encoding model for the sender can be used to implement semantic encoding of data. In some embodiments, the semantic encoding model for the sender can be referred to as a semantic encoder.

[0220] In some embodiments, the semantic encoding model for the receiver can be used to implement semantic decoding of data. In some embodiments, the semantic encoding model for the sender can be referred to as a semantic decoder.

[0221] In some embodiments, the semantic encoding model for the sender and the semantic encoding model for the receiver can have symmetry.

[0222] In some embodiments, the training of the semantic encoding model for the sender and the training of the semantic encoding model for the receiver can be performed separately or jointly. In an example, the same first sample can be used to implement the training of the semantic encoding model for the sender and the training of the semantic encoding model for the receiver respectively. In an example, the first sample can be used to train one semantic encoding model. The trained semantic encoding model can be used for both the sender and the receiver.

[0223] In some embodiments, by step S2105, the trained semantic encoding model can be obtained.

[0224] In step S2106, the first node 101 sends the semantic encoding model to the third node 103.

[0225] In some embodiments, the third node 103 can receive the semantic encoding model.

[0226] In some embodiments, the first node 101 can send the semantic encoding model to the third node 103, so as to deploy the semantic encoding model to the third node 103.

[0227] In some embodiments, the third node 103 can be a network node participating in semantic communication. In an example, the third node 103 can be a sender in semantic communication. For example, the first node 101 can send the trained semantic encoding model for the sender to the third node 103. In an example, the third node 103 can be a receiver in semantic communication. For example, the first node 101 can send the trained semantic encoding model for the receiver to the third node 103.

[0228] In some embodiments, the semantic encoding model deployed on the third node 103 can be used for the third node 103 to implement semantic communication. For example, the semantic encoding model for the sender can be used for the third node 103 to send semantic information. For example, the semantic encoding model for the receiver can be used for the third node 103 to receive semantic information.

[0229] In some embodiments, the first node 101 can send the semantic encoding model through at least one of a control plane, a user plane, a data plane, and a computing plane.

[0230] Through steps S2101 to S2106, the semantic communication method of the embodiments of the present disclosure can be implemented.

[0231] The semantic communication method related by the embodiments of the present disclosure can include at least one of steps S2101 to S2106. For example, step S2102 can be implemented as an independent embodiment. For example, step S2104 can be implemented as an independent embodiment. For example, step S2105 can be implemented as an independent embodiment. For example, a combination of steps S2102 and S2104 can be implemented as an independent embodiment. For example, a combination of steps S2104 and S2105 can be implemented as an independent embodiment. For example, a combination of steps S2102, S2104, and S2105 can be implemented as an independent embodiment. It should be noted that the possible independent embodiments composed of one or more of steps S2101 to S2106 are not limited to this.

[0232] In some embodiments, steps S2101, S2102, S2103, S2105, and S2106 are optional, and one or more of these steps can be omitted or replaced in different embodiments.

[0233] In some embodiments, other optional implementation manners described before or after the description of FIG. 2A can be referred to.

[0234] FIG. 2B is an interaction schematic diagram of a semantic communication method according to an embodiment of the present disclosure. The semantic communication method related by the embodiments of the present disclosure can be applied to the communication system 100. As shown in FIG. 2B, the semantic communication method of the embodiments of the present disclosure includes steps S2201 to S2207.

[0235] In step S2201, the second node 102 obtains second data.

[0236] The optional implementation manner of step S2201 can refer to the optional implementation manner of step S2101 of FIG. 2A and other associated parts in the embodiments related by FIG. 2A, which will not be described herein again.

[0237] In some embodiments, the second data can be acquired by the second node 102.

[0238] In some embodiments, the second node 102 can acquire the second data locally. In some embodiments, the second node 102 can acquire the second data from other nodes.

[0239] In step S2202, the second node 102 trains the GAN.

[0240] Optional implementation of step S2202 can refer to optional implementation of step S2102 of FIG. 2A, and other associated parts in the embodiments involved by FIG. 2A, which will not be repeated here.

[0241] In some embodiments, the second node 102 can implement the training of the GAN based on the second data.

[0242] In step S2203, the second node 102 acquires the first data.

[0243] Optional implementation of step S2203 can refer to optional implementation of step S2103 of FIG. 2A, and other associated parts in the embodiments involved by FIG. 2A, which will not be repeated here.

[0244] In some embodiments, the first data can be acquired by the second node 102.

[0245] In some embodiments, the second node 102 can acquire the first data locally. In some embodiments, the second node 102 can acquire the first data from other nodes.

[0246] In step S2204, the second node 102 determines the first sample.

[0247] Optional implementation of step S2204 can refer to optional implementation of step S2104 of FIG. 2A, and other associated parts in the embodiments involved by FIG. 2A, which will not be repeated here.

[0248] In some embodiments, the second node 102 can determine the first sample based on the first data.

[0249] In step S2205, the second node 102 sends the first sample to the first node 101.

[0250] In some embodiments, the first node 101 can receive the first sample.

[0251] In some embodiments, the second node 102 can send the first sample through at least one of a control plane, a user plane, a data plane, and a computing plane.

[0252] In step S2206, the first node 101 trains the semantic encoding model.

[0253] Optional implementation of step S2206 can refer to the optional implementation of step S2105 in FIG. 2A, and other associated parts in the embodiments involved in FIG. 2A, which will not be repeated here.

[0254] In some embodiments, the first node 101 can implement the training of the semantic encoding model based on the received first sample.

[0255] In step S2207, the first node 101 sends the semantic encoding model to the third node 103.

[0256] Optional implementation of step S2207 can refer to the optional implementation of step S2106 in FIG. 2A, and other associated parts in the embodiments involved in FIG. 2A, which will not be repeated here.

[0257] Through steps S2201 to S2207, the semantic communication method of the embodiments of the present disclosure can be implemented.

[0258] The semantic communication method involved in the embodiments of the present disclosure can include at least one of steps S2201 to S2207. For example, step S2202 can be implemented as an independent embodiment. For example, step S2204 can be implemented as an independent embodiment. For example, step S2206 can be implemented as an independent embodiment. For example, the combination of steps S2202 and S2204 can be implemented as an independent embodiment. For example, the combination of steps S2204 and S2206 can be implemented as an independent embodiment. For example, the combination of steps S2202, S2204 and S2206 can be implemented as an independent embodiment. It should be noted that the possible independent embodiments composed of one or more of steps S2201 to S2207 are not limited to this.

[0259] In some embodiments, steps S2201, S2202, S2203, S2205, S2206 and S2207 are optional, and one or more of these steps can be omitted or replaced in different embodiments.

[0260] In some embodiments, steps S2201, S2202, S2203, S2204, S2205 and S2207 are optional, and one or more of these steps can be omitted or replaced in different embodiments.

[0261] In some embodiments, other optional implementations can be recorded before or after the description corresponding to FIG. 2B.

[0262] FIG. 2C is an interaction schematic diagram of a semantic communication method according to an embodiment of the present disclosure. The semantic communication method related by the embodiments of the present disclosure can be applied to the communication system 100. As shown in FIG. 2C, the semantic communication method of the embodiments of the present disclosure includes steps S2301 to S2307.

[0263] In step S2301, the second node 102 obtains second data.

[0264] The optional implementation of step S2301 can refer to the optional implementation of step S2101 of FIG. 2A and other associated parts in the embodiments related by FIG. 2A, which will not be repeated here.

[0265] In some embodiments, the second data can be obtained by the second node 102.

[0266] In some embodiments, the second node 102 can obtain the second data locally. In some embodiments, the second node 102 can obtain the second data from other nodes.

[0267] In step S2302, the second node 102 trains a GAN.

[0268] The optional implementation of step S2302 can refer to the optional implementation of step S2102 of FIG. 2A and other associated parts in the embodiments related by FIG. 2A, which will not be repeated here.

[0269] In some embodiments, the second node 102 can train the GAN based on the second data.

[0270] In step S2303, the second node 102 sends the GAN to the first node 101.

[0271] In some embodiments, the first node 101 can receive the GAN.

[0272] In some embodiments, the second node 102 can send the GAN to the first node 101, so as to deploy the GAN to the first node 101.

[0273] In step S2304, the first node 101 obtains first data.

[0274] The optional implementation of step S2304 can refer to the optional implementation of step S2103 of FIG. 2A and other associated parts in the embodiments related by FIG. 2A, which will not be repeated here.

[0275] In step S2305, the first node 101 determines a first sample.

[0276] The optional implementation of step S2305 can refer to the optional implementation of step S2104 in FIG. 2A and other associated parts in the embodiments involved in FIG. 2A, which will not be repeated here.

[0277] In step S2306, the first node 101 trains the semantic encoding model.

[0278] The optional implementation of step S2306 can refer to the optional implementation of step S2105 in FIG. 2A and other associated parts in the embodiments involved in FIG. 2A, which will not be repeated here.

[0279] In step S2307, the first node 101 sends the semantic encoding model to the third node 103.

[0280] The optional implementation of step S2307 can refer to the optional implementation of step S2106 in FIG. 2A and other associated parts in the embodiments involved in FIG. 2A, which will not be repeated here.

[0281] Through steps S2301 to S2307, the semantic communication method of the embodiments of the present disclosure can be implemented.

[0282] The semantic communication method involved in the embodiments of the present disclosure can include at least one of steps S2301 to S2307. For example, step S2302 can be implemented as an independent embodiment. For example, step S2305 can be implemented as an independent embodiment. For example, step S2306 can be implemented as an independent embodiment. For example, the combination of steps S2302 and S2305 can be implemented as an independent embodiment. For example, the combination of steps S2305 and S2306 can be implemented as an independent embodiment. For example, the combination of steps S2302, S2305 and S2306 can be implemented as an independent embodiment. It should be noted that the possible independent embodiments composed of one or more of steps S2301 to S2307 are not limited to this.

[0283] In some embodiments, steps S2301, S2302, S2303, S2304, S2306 and S2307 are optional, and one or more of these steps can be omitted or replaced in different embodiments.

[0284] In some embodiments, steps S2301, S2302, S2303, S2304, S2305 and S2307 are optional, and one or more of these steps can be omitted or replaced in different embodiments.

[0285] In some embodiments, other optional implementations can be recorded before or after the description corresponding to FIG. 2C.

[0286] In some embodiments, the name of information and the like is not limited to the name described in the embodiments, and the terms of "information", "message", "signal", "signaling", "report", "configuration", "indication", "instruction", "command", "channel", "parameter", "domain", "field", "symbol", "symbol", "codebook", "codeword", "code point", "bit", "data", "program", "chip", and the like can be replaced with each other.

[0287] In some embodiments, the terms of "radio", "wireless", "radio access network (RAN)", "access network (AN)", "RAN-based", and the like can be replaced with each other.

[0288] In some embodiments, the terms of "time", "time point", "time", "time position", and the like can be replaced with each other, and the terms of "time length", "time period", "time window", "window", "time", and the like can be replaced with each other.

[0289] In some embodiments, "acquire", "obtain", "get", "receive", "transmit", "bidirectional transmission", "send and / or receive", and the like can be replaced with each other, which can be interpreted as receiving from other subjects, acquiring from protocols, acquiring from higher layers, obtaining by self-processing, and the like.

[0290] In some embodiments, the terms of "send", "transmit", "report", "issue", "transmit", "bidirectional transmission", "send and / or receive", and the like can be replaced with each other.

[0291] In some embodiments, the terms "certain", "preset", "pre-set", "set", "indicated", "any", "first", and the like can be replaced with each other, "certain A", "preset A", "pre-set A", "set A", "indicated A", "any A", "first A" can be interpreted as A predetermined in a protocol or the like, can be interpreted as A obtained by setting, configuring, or indicating, or the like, can be interpreted as certain A, any A, or first A, and the like, but are not limited thereto.

[0292] In some embodiments, the determination or judgment can be made by a value represented by 1 bit (0 or 1), can be made by a true or false value (Boolean value) represented by true or false, can be made by comparison of numerical values (for example, comparison with a predetermined value), but is not limited thereto.

[0293] FIG. 4A is a flow diagram of a semantic communication method according to an embodiment of the present disclosure. The embodiment of the present disclosure relates to a semantic communication method. The semantic communication method is performed by a first node 101. As shown in FIG. 4A, the method includes steps S4101 to S4106.

[0294] In step S4101, second data is obtained.

[0295] The optional implementation of step S4101 can refer to the optional implementation of step S2101 in FIG. 2A and other associated parts in the embodiments involved in FIG. 2A, which will not be repeated here.

[0296] In step S4102, a GAN is trained.

[0297] The optional implementation of step S4102 can refer to the optional implementation of step S2102 in FIG. 2A and other associated parts in the embodiments involved in FIG. 2A, which will not be repeated here.

[0298] In some embodiments, the training of the GAN can be implemented based on the second data.

[0299] In step S4103, first data is obtained.

[0300] The optional implementation of step S4103 can refer to the optional implementation of step S2103 in FIG. 2A and other associated parts in the embodiments involved in FIG. 2A, which will not be repeated here.

[0301] In step S4104, a first sample is determined.

[0302] The optional implementation of step S4104 can refer to the optional implementation of step S2104 in FIG. 2A and other associated parts in the embodiments related to FIG. 2A, which will not be repeated here.

[0303] In some embodiments, the first sample can be determined based on the first data.

[0304] In step S4105, the semantic encoding model is trained.

[0305] The optional implementation of step S4105 can refer to the optional implementation of step S2105 in FIG. 2A and other associated parts in the embodiments related to FIG. 2A, which will not be repeated here.

[0306] In some embodiments, the training of the semantic encoding model can be implemented based on the first sample.

[0307] In step S4106, the semantic encoding model is sent.

[0308] The optional implementation of step S4106 can refer to the optional implementation of step S2106 in FIG. 2A and other associated parts in the embodiments related to FIG. 2A, which will not be repeated here.

[0309] In some embodiments, the first node 101 can send the semantic encoding model to the third node 103, but is not limited thereto, and can also send the semantic encoding model to other subjects.

[0310] In some embodiments, the semantic encoding model can be used by the third node 103 to implement semantic communication.

[0311] The semantic communication method related to the embodiments of the present disclosure can include at least one of steps S4101 to S4106. For example, step S4102 can be implemented as an independent embodiment. For example, step S4104 can be implemented as an independent embodiment. For example, step S4105 can be implemented as an independent embodiment. For example, a combination of steps S4102 and S4104 can be implemented as an independent embodiment. For example, a combination of steps S4104 and S4105 can be implemented as an independent embodiment. For example, a combination of steps S4102, S4104 and S4105 can be implemented as an independent embodiment. It should be noted that the possible independent embodiments composed of one or more of steps S4101 to S4106 are not limited thereto.

[0312] In some embodiments, steps S4101, S4102, S4103, S4105 and S4106 are optional, and one or more of these steps can be omitted or replaced in different embodiments.

[0313] FIG. 4B is a flow diagram of a semantic communication method according to an embodiment of the present disclosure. The present embodiment relates to a semantic communication method. The semantic communication method is performed by the first node 101. As shown in FIG. 4B, the above method includes steps S4201 to S4203.

[0314] In step S4201, a first sample is obtained.

[0315] The optional implementation of step S4201 can refer to the optional implementation of step S2205 in FIG. 2B and other associated parts in the embodiments involved in FIG. 2B, which will not be repeated here.

[0316] In some embodiments, the first node 101 can receive the first sample sent by the second node 102, but is not limited thereto, and can also receive the first sample sent by other subjects.

[0317] In step S4202, a semantic encoding model is trained.

[0318] The optional implementation of step S4202 can refer to the optional implementation of step S2206 in FIG. 2B and other associated parts in the embodiments involved in FIG. 2B, which will not be repeated here.

[0319] In some embodiments, the training of the semantic encoding model can be implemented based on the first sample.

[0320] In step S4203, the semantic encoding model is sent.

[0321] The optional implementation of step S4203 can refer to the optional implementation of step S2207 in FIG. 2B and other associated parts in the embodiments involved in FIG. 2B, which will not be repeated here.

[0322] In some embodiments, the first node 101 can send the semantic encoding model to the third node 103, but is not limited thereto, and can also send the semantic encoding model to other subjects.

[0323] In some embodiments, the semantic encoding model can be used by the third node 103 to implement semantic communication.

[0324] The semantic communication method related to the present embodiment can include at least one of steps S4201 to S4203. For example, step S4202 can be implemented as an independent embodiment. For example, the combination of steps S4201 and S4202 can be implemented as an independent embodiment. It should be noted that the possible independent embodiments composed of one or more of steps S4201 to S4203 are not limited thereto.

[0325] In some embodiments, steps S4201 and S4203 are optional, and one or more of these steps can be omitted or replaced in different embodiments.

[0326] FIG. 4C is a flow diagram of a semantic communication method according to embodiments of the present disclosure. Embodiments of the present disclosure relate to a semantic communication method. The semantic communication method is performed by a first node 101. As shown in FIG. 4C, the above method includes steps S4301 to S4305.

[0327] In step S4301, a GAN is obtained.

[0328] Optional implementation of step S4301 can refer to optional implementation of step S2303 in FIG. 2C and other associated parts in embodiments related to FIG. 2C, which will not be repeated here.

[0329] In some embodiments, the first node 101 can receive the GAN sent by the second node 102, but is not limited thereto, and can also receive the GAN sent by other subjects.

[0330] In step S4302, first data is obtained.

[0331] Optional implementation of step S4302 can refer to optional implementation of step S2304 in FIG. 2C and other associated parts in embodiments related to FIG. 2C, which will not be repeated here.

[0332] In step S4303, a first sample is determined.

[0333] Optional implementation of step S4303 can refer to optional implementation of step S2305 in FIG. 2C and other associated parts in embodiments related to FIG. 2C, which will not be repeated here.

[0334] In some embodiments, the first sample can be determined based on the first data.

[0335] In step S4304, a semantic encoding model is trained.

[0336] Optional implementation of step S4304 can refer to optional implementation of step S2306 in FIG. 2C and other associated parts in embodiments related to FIG. 2C, which will not be repeated here.

[0337] In some embodiments, the training of the semantic encoding model can be implemented based on the first sample.

[0338] In step S4305, the semantic encoding model is sent.

[0339] The optional implementation of step S4305 can refer to the optional implementation of step S2307 in FIG. 2C, and other associated parts in the embodiments related to FIG. 2B, which are not described herein again.

[0340] In some embodiments, the first node 101 can send the semantic coding model to the third node 103, but is not limited thereto, and can also send the semantic coding model to other subjects.

[0341] In some embodiments, the semantic coding model can be used by the third node 103 to implement semantic communication.

[0342] The semantic communication method related to the embodiments of the present disclosure can include at least one of steps S4301 to S4305. For example, step S4303 can be implemented as an independent embodiment. For example, step S4304 can be implemented as an independent embodiment. For example, a combination of steps S4303 and S4304 can be implemented as an independent embodiment. It should be noted that the possible independent embodiments composed of one or more of steps S4301 to S4305 are not limited thereto.

[0343] In some embodiments, steps S4301, S4302, S4304, and S4305 are optional, and one or more of these steps can be omitted or replaced in different embodiments.

[0344] In some embodiments, steps S4301, S4302, S4303, and S4305 are optional, and one or more of these steps can be omitted or replaced in different embodiments.

[0345] FIG. 5A is a flow diagram of a semantic communication method according to an embodiment of the present disclosure. The embodiments of the present disclosure relate to a semantic communication method. The semantic communication method is performed by the second node 102. As shown in FIG. 5A, the above method includes steps S5101 to S5105.

[0346] In step S5101, second data is obtained.

[0347] The optional implementation of step S5101 can refer to the optional implementation of step S2201 in FIG. 2B, and other associated parts in the embodiments related to FIG. 2B, which are not described herein again.

[0348] In step S5102, a GAN is trained.

[0349] The optional implementation of step S5102 can refer to the optional implementation of step S2202 in FIG. 2B, and other associated parts in the embodiments related to FIG. 2B, which are not described herein again.

[0350] In some embodiments, the training of the GAN can be implemented based on the second data.

[0351] In step S5103, the first data is acquired.

[0352] Optional implementation of step S5103 can refer to the optional implementation of step S2203 in FIG. 2B and other associated parts in the embodiments involved in FIG. 2B, which will not be repeated here.

[0353] In step S5104, the first sample is determined.

[0354] Optional implementation of step S5104 can refer to the optional implementation of step S2204 in FIG. 2B and other associated parts in the embodiments involved in FIG. 2B, which will not be repeated here.

[0355] In some embodiments, the first sample can be determined based on the first data.

[0356] In step S5105, the first sample is sent.

[0357] Optional implementation of step S5105 can refer to the optional implementation of step S2205 in FIG. 2B and other associated parts in the embodiments involved in FIG. 2B, which will not be repeated here.

[0358] In some embodiments, the second node 102 can send the first sample to the first node 101, but is not limited thereto, and can also send the first sample to other subjects.

[0359] The semantic communication method involved in the embodiments of the present disclosure can include at least one of steps S5101 to S5105. For example, step S5102 can be implemented as an independent embodiment. For example, step S5103 can be implemented as an independent embodiment. For example, step S5104 can be implemented as an independent embodiment. For example, the combination of steps S5103 and S5104 can be implemented as an independent embodiment. It should be noted that the possible independent embodiments composed of one or more of steps S5101 to S5105 are not limited thereto.

[0360] In some embodiments, steps S5101, S5103, S5104 and S5105 are optional, and one or more of these steps can be omitted or replaced in different embodiments.

[0361] In some embodiments, steps S5101, S5102, S5104 and S5105 are optional, and one or more of these steps can be omitted or replaced in different embodiments.

[0362] In some embodiments, steps S5101, S5102, S5103, S5105 are optional, and one or more of these steps can be omitted or replaced in different embodiments.

[0363] FIG. 5B is a flow diagram of a semantic communication method according to embodiments of the present disclosure. Embodiments of the present disclosure relate to a semantic communication method. The semantic communication method is performed by the second node 102. As shown in FIG. 5B, the above method includes steps S5201 to S5203.

[0364] In step S5201, the second data is obtained.

[0365] Optional implementation of step S5201 can refer to optional implementation of step S2301 in FIG. 2C and other associated parts in embodiments related to FIG. 2C, which will not be described here.

[0366] In step S5202, the GAN is trained.

[0367] Optional implementation of step S5202 can refer to optional implementation of step S2302 in FIG. 2C and other associated parts in embodiments related to FIG. 2C, which will not be described here.

[0368] In some embodiments, the training of the GAN can be implemented based on the second data.

[0369] In step S5203, the GAN is sent.

[0370] Optional implementation of step S5203 can refer to optional implementation of step S2203 in FIG. 2C and other associated parts in embodiments related to FIG. 2C, which will not be described here.

[0371] In some embodiments, the second node 102 can send the GAN to the first node 101, but is not limited thereto, and can also send the GAN to other subjects.

[0372] The semantic communication method related to embodiments of the present disclosure can include at least one of steps S5201 to S5203. For example, step S5202 can be implemented as an independent embodiment. For example, the combination of steps S5202 and S5203 can be implemented as an independent embodiment. It should be noted that possible independent embodiments composed of one or more of steps S5201 to S5203 are not limited thereto.

[0373] In some embodiments, steps S5201, S5203 are optional, and one or more of these steps can be omitted or replaced in different embodiments.

[0374] FIG. 6A is a flow diagram of a semantic communication method according to an embodiment of the present disclosure. Embodiments of the present disclosure relate to a communication method. As shown in FIG. 6A, the above method comprises step S6101 and step S6102.

[0375] In step S6101, the first node obtains a first sample.

[0376] Optional implementation of step S6101 can be seen in optional implementation of step S2104 of FIG. 2A, step S2205 of FIG. 2B, step S2305 of FIG. 2C, and other associated parts in the embodiments related to FIG. 2A, FIG. 2B, and FIG. 2C, which will not be repeated here.

[0377] In step S6102, the first node trains a semantic encoding model based on the first sample.

[0378] Optional implementation of step S6102 can be seen in optional implementation of step S2105 of FIG. 2A, step S2206 of FIG. 2B, step S2306 of FIG. 2C, and other associated parts in the embodiments related to FIG. 2A, FIG. 2B, and FIG. 2C, which will not be repeated here.

[0379] FIG. 6B is an interaction diagram of a semantic communication method according to an embodiment of the present disclosure. Embodiments of the present disclosure relate to a communication method. As shown in FIG. 6B, the above method comprises step S6201 and step S6202.

[0380] In step S6201, the second node obtains a first sample based on first data through a GAN.

[0381] Optional implementation of step S6201 can be seen in optional implementation of step S2204 of FIG. 2B, and other associated parts in the embodiments related to FIG. 2B, which will not be repeated here.

[0382] In step S6202, the second node sends the first sample to the first node.

[0383] Optional implementation of step S6202 can be seen in optional implementation of step S2205 of FIG. 2B, and other associated parts in the embodiments related to FIG. 2B, which will not be repeated here.

[0384] In the following, the technical solutions of the embodiments of the present disclosure are exemplarily described through specific embodiments.

[0385] In some embodiments, a GAI-assisted semantic communication framework is proposed, and a GAI-based adversarial training auxiliary sample generation strategy is introduced.

[0386] In some embodiments, the GAI-assisted semantic communication framework can be referred to as Gen-SC (generative-sematic communication), which is sufficient to generate samples according to user context information using a GAI model, and then train the transceiver for semantic communication. In addition, in order to guide the GAI model to generate context-related content, a discriminator is added in Gen-SC to measure the difference between the generated samples and the actual samples (i.e., real samples), so as to achieve higher semantic accuracy.

[0387] In some embodiments, the GAI-based end-to-end Gen-SC uses a GAI model to accurately generate user-related contextualized supplementary samples for training of a semantic communication model (SCM).

[0388] In some embodiments, the GAI knowledge enhancement model: by calculating the uncertainty of background knowledge (e.g., context information), the interactive information of the background knowledge can be quantified, and the effectiveness of knowledge enhancement can be measured according to the increment of the background knowledge interactive information.

[0389] In some embodiments, the adversarial training assisted sample generation strategy: a sample generation strategy based on GAI-based question and answer mode is proposed. The generated samples are contextualized by adjusting the context of the question, and a discriminator is designed to evaluate the similarity between the generated answers and the actual answers. This method enables Gen-SC to train the encoder and decoder using contextualized samples, ultimately improving the inference ability of the semantic communication model.

[0390] FIG. 7A is a schematic diagram of a scenario of semantic communication according to an embodiment of the present disclosure. In some embodiments, as shown in FIG. 7A, the end-to-end semantic communication is composed of a pair of transmission nodes (i.e., the third node) and a cloud server (e.g., the first node and / or the second node).

[0391] In some embodiments, the SCM (including the encoder and / or the decoder) is trained on the cloud server. The trained encoder and decoder are deployed on the user end to perform the end-to-end semantic communication task. In the semantic communication model training part, it is assumed that the user interacts with the server that retains the historical context information, and the server generates user-related contextualized training samples. The generated samples are added to the original dataset to train the encoder and the decoder.

[0392] In some embodiments, an adversarial training assisted sample generation strategy is proposed. In this strategy, new data conforming to a specific distribution is generated through adversarial training. In some embodiments, a discriminator is added. By changing the context of the question input to the GAI, the output of the GAI is adjusted to overcome the problem that the GAI network cannot adjust.

[0393] FIG. 7B is a schematic diagram of a GAN-based training scenario, according to an embodiment of the present disclosure. As shown in FIG. 7B, the adversarial training auxiliary sample generation strategy consists of a generator and a discriminator.

[0394] In some embodiments, the generator includes a context adjuster and a GAI model. The context adjuster adjusts the input context of the GAI, and the GAI model is used to generate answers. The discriminator scores the generated answers and the actual answers. The training of the generator and the training of the discriminator are performed alternately to facilitate adversarial training.

[0395] Continuing to refer to FIG. 7B, the question and answer manner in the GAI is explained.

[0396] In some embodiments, each question input into the GAI contains a stem q e Q and a corresponding context h e H, where Q is the set of all stems that can be asked, and H is the set of all contexts. The stem is a description of the question, and the context represents relevant background information related to the question. For a given q and h q , there exists an actual answer a q derived from real scenarios and an answer a q generated by the GAI.

[0397] In the generator, the context adjuster stores the context and the stem of the question. Before inputting the stem into the GAI each time, the context adjuster provides a specific context to the GAI to generate different answers. Considering that we cannot change the parameters of the GAI network in this work, the only way to generate qualified answers is to set an appropriate context for the system.

[0398] To get the optimal context, the discriminator can be used to drive the context adjuster to gradually adjust the context, so as to generate answers close to the actual answers in topic. In an example set, word vectors can be used to represent the probability distribution of all generated answers and all actual answers, and the distance between the actual answers and the generated answers is measured based on the word vectors. Given q and h q , the output answer a q generated by the GAI model M is determined by a probability distribution p(q, a q |h q , M). For the stem q, the probability distribution of the actual answer a from the real sample is M represents the GAI model. Given H and Q, according to the probability distribution of each single generated answer, the distribution P(H, Q) of all generated answers and the distribution of all actual answers can be obtained. When P(H, Q) approaches ​This means that the generated answer's theme and scenario are similar to the actual answer. Therefore, if the generated answer is thematically close to the actual answer, the discriminator will give it a high score; if the generated answer is irrelevant, the discriminator will give it a low score.

[0399] In this way, the training of the generator and the discriminator run alternately. Finally, the context adjuster, driven by the discriminator, takes the most suitable context as input and generates samples with less redundant knowledge to train the encoder and decoder. Therefore, the training objective of a given GAI network can be expressed as:

[0400] in, Let Q(a) be the probability that the network correctly determines the actual answer. q This determines the probability that the network generates an answer that is the actual answer. In one example, Given the probability of the sample label, Q(a) q The probability is given by the output of the discriminator. The goal of the discriminator is to maximize L(G,D) to distinguish the actual answer, while the goal of the generator is to minimize L(G,D) to generate the answer accurately.

[0401] In the embodiments disclosed herein, some or all of the steps and their optional implementations may be arbitrarily combined with some or all of the steps in other embodiments, or may be arbitrarily combined with the optional implementations in other embodiments.

[0402] This disclosure also provides semantic communication apparatuses for implementing any of the above methods. For example, this disclosure provides a semantic communication apparatus including units or modules for implementing the steps performed by a first node in any of the above methods. For example, this disclosure provides a semantic communication apparatus including units or modules for implementing the steps performed by a second node in any of the above methods. For example, this disclosure provides a semantic communication apparatus including units or modules for implementing the steps performed by a third node in any of the above methods.

[0403] It should be understood that the division of each unit or module in the above apparatus is only a logical function division, and all or part of them can be integrated into a physical entity or physically separated in actual implementation. In addition, the units or modules in the apparatus can be implemented in the form of processor calling software: for example, the apparatus includes a processor connected with a memory, the memory stores instructions, and the processor calls the instructions stored in the memory to implement any of the above methods or realize the functions of each unit or module of the above apparatus, wherein the processor is, for example, a general processor such as a central processing unit (CPU) or a microprocessor, and the memory is a memory in the apparatus or a memory outside the apparatus. Alternatively, the units or modules in the apparatus can be implemented in the form of hardware circuit, and the functions of part or all of the units or modules can be realized by the design of hardware circuit. The above hardware circuit can be understood as one or more processors; for example, in one implementation, the above hardware circuit is an application-specific integrated circuit (ASIC), and the functions of part or all of the above units or modules are realized by the design of the logical relationship of elements in the circuit; for example, in another implementation, the above hardware circuit is a programmable logic device (PLD), and a field programmable gate array (FPGA) is taken as an example, which can include a large number of logic gate circuits, and the connection relationship between the logic gate circuits is configured by a configuration file, so as to realize the functions of part or all of the above units or modules. All units or modules of the above apparatus can be implemented in the form of processor calling software, or all units or modules can be implemented in the form of hardware circuit, or part of the units or modules are implemented in the form of processor calling software, and the remaining part is implemented in the form of hardware circuit.

[0404] In the embodiments of the present disclosure, the processor is a circuit with signal processing capability. In one implementation, the processor can be a circuit with instruction reading and running capability, such as a central processing unit, a microprocessor, a graphics processing unit (GPU) (which can be understood as a microprocessor), a digital signal processor (DSP), and the like. In another implementation, the processor can implement certain functions through a logical relationship of a hardware circuit, and the logical relationship of the hardware circuit is fixed or reconfigurable. For example, the processor is a hardware circuit implemented by a special-purpose integrated circuit or a programmable logic device, such as an FPGA. In the reconfigurable hardware circuit, the processor loads a configuration document to implement the configuration of the hardware circuit. It can be understood that the processor loads an instruction to implement the functions of the above part or all units or modules. In addition, the hardware circuit can also be designed for artificial intelligence, which can be understood as an ASIC, such as a neural network processing unit (NPU), a tensor processing unit (TPU), a deep learning processing unit (DPU), and the like.

[0405] FIG. 8 is a structural schematic diagram of a semantic communication apparatus provided by an embodiment of the present disclosure. As shown in FIG. 8, the semantic communication apparatus 800 can include at least one of a transceiver module 801 and a processing module 802.

[0406] In some embodiments, the communication apparatus 800 can be the first node 101. In some embodiments, the transceiver module 801 can be configured to obtain a first sample, where the first sample is obtained by a GAN, and train a semantic encoding model based on the first sample, where the semantic encoding model is used to implement semantic communication. Optionally, the transceiver module 801 can be configured to perform at least one of the communication steps (for example, steps S2106, S2205, S2207, S2303, S2307, but not limited thereto) of the sending and / or receiving performed by the first node 101 in any of the above methods, which will not be described herein. Optionally, the processing module 802 can be configured to perform at least one of the steps (for example, steps S2101, S2102, S2103, S2104, S2105, S2206, S2304, S2305, S2306, but not limited thereto) other than the communication steps of sending and receiving performed by the first node 101 in any of the above methods.

[0407] In some embodiments, the communication apparatus 800 can be the second node 102. In some embodiments, the processing module 802 can be configured to obtain the first sample by the GAN based on the first data; and the transceiver module 801 can be configured to transmit the first sample to the first node. Optionally, the transceiver module 801 can be configured to perform at least one of the communication steps (for example, steps S2205, S2303, but not limited to) of transmitting and / or receiving performed by the second node 102 in any one of the methods, which will not be described herein again. Optionally, the processing module 802 can be configured to perform at least one of the steps (for example, steps S2201, S2202, S2203, S2204, S2301, S2302, but not limited to) other than the communication steps of transmitting and receiving performed by the second node 102 in any one of the methods.

[0408] In some embodiments, the transceiver module can include a transmitting module and / or a receiving module. The transmitting module and the receiving module can be separate or integrated together. Optionally, the transceiver module can be mutually replaced with the transceiver.

[0409] In some embodiments, the processing module can be one module or include multiple sub-modules. Optionally, the multiple sub-modules perform all or part of the steps required to be performed by the processing module. Optionally, the processing module can be mutually replaced with the processor.

[0410] FIG. 9A is a structural schematic diagram of a communication device according to embodiments of the present disclosure. The communication device 9100 can be a network device (for example, an access network device, a core network device, etc.), a terminal (for example, a user equipment, etc.), a chip, a chip system, or a processor supporting the network device to implement any one of the above methods, or a chip, a chip system, or a processor supporting the terminal to implement any one of the above methods. The communication device 9100 can be used to implement the methods described in the above method embodiments, and details can be referred to the descriptions in the above method embodiments.

[0411] As shown in FIG. 9A, the communication device 9100 includes one or more processors 9101. The processor 9101 can be a general-purpose processor or a special-purpose processor, for example, a baseband processor or a central processing unit. The baseband processor can be used to process communication protocols and communication data, and the central processing unit can be used to control the communication device (for example, a base station, a baseband chip, a terminal device, a terminal device chip, a DU or a CU, etc.), execute programs, and process data of the programs. Optionally, the communication device 9100 is used to execute any one of the above methods. Optionally, the one or more processors 9101 are used to call instructions to enable the communication device 9100 to execute any one of the above methods.

[0412] In some embodiments, the communication device 9100 further includes one or more transceivers 9102. When the communication device 9100 includes one or more transceivers 9102, the transceiver 9102 performs at least one of the communication steps (e.g., steps S2106, S2205, S2207, S2303, S2307, but not limited to) in the above-described methods, and the processor 9101 performs at least one of the other steps (e.g., steps S2101, S2102, S2103, S2104, S2105, S2201, S2202, S2203, S2204, S2206, S2301, S2302, S2304, S2305, S2306, but not limited to). In optional embodiments, the transceiver can include a receiver and / or a transmitter, which can be separate or integrated together. Optionally, the terms transceiver, transceiving unit, transceiver, transceiving circuit, interface circuit, interface, etc. can be replaced with each other, and the terms transmitter, transmitting unit, transmitter, transmitting circuit, etc. can be replaced with each other, and the terms receiver, receiving unit, receiver, receiving circuit, etc. can be replaced with each other.

[0413] In some embodiments, the communication device 9100 further includes one or more memories 9103 for storing data. Optionally, all or part of the memory 9103 can also be outside the communication device 9100. In optional embodiments, the communication device 9100 can include one or more interface circuits 9104. Optionally, the interface circuit 9104 is connected to the memory 9103, and the interface circuit 9104 can be used to receive data from the memory 9103 or other devices, and can be used to send data to the memory 9103 or other devices. For example, the interface circuit 9104 can read the data stored in the memory 9103 and send the data to the processor 9101.

[0414] The communication device 9100 described in the above embodiments can be a network device or a terminal, but the scope of the communication device 9100 described in the present disclosure is not limited thereto, and the structure of the communication device 9100 can not be limited by FIG. 9A. The communication device can be a standalone device or can be part of a larger device. For example, the communication device can be: 1) a standalone integrated circuit (IC), or a chip, or a chip system or subsystem; (2) a set of one or more ICs, which can optionally include a storage component for storing data, programs; (3) an ASIC, such as a Modem; (4) a module that can be embedded in other devices; (5) a receiver, a terminal device, a smart terminal device, a cellular phone, a wireless device, a handset, a mobile unit, a vehicle-mounted device, a network device, a cloud device, an artificial intelligence device, etc.; (6) others, etc.

[0415] FIG. 9B is a structural schematic diagram of a chip according to an embodiment of the present disclosure. For the case that the communication device 9100 can be a chip or a chip system, reference can be made to the structural schematic diagram of a chip 9200 shown in FIG. 9B, but the disclosure is not limited thereto.

[0416] The chip 9200 comprises one or more processors 9201. The chip 9200 is configured to perform any of the above methods.

[0417] In some embodiments, the chip 9200 further comprises one or more interface circuits 9202. Optionally, the terms interface circuit, interface, transceiver pin, etc. can replace each other. In some embodiments, the chip 9200 further comprises one or more memories 9203 for storing data. Optionally, all or part of the memory 9203 can be outside the chip 9200. Optionally, the interface circuit 9202 is connected with the memory 9203, the interface circuit 9202 can be configured to receive data from the memory 9203 or other devices, and the interface circuit 9202 can be configured to send data to the memory 9203 or other devices. For example, the interface circuit 9202 can read the data stored in the memory 9203 and send the data to the processor 9201.

[0418] In some embodiments, the interface circuit 9202 performs at least one of the communication steps (for example, steps S2106, S2205, S2207, S2303, S2307, but the disclosure is not limited thereto) of transmitting and / or receiving in the above methods. The interface circuit 9202 performing the communication steps such as transmitting and / or receiving in the above methods means that the interface circuit 9202 performs data interaction between the processor 9201, the chip 9200, the memory 9203 or the transceiver device. In some embodiments, the processor 9201 performs at least one of the other steps (for example, steps S2101, S2102, S2103, S2104, S2105, S2201, S2202, S2203, S2204, S2206, S2301, S2302, S2304, S2305, S2306, but the disclosure is not limited thereto).

[0419] The modules and / or devices described in each of the embodiments of the virtual device, the physical device, the chip, etc. can be combined or separated as appropriate. Optionally, part or all of the steps can also be performed by a plurality of modules and / or devices in cooperation, which is not limited herein.

[0420] The embodiment of the disclosure further provides a storage medium, and instructions are stored on the storage medium. When the instructions are executed on the communication device 9100, the communication device 9100 performs any one of the above methods. Alternatively, the storage medium is an electronic storage medium. Alternatively, the storage medium is a computer readable storage medium, but is not limited to this, and can also be a storage medium readable by other devices. Alternatively, the storage medium can be a non-transitory storage medium, but is not limited to this, and can also be a transitory storage medium.

[0421] The embodiment of the disclosure further provides a program product, and the program product is executed by the communication device 9100, so that the communication device 9100 performs any one of the above methods. Alternatively, the program product is a computer program product.

[0422] The embodiment of the disclosure further provides a computer program, and when the computer program is executed on a computer, the computer performs any one of the above methods.

[0423] Other embodiments of the application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. The specification and examples given are intended as illustrative only and not limiting of the true scope and spirit of the application. Various changes and modifications can be suggested to those skilled in the art, and it is intended that the present application encompass such changes and modifications as fall within the scope of the appended claims. The specification and examples give are intended as illustrative only and not limiting of the true scope and spirit of the application.

[0424] It should be understood that the application is not limited to the precise construction that has been described above and shown in the accompanying drawings, and that various modifications and changes can be effected therein by those skilled in the art without departing from the scope of the application. The scope of the application is to be limited only by the appended claims.

Claims

A method of semantic communication, performed by a first node, wherein The method comprises: obtaining a first sample, wherein the first sample is obtained by a generative adversarial network (GAN); training a semantic coding model based on the first sample, wherein the semantic coding model is used to realize semantic communication. The method of claim 1, wherein, The obtaining of the first sample comprises at least one of: receiving the first sample sent by a second node; obtaining the first sample by the GAN based on first data. The method of claim 2, wherein, The GAN comprises a generator; wherein the obtaining of the first sample by the GAN based on the first data comprises: obtaining the first sample by the generator based on the first data. The method of claim 3, wherein, The generator comprises a context adjuster and a generative artificial intelligence (GAI) model; wherein the obtaining of the first sample by the generator based on the first data comprises: determining a first context by the context adjuster; obtaining the first sample by the GAI model based on a first stem in the first data and the first context. The method of any one of claims 1 to 4, wherein The method further comprises: training the GAN based on second data. The method of claim 5, wherein, The training of the GAN comprises at least one of: training of a generator in the GAN; training of a discriminator in the GAN. The method of claim 6, wherein, The training of the generator comprises: obtaining a second sample by the generator based on the second data; discriminating the second sample by the discriminator; determining a first loss function value of the generator according to a discrimination result; adjusting a parameter value of the generator in a case where the first loss function value meets a first condition. The method according to claim 6 or 7, wherein The training of the discriminator comprises: obtaining a second sample by the generator based on the second data; discriminating the second sample and a third sample in the second data by the discriminator; determining a second loss function value of the discriminator according to a discrimination result; adjusting a parameter value of the discriminator in a case where the second loss function value meets a second condition. The method of any one of claims 6 to 8, wherein, The training of the generator and the training of the discriminator are alternately performed. The method of any one of claims 1 to 9, wherein, The method further comprises: sending the trained semantic coding model to a third node, wherein the trained semantic coding model is used by the third node to realize semantic communication. A method of semantic communication, performed by a second node, wherein, The method comprises: obtaining a first sample by a generative adversarial network (GAN) based on first data; sending the first sample to a first node, wherein the first sample is used to train a semantic coding model, and the semantic coding model is used to realize semantic communication. The method of claim 11, wherein, The GAN comprises a generator; wherein the obtaining of the first sample by the GAN based on the first data comprises: obtaining the first sample by the generator based on the first data. The method of claim 12, wherein, The generator comprises a context adjuster and a generative artificial intelligence (GAI) model; wherein the obtaining of the first sample by the generator based on the first data comprises: determining a first context by the context adjuster; obtaining the first sample by the GAI model based on a first stem in the first data and the first context. The method of any one of claims 11 to 13, wherein, The method further comprises: The GAN is trained based on the second data. The method of claim 14, wherein, The training for the GAN comprises at least one of: the training for a generator in the GAN; the training for a discriminator in the GAN. The method of claim 15, wherein, The training for the generator comprises: obtaining, by the generator, a second sample based on the second data; discriminating, by the discriminator, the second sample; determining a first loss function value of the generator according to a discrimination result; adjusting a parameter value of the generator in a case where the first loss function value satisfies a first condition. The method according to claim 15 or 16, wherein The training for the discriminator comprises: obtaining, by the generator, a second sample based on the second data; discriminating, by the discriminator, the second sample and a third sample in the second data; determining a second loss function value of the discriminator according to a discrimination result; adjusting a parameter value of the discriminator in a case where the second loss function value satisfies a second condition. The method of any one of claims 15 to 17, wherein, The training for the generator and the training for the discriminator are alternately performed. A semantic communication device is provided at a first node, wherein The apparatus comprises: a processing module configured to: obtain a first sample, wherein the first sample is obtained by a generative adversarial network (GAN); train a semantic encoding model based on the first sample, wherein the semantic encoding model is used to implement semantic communication. A semantic communication device is provided at a second node, wherein The apparatus comprises: a processing module configured to obtain a first sample by a generative adversarial network (GAN) based on first data; a transceiver module configured to send the first sample to a first node, wherein the first sample is used to train a semantic encoding model, and the semantic encoding model is used to implement semantic communication. A communication device comprises: one or more processors; a memory storing instructions; wherein the instructions, when executed by the communication device, cause the communication device to implement the semantic communication method according to any one of claims 1 to 10. A communication device comprises: one or more processors; a memory storing instructions; wherein the instructions, when executed by the communication device, cause the communication device to implement the semantic communication method according to any one of claims 11 to 18. A communication system comprises a first node and a second node; wherein, the first node is configured to implement the semantic communication method according to any one of claims 1 to 10, and the second node is configured to implement the semantic communication method according to any one of claims 11 to 18. A storage medium storing instructions, wherein, The instructions, when executed on a communication device, cause the communication device to implement at least one of: the semantic communication method according to any one of claims 1 to 10; the semantic communication method according to any one of claims 11 to 18. A computer program product comprising instructions, wherein the instructions, when executed on a communication device, cause the communication device to implement at least one of: the semantic communication method according to any one of claims 1 to 10; the semantic communication method according to any one of claims 11 to 18.

Images