Model training, information sending, receiving method and device

Abstract

The present disclosure relates to the technical field of communication, in particular to a model training method, an information sending method, a receiving method and apparatuses. The model training method comprises: constructing a training sample set, wherein a sample in the training sample set comprises sample information; training a first initial model and a second initial model based on the training sample set; wherein an input of the first initial model comprises the sample information, an output of the first initial model comprises sample coding features extracted from the sample information, an input of the second initial model comprises the sample coding features, and an output of the second initial model comprises sample reconstruction data obtained by decoding the sample coding features; obtaining a coding model by training the first initial model, and obtaining a decoding model by training the second initial model. According to the present disclosure, the communication resources occupied in the transmission process are relatively less, which is beneficial to saving the resource overhead of the communication between the sending end and the receiving end.

Description

Model training, information sending, receiving method and device TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of communication, and in particular to a model training method, an information sending method, an information receiving method, a model training device, an information sending device, an information receiving device, a communication device and a storage medium. BACKGROUND

[0002] To meet the challenges of communication efficiency and intelligent connectivity, wireless networks need a new communication framework to achieve efficient information exchange and integration of artificial intelligence (AI) communication. However, the current semantic communication framework relies on sending original semantic information, which can improve semantic fidelity, but it is difficult to achieve efficient semantic communication.

[0003] SUMMARY

[0004] Embodiments of the present disclosure provide a model training, information sending, receiving method and device to solve the technical problems in the related art.

[0005] According to a first aspect of embodiments of the present disclosure, a model training method is provided, the method comprising: constructing a training sample set, wherein a sample in the training sample set comprises sample information; training a first initial model and a second initial model based on the training sample set; wherein the input of the first initial model comprises the sample information, the output of the first initial model comprises sample encoding features extracted from the sample information, the input of the second initial model comprises the sample encoding features, and the output of the second initial model comprises sample reconstruction data obtained by decoding the sample encoding features; obtaining an encoding model by training the first initial model, and obtaining a decoding model by training the second initial model.

[0006] According to a second aspect of embodiments of the present disclosure, an information sending method is provided, executed by a sending end, the method comprising: inputting first information to be transmitted into the encoding model of any one of the first aspect and the optional embodiments of the first aspect to obtain encoding features; and sending the encoding features to a receiving end.

[0007] According to a third aspect of embodiments of the present disclosure, an information receiving method is provided, executed by a receiving end, the method comprising: receiving encoding features transmitted by a sending end, wherein the encoding features are output obtained by the sending end inputting first information to be transmitted into the encoding model of any one of the first aspect and the optional embodiments of the first aspect; and inputting the encoding features into the decoding model to obtain first reconstruction data obtained by decoding the encoding features.

[0008] According to a fourth aspect of the embodiments of the present disclosure, a model training apparatus is provided. The apparatus comprises: a sample set construction module configured to construct a training sample set, wherein a sample in the training sample set comprises sample information; a training module configured to train a first initial model and a second initial model based on the training sample set; wherein an input of the first initial model comprises the sample information, an output of the first initial model comprises sample encoding features extracted from the sample information, an input of the second initial model comprises the sample encoding features, and an output of the second initial model comprises sample reconstruction data obtained by decoding the sample encoding features; the first initial model is trained to obtain an encoding model, and the second initial model is trained to obtain a decoding model.

[0009] According to a fifth aspect of the embodiments of the present disclosure, an information sending apparatus is provided. The apparatus is executed by a sending end and comprises: a processing module configured to input first information to be transmitted into an encoding model in any one of the optional embodiments of the first aspect to obtain encoding features; and a sending module configured to send the encoding features to a receiving end.

[0010] According to a sixth aspect of the embodiments of the present disclosure, an information receiving apparatus is provided. The apparatus is executed by a receiving end and comprises: a receiving module configured to receive encoding features transmitted by a sending end, wherein the encoding features are output obtained by the sending end inputting first information to be transmitted into an encoding model in any one of the optional embodiments of the first aspect; and a processing module configured to input the encoding features into the decoding model to obtain first reconstruction data obtained by decoding the encoding features.

[0011] According to a seventh aspect of the embodiments of the present disclosure, a communication device is provided. The communication device comprises one or more processors, and the communication device is configured to execute the method in any one of the first aspect, the second aspect, and the third aspect.

[0012] According to an eighth aspect of the embodiments of the present disclosure, a communication system is provided. The communication system comprises a sending end and a receiving end, wherein the sending end is configured to implement the information sending method in the second aspect, and the receiving end is configured to implement the information receiving method in the third aspect.

[0013] According to a ninth aspect of the embodiments of the present disclosure, a storage medium is provided. The storage medium stores instructions, and when the instructions run on a communication device, the communication device is caused to execute the method in any one of the first aspect, the optional embodiments of the first aspect, the second aspect, the optional embodiments of the second aspect, the third aspect, and the optional embodiments of the third aspect.

[0014] According to a tenth 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 method of any one of the first aspect, the optional implementation of the first aspect, the second aspect, the optional implementation of the second aspect, the third aspect, or the optional implementation of the third aspect.

[0015] According to an embodiment of the present disclosure, when the sending end needs to send the first information to the receiving end, the sending end can not send the original data corresponding to the first information, but encode the original data into the encoded feature through the encoding model, and then send the encoded feature to the receiving end. Since the encoded feature is a low-dimensional feature vector corresponding to the first information, the data amount is smaller than that of the original data corresponding to the first information, and therefore the communication resources occupied in the transmission process are relatively small, which is beneficial to saving the resource overhead of the communication between the sending end and the receiving end. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the drawings needed to be used in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present disclosure, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0017] FIG. 1A is a schematic diagram of an architecture of a communication system according to an embodiment of the present disclosure.

[0018] FIG. 1B is a schematic diagram of communication between a sending end and a receiving end according to an embodiment of the present disclosure.

[0019] FIG. 2 is an interaction schematic diagram of an information sending method according to an embodiment of the present disclosure.

[0020] FIG. 3 is a schematic diagram of task-oriented communication according to an embodiment of the present disclosure.

[0021] FIG. 4 is a schematic diagram of semantic communication according to an embodiment of the present disclosure.

[0022] FIG. 5 is a schematic diagram of a perception scene according to an embodiment of the present disclosure.

[0023] FIG. 6 is a schematic diagram of a communication-perception integrated scene according to an embodiment of the present disclosure.

[0024] FIG. 7 is a schematic diagram of a semantic communication-perception integrated scene according to an embodiment of the present disclosure.

[0025] FIG. 8A is a schematic diagram of a deep neural network according to an embodiment of the present disclosure.

[0026] FIG. 8B is a schematic diagram of another deep neural network, according to an embodiment of the present disclosure.

[0027] FIG. 8C is a schematic diagram of yet another deep neural network, according to an embodiment of the present disclosure.

[0028] FIG. 8D is a schematic diagram of yet another deep neural network, according to an embodiment of the present disclosure.

[0029] FIG. 9 is a schematic flowchart of a model training method, according to an embodiment of the present disclosure.

[0030] FIG. 10 is a schematic flowchart of an information sending method, according to an embodiment of the present disclosure.

[0031] FIG. 11 is a schematic flowchart of an information receiving method, according to an embodiment of the present disclosure.

[0032] FIG. 12 is a schematic block diagram of a model training apparatus, according to an embodiment of the present disclosure.

[0033] FIG. 13 is a schematic block diagram of an information sending apparatus, according to an embodiment of the present disclosure.

[0034] FIG. 14 is a schematic block diagram of an information receiving apparatus, according to an embodiment of the present disclosure.

[0035] FIG. 15A is a schematic diagram of a structure of a communication device, according to an embodiment of the present disclosure.

[0036] FIG. 15B is a schematic diagram of a structure of a chip, according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0037] Embodiments of the present disclosure provide a model training method, an information sending method, an information receiving method, and an apparatus.

[0038] In a first aspect, embodiments of the present disclosure provide a model training method. The method comprises: constructing a training sample set, wherein a sample in the training sample set comprises sample information; training a first initial model and a second initial model based on the training sample set; wherein an input of the first initial model comprises the sample information, an output of the first initial model comprises sample encoding features extracted from the sample information, an input of the second initial model comprises the sample encoding features, and an output of the second initial model comprises sample reconstruction data obtained by decoding the sample encoding features; obtaining an encoding model by training the first initial model, and obtaining a decoding model by training the second initial model.

[0039] In the above embodiments, when the sending end needs to send the first information to the receiving end, the sending end can not send the original data corresponding to the first information, but can encode the original data into the encoded feature through the encoding model, and then send the encoded feature to the receiving end. Since the encoded feature is a low-dimensional feature vector corresponding to the first information, the data amount is smaller than that of the original data corresponding to the first information, and therefore, the communication resources occupied in the transmission process are relatively small, which is beneficial to saving the resource overhead of the communication between the sending end and the receiving end.

[0040] Some embodiments are combined with the first aspect. In some embodiments, the second initial model includes an initial classification sub-model and at least one initial decoding sub-model; wherein the input of the initial classification model includes the sample encoded feature, and the output of the initial classification model includes a sample classification result of the sample encoded feature; the input of the initial decoding sub-model includes the sample encoded feature, and the output of the initial decoding sub-model includes sample reconstruction data obtained by decoding the sample encoded feature; the initial classification sub-model is trained to obtain a classification sub-model, and the initial decoding sub-model is trained to obtain a decoding sub-model.

[0041] Some embodiments are combined with the first aspect. In some embodiments, the loss function for training the first initial model and the second initial model based on the training sample set includes a classification cross-entropy.

[0042] Some embodiments are combined with the first aspect. In some embodiments, the training of the first initial model and the second initial model based on the training sample set includes training the first initial model, the second initial model, and a third initial model based on the training sample set; wherein the input of the third initial model includes sample perception information of the sample encoded feature after being reflected by a sample reflection device, and the output of the third initial model includes a perception result of the sample reflection device, and the third initial model is trained to obtain a perception model.

[0043] Some embodiments are combined with the first aspect. In some embodiments, the first initial model and / or the second initial model includes a deep neural network (DNN).

[0044] In the second aspect, embodiments of the present disclosure provide an information sending method, executed by a sending end, the method comprising: inputting first information to be transmitted into an encoding model in the first aspect or any one of the optional embodiments of the first aspect to obtain an encoded feature; and sending the encoded feature to a receiving end.

[0045] Some embodiments combine the second aspect. In some embodiments, the training the first initial model and the second initial model based on the training sample set comprises: training the first initial model, the second initial model, and a third initial model based on the training sample set; wherein an input of the third initial model comprises sample perceptual information of the sample encoding feature after being reflected by a sample reflection device, and an output of the third initial model comprises a perception result of the sample reflection device, and the third initial model is trained to obtain a perception model; and the information sending method further comprises: receiving perceptual information of the encoding feature after being reflected by a target device; inputting the perceptual information into the perception model to obtain a perception result of the target device.

[0046] In a third aspect, embodiments of the present disclosure provide an information receiving method, executed by a receiving end, comprising: receiving an encoding feature transmitted by a sending end, wherein the encoding feature is output obtained by inputting first information to be transmitted by the sending end into the encoding model of the first aspect or any one of the optional embodiments of the first aspect; and inputting the encoding feature into the decoding model to obtain first reconstruction data obtained by decoding the encoding feature.

[0047] Some embodiments combine the third aspect. In some embodiments, the second initial model comprises an initial classification sub-model and at least one initial decoding sub-model; wherein an input of the initial classification model comprises the sample encoding feature, and an output of the initial classification model comprises a sample classification result of the sample encoding feature; an input of the initial decoding sub-model comprises the sample encoding feature, and an output of the initial decoding sub-model comprises sample reconstruction data obtained by decoding the sample encoding feature; the initial classification sub-model is trained to obtain a classification sub-model, and the initial decoding sub-model is trained to obtain a decoding sub-model; and inputting the encoding feature into the decoding model comprises: inputting the encoding feature into the classification sub-model to obtain a first classification result of the encoding feature; determining a decoding sub-model corresponding to the first classification result from the at least one decoding sub-model, and inputting the encoding feature into the decoding sub-model.

[0048] In a fourth aspect, the embodiments of the present disclosure provide a model training apparatus, the apparatus comprising: a sample set construction module configured to construct a training sample set, wherein a sample in the training sample set comprises sample information; a training module configured to train a first initial model and a second initial model based on the training sample set; wherein an input of the first initial model comprises the sample information, an output of the first initial model comprises sample encoding features extracted from the sample information, an input of the second initial model comprises the sample encoding features, and an output of the second initial model comprises sample reconstruction data obtained by decoding the sample encoding features; the first initial model is trained to obtain an encoding model, and the second initial model is trained to obtain a decoding model.

[0049] In a fifth aspect, the embodiments of the present disclosure provide an information sending apparatus executed by a sending end, the apparatus comprising: a processing module configured to input first information to be transmitted into the encoding model of any one of the first aspect and the optional embodiments of the first aspect to obtain encoding features; and a sending module configured to send the encoding features to a receiving end.

[0050] In a sixth aspect, the embodiments of the present disclosure provide an information receiving apparatus executed by a receiving end, the apparatus comprising: a receiving module configured to receive encoding features transmitted by a sending end, wherein the encoding features are output obtained by the sending end inputting first information to be transmitted into the encoding model of any one of the first aspect and the optional embodiments of the first aspect; and a processing module configured to input the encoding features into the decoding model to obtain first reconstruction data obtained by decoding the encoding features.

[0051] In a seventh aspect, the embodiments of the present disclosure provide a communication device, comprising: one or more processors; wherein the communication device is configured to execute the method of any one of the first aspect, the optional embodiments of the first aspect, the second aspect, the optional embodiments of the second aspect, the third aspect, and the optional embodiments of the third aspect.

[0052] In an eighth aspect, the embodiments of the present disclosure provide a communication system comprising a sending end and a receiving end, wherein the sending end is configured to implement the information sending method of any one of the second aspect and the optional embodiments of the second aspect, and the receiving end is configured to implement the information receiving method of any one of the third aspect and the optional embodiments of the third aspect.

[0053] In a ninth aspect, the embodiments of the present disclosure provide a storage medium, which stores instructions, when the instructions are executed on a communication device, cause the communication device to perform the method in any one of the first aspect, the optional implementation of the first aspect, the second aspect, the optional implementation of the second aspect, the third aspect, or the optional implementation of the third aspect.

[0054] In a tenth aspect, the embodiments of the present disclosure provide a program product, when the program product is executed by a communication device, cause the communication device to perform the method in any one of the first aspect, the optional implementation of the first aspect, the second aspect, the optional implementation of the second aspect, the third aspect, or the optional implementation of the third aspect.

[0055] In an eleventh aspect, the embodiments of the present disclosure provide a computer program, when the computer program is executed on a computer, cause the computer to perform the method in any one of the first aspect, the optional implementation of the first aspect, the second aspect, the optional implementation of the second aspect, the third aspect, or the optional implementation of the third aspect.

[0056] It can be understood that the above model training, information sending, receiving device, communication device, communication system, storage medium, program product, and computer program are used to perform the method proposed in the embodiments of the present disclosure. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding method, which will not be described here.

[0057] The embodiments of the present disclosure propose a model training, information sending, and receiving method and device. In some embodiments, the terms of model training, information sending, and receiving method, information processing method, and communication method can be replaced with each other, the terms of model training, information sending, and receiving device, information processing device, and communication device can be replaced with each other, and the terms of information processing system and communication system can be replaced with each other.

[0058] The 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 in an embodiment can be combined arbitrarily; in addition, the embodiments can be combined arbitrarily, for example, the steps of different embodiments or part of the steps of different embodiments can be combined arbitrarily, an embodiment can be combined with the optional implementation of other embodiments arbitrarily.

[0059] 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.

[0060] 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.

[0061] In the embodiments of the present disclosure, unless otherwise specified and logically conflicted, the elements expressed in singular form, such as "one", "a", "the", "above", "said", "preceding", "this" and the like, can represent "one and only one", or "one or more", "at least one" and the like.

[0062] For example, in the case of using articles such as "a", "an", "the" and the like in translation, the noun after the article can be understood as a singular expression, or can be understood as a plural expression.

[0063] In the embodiments of the present disclosure, "plurality" means two or more.

[0064] In some embodiments, the terms "at least one of", "one or more", "a plurality of", "multiple" and the like can be replaced with each other.

[0065] In some embodiments, the description manner such as "at least one of A, B", "A and / or B", "A in one case and B in another case", "A in response to one case and B in response to another case" and the like can include the following technical solutions according to the case: A in some embodiments (A is executed regardless of B); B in some embodiments (B is executed regardless of A); A and B are selectively executed in some embodiments (A and B are selected from A and B); A and B are executed in some embodiments (A and B are executed). When there are more branches such as A, B, C and the like, it is similar to the above.

[0066] In some embodiments, the description manner such as "A or B" and the like can include the following technical solutions according to the case: A in some embodiments (A is executed regardless of B); B in some embodiments (B is executed regardless of A); A and B are selectively executed in some embodiments (A and B are selected from A and B). When there are more branches such as A, B, C and the like, it is similar to the above.

[0067] The prefix words "first", "second", etc. 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 of the description objects should refer to the description in the context of the claims or embodiments, and should not constitute redundant limitation because of the use of the prefix words.

[0068] 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". "First" and "second" do not limit whether the "fields" modified thereby are in the same message, nor do they 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, 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 "first information" and "second information" can be the same information or different information, and the content thereof can be the same or different.

[0069] In some embodiments, "including A", "containing A", "for indicating A", "carrying A" can be interpreted as directly carrying A, or indirectly indicating A.

[0070] In some embodiments, the terms "in response to", "in response to determining", "in the case of", "when", "when", "if", "if" and the like can be replaced with each other.

[0071] In some embodiments, the terms "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 "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.

[0072] 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, and 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.

[0073] 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.

[0074] 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.

[0075] 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.

[0076] 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.

[0077] 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.

[0078] In some embodiments, obtaining data, information, etc. can comply with laws and regulations of the country where the location is.

[0079] In some embodiments, data, information, etc. can be obtained after obtaining user consent.

[0080] 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.

[0081] FIG. 1A is a schematic diagram of an architecture of a communication system according to an embodiment of the present disclosure.

[0082] As shown in FIG. 1A, the communication system 100 includes a sending end 101 and a receiving end 102, for example, the sending end and the receiving end can include a terminal, a network device, etc. communication equipment, wherein the network device includes at least one of the following: an access network device, a core network device.

[0083] In some embodiments, the terminal includes at least one of the following: a mobile phone, a wearable device, an Internet of Things device, a car with communication function, a smart car, a Pad, a computer with wireless transceiver function, 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 smart grid, a wireless terminal device in transportation safety, a wireless terminal device in smart city, a wireless terminal device in smart home, but not limited to this.

[0084] In some embodiments, the access network device is, for example, a node or device that accesses a terminal to a wireless network, and 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.

[0085] In some embodiments, the core network device can be one device including one or more network elements, or can be multiple devices or device groups including all or part of the one or more network elements described above. The network element can be virtual or physical. The core network includes, for example, at least one of an evolved packet core (EPC), a 5G core network (5GCN), and a next generation core (NGC).

[0086] 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 realized through software or programs.

[0087] 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, and some of the protocol layers are controlled by the CU, and the remaining or all of the protocol layers are distributed in the DU and controlled by the CU, but is not limited thereto.

[0088] It can be understood that the communication system described in the embodiments of the present disclosure is for more clearly illustrating the technical solutions of the embodiments of the present disclosure, and does not constitute a limitation on the technical solutions proposed by the embodiments of the present disclosure. Those skilled in the art can know that, with the evolution of system architecture and the appearance of new business scenarios, the technical solutions proposed by the embodiments of the present disclosure are also applicable to similar technical problems.

[0089] The following embodiments of the present disclosure can be applied to the communication system 100 shown in FIG. 1A or part of the subject, but are not limited thereto. The subjects shown in FIG. 1A are exemplary, and the communication system can include all or part of the subjects in FIG. 1A, or other subjects other than FIG. 1A. The number and form of each subject is arbitrary, each subject can be physical or virtual, the connection relationship between each subject is exemplary, each subject can not be connected or can be connected, the connection can be in any way, can be direct connection or indirect connection, can be wired connection or wireless connection.

[0090] Embodiments of the present disclosure can be applied to Long Term Evolution (LTE), LTE-Advanced (LTE-A), LTE-Beyond (LTE-B), SUPER 3G, IMT-Advanced, 4th generation mobile communication system (4G), 5th generation mobile communication system (5G), 5G new radio (NR), Future Radio Access (FRA), New-Radio Access Technology (RAT), New Radio (NR), New radio access (NX), Future generation radio access (FX), Global System for Mobile communications (GSM (registered trademark)), CDMA2000, Ultra Mobile Broadband (UMB), IEEE 802.11 (Wi-Fi (registered trademark)), IEEE 802.16 (WiMAX (registered trademark)), IEEE 802.20, Ultra-WideBand (UWB), Bluetooth (Bluetooth (registered trademark)), Public Land Mobile Network (PLMN) network, Device-to-Device (D2D) system, Machine to Machine (M2M) system, Internet of Things (IoT) system, Vehicle-to-Everything (V2X), system using other communication methods, next-generation system expanded based on them, or the like. In addition, a plurality of systems can be combined (for example, combination of LTE or LTE-A and 5G, or the like).

[0091] In some embodiments, in a scenario of communication between a sending end and a receiving end, the sending end can directly send original data (such as a picture, voice, text, video, etc.) to the receiving end. The sending end can send the original data to the receiving end through a channel.

[0092] FIG. 1B is a schematic diagram of communication between a sending end and a receiving end according to an embodiment of the present disclosure.

[0093] As shown in FIG. 1B, for the input original data, the sending end can first perform source coding processing to compress the original data into smaller data packets for transmission.

[0094] Further, the sending end can perform channel coding on the source coded data, which can add some redundant information to improve the anti-interference capability. Then, the channel coded data is modulated by a modulator to convert the channel coded data into a signal form suitable for channel transmission, and then transmitted to the receiving end through the channel. The receiving end can perform source decoding, channel decoding, demodulation and the like, and the operation sequence is not limited by the present disclosure, so as to obtain the reconstructed data.

[0095] In the above embodiment, since the source coding is performed on all original data, the amount of coded data is still large, and further, the amount of data obtained after channel coding, modulator modulation and the like is also relatively large, and the occupation of communication resources is relatively large during transmission to the receiving end.

[0096] FIG. 2 is an interaction schematic diagram of an information sending method according to an embodiment of the present disclosure.

[0097] As shown in FIG. 2, the model training method can include the following steps:

[0098] In step S201, the sending end inputs the first information to be transmitted into an encoding model to obtain an encoding feature. For example, the encoding feature can include semantic features of the first information.

[0099] In some embodiments, the first information includes at least one of the following: a picture, a voice, a text, and a video.

[0100] It should be noted that the first information is not limited to these types of information, and can also include other information that can be sent by the sending end, and the present disclosure does not limit this.

[0101] In some embodiments, the encoding model can be trained based on machine learning, deep learning and the like on the basis of a training sample set. The training method of the encoding model is described in subsequent embodiments.

[0102] In step S202, the sending end sends the encoding feature to the receiving end.

[0103] In some embodiments, the receiving end receives the encoding feature.

[0104] In some embodiments, the sending end can be a terminal or a network device, and the receiving end can be a terminal or a network device. For example, one of the sending end and the receiving end is a terminal, and the other is a network device; or, both the sending end and the receiving end are terminals; or, both the sending end and the receiving end are network devices.

[0105] In some embodiments, after receiving the encoded feature, the receiving end can input the encoded feature into a decoding model. The decoding model can decode the input encoded feature, and the decoded data can be used as the reconstruction of the first information, for example, can be referred to as first reconstructed data, which is the same as or similar to the first information.

[0106] In some embodiments, the decoding model can be trained based on machine learning and deep learning on the basis of the training sample set. The training and determination of the decoding model are described in subsequent embodiments.

[0107] According to embodiments of the present disclosure, when the sending end needs to send the first information to the receiving end, the sending end can not send the original data corresponding to the first information, but can encode the original data into an encoded feature through an encoding model and then send the encoded feature to the receiving end. Since the encoded feature is a low-dimensional feature vector corresponding to the first information, the data amount is smaller than that of the original data corresponding to the first information, and thus the communication resources occupied in the transmission process are relatively small, which is beneficial to saving the resource overhead of the communication between the sending end and the receiving end.

[0108] Moreover, although the sending end in the embodiments of the present disclosure does not transmit the original data corresponding to the first information, for example, does not process and transmit the original data of the first information based on the embodiment shown in FIG. 1B, but transmits the encoded feature corresponding to the first information, the receiving end can decode the encoded feature through a decoding model to obtain the first reconstructed data corresponding to the first information, thereby realizing the end-to-end (sending end to receiving end) transmission of the first information.

[0109] The training processes of the encoding model and the decoding model are exemplarily described below through several embodiments.

[0110] In some embodiments, the present disclosure provides a model training method, which includes:

[0111] constructing a training sample set, wherein the samples in the training sample set include sample information;

[0112] training a first initial model and a second initial model based on the training sample set;

[0113] The input of the first initial model includes sample information, the output of the first initial model includes sample encoding features extracted from the sample information, the input of the second initial model includes the sample encoding features, and the output of the second initial model includes sample reconstruction data obtained by decoding the sample encoding features; the first initial model is trained to obtain an encoding model, and the second initial model is trained to obtain a decoding model.

[0114] In some embodiments, the first initial model and / or the second initial model includes a deep neural network (DNN).

[0115] In the embodiment, the DNN can generate and transmit a low-dimensional feature vector. In the embodiment, the encoding model can be an encoder, the decoding model can be a decoder, and the encoder-decoder pair can be modeled as an end-to-end DNN. For example, the first initial model can be a DNN, and the second initial model can also be a DNN.

[0116] For example, when the first initial model extracts semantic features of the sample information as the sample encoding features, the encoding model obtained by training can also extract semantic features from the encoding features. In this case, the scenario in which the sender transmits the encoding features to the receiver can be a semantic communication (SemCom) scenario.

[0117] For example, in the SemCom scenario, the sender extracts first feature encoding from the first information and then transmits the encoding features to the receiver in the communication process (for example, the process can also include a subsequent step of obtaining the first reconstruction data by the receiver). The process can be referred to as task-oriented communication (TOC).

[0118] In the communication process, the sender and the receiver use the models (for example, the encoding model and the decoding model) trained based on the DNN to transmit semantic information. In the process of training the models, the encoding model of the sender (for example, which can be an encoder) and the decoding model of the receiver (for example, which can be a decoder) can be trained to perform source encoding, source decoding, channel encoding, channel decoding, modulation, demodulation, and the like in cooperation without the need to transmit and reconstruct all data of the sample to the receiver.

[0119] In some embodiments, the above communication process can also complete a specific task, for example, a classification task for the sample (described in subsequent embodiments).

[0120] In addition, in the semantic communication scenario, the receiving end also needs to complete a semantic reservation task, for example, judging whether the information (for example, the sample reconstruction data) of the receiving end reserves the semantic of the information (for example, the sample information) of the sending end. If the sample reconstruction data has the same meaning as the sample information, it can be determined that the information of the receiving end reserves the semantic of the information of the sending end, and the sending end correctly conveys the semantic to the receiving end.

[0121] In the semantic communication scenario, in addition to the task of judging whether the information of the receiving end reserves the semantic of the information of the sending end, the receiving end also involves a data reconstruction task, for example, determining the corresponding sample reconstruction data for the sample information. For these two tasks, the encoding model on the sending end and the decoding model on the receiving end can be jointly trained to perform together, for example, the machine learning method involved in the training process can include multi-task learning (MTL), which combines the two tasks of data reconstruction and semantic reservation.

[0122] In some embodiments, the second initial model includes an initial classification sub-model and at least one initial decoding sub-model.

[0123] The input of the initial classification model includes the sample encoding feature, and the output of the initial classification model includes the sample classification result of the sample encoding feature. The input of the initial decoding sub-model includes the sample encoding feature, and the output of the initial decoding sub-model includes the sample reconstruction data obtained by decoding the sample encoding feature. The initial classification sub-model is trained to obtain a classification sub-model, and the initial decoding sub-model is trained to obtain a decoding sub-model.

[0124] In some embodiments, the decoding model deployed on the receiving end can not only perform data reconstruction (for example, determining the first reconstruction data corresponding to the first information), but also perform classification, for example, determining the type corresponding to the information (for example, the encoding feature) received from the sending end. Taking the first information including a picture as an example, the type of classification can be the type of object in the picture, for example, car, person, airplane, animal, building, etc.

[0125] In some embodiments, in order to realize data reconstruction and classification on the receiving end, two models can be deployed on the receiving end, for example, the encoding model can include two sub-models, one is a classification sub-model for classification, and the other is a decoding sub-model for data reconstruction.

[0126] In this case, in order to train the classification sub-model and the decoding sub-model, for the second initial model that needs to be trained in the training stage, the second initial model can be set to include an initial classification sub-model and at least one initial decoding sub-model. In the training process, the sample encoding feature can be used as the input of the initial decoding sub-model and the input of the initial decoding sub-model, and the two sub-models can be jointly trained.

[0127] Unlike the conventional receiver chain focusing on data reconstruction, the task-oriented receiving device in the embodiment directly performs the task of classification, such as classifying the received signal (e.g., encoded features), using a dedicated decoder (e.g., a trained classification submodel). Since the input of the classification submodel is the encoded features, rather than the original data corresponding to the first information, the process of reconstructing the sample data (e.g., determining the sample reconstruction data) can be avoided in the classification process, and correspondingly, the process of reconstructing the sample data can also be avoided in the training process.

[0128] In some embodiments, the loss function for training the first initial model and the second initial model based on the training sample set includes a classification cross-entropy.

[0129] In the embodiment, the encoding model and the decoding model can be used as an encoder and a decoder, respectively, and the encoder-decoder pair can be modeled as an end-to-end DNN. For example, the first initial model can be a DNN, and the second initial model can also be a DNN. The DNN is trained to optimize the task performance by considering the channel and the data features. The success of optimization is quantified by machine learning metrics, such as the loss function in the classification task. For example, the classification cross-entropy can be used as the loss function for training.

[0130] It should be noted that the loss function of the present disclosure is not limited to the classification cross-entropy. For example, a loss function considering semantics can also be considered. The construction of the specific loss function is not limited by the present disclosure.

[0131] FIG. 3 is a schematic diagram illustrating a task-oriented communication according to an embodiment of the present disclosure.

[0132] As shown in FIG. 3, in the task-oriented communication (TOC), an encoder (e.g., trained based on a DNN) is deployed at the sending end, and a decoder (e.g., trained based on a DNN) is deployed at the receiving end.

[0133] The encoder can be used to obtain the encoded features of the input information and generate a waveform based on the encoded features. For example, the waveform generation can include converting the encoded features into a waveform signal for transmission. In turn, the encoder can transmit the encoded features in the form of a waveform signal to the receiving end through the channel.

[0134] For example, the TOC can complete a classification task, and a decoder deployed at the receiving end can serve as a task classifier, where the task can include the received encoded feature. The classification of the encoded feature by the encoder can include, for example, determining a task label of the encoded feature, where the task label represents the classification result (e.g., a type). Taking a picture as an example of the first information, the type of the classification can refer to the type of an object in the picture, such as a car, a person, an airplane, an animal, a building, and the like. As shown in FIG. 3, the classification result can be an automobile.

[0135] FIG. 4 is a schematic diagram of semantic communication according to an embodiment of the present disclosure.

[0136] According to an embodiment of the present disclosure, as shown in FIG. 4, a decoding model deployed at the receiving end can include two sub-models, a classification sub-model and a decoding sub-model, where the classification sub-model can serve as a decoder for task classification, for example, a task classifier, and the decoding sub-model can serve as a decoder for data reconstruction.

[0137] In some embodiments, a classification sub-model and a decoding sub-model can be trained, and the decoding sub-model can perform data reconstruction on the encoded feature as input without distinguishing the classification result of the encoded feature.

[0138] In some embodiments, a classification sub-model and multiple decoding sub-models can be trained, where the decoding sub-models can correspond to the classification result (e.g., a determined type) of the classification sub-model, and the decoding sub-models can be dedicated to data reconstruction of the encoded feature corresponding to the classification result.

[0139] For example, in an application process, the sending end can obtain the encoded feature of the first information through the encoding model, and then send the encoded feature to the receiving end through a channel.

[0140] The receiving end can input the encoded feature into the classification sub-model and the first decoding sub-model, respectively. The encoded feature can be input into the classification sub-model to obtain a first classification result of the encoded feature. In at least one decoding sub-model, the decoding sub-model corresponding to the first classification result is determined, and the encoded feature is input into the decoding sub-model.

[0141] For example, decoding sub-model #A is dedicated to data reconstruction of encoded features with a classification result of a car, decoding sub-model #B is dedicated to data reconstruction of encoded features with a classification result of a person, and decoding sub-model #C is dedicated to data reconstruction of encoded features with a classification result of a building. In the case of a first classification result of a car, decoding sub-model #A corresponding to the determination of the first classification result can be selected, so that the encoded features are input into decoding sub-model #A, and data reconstruction is completed through decoding sub-model #A.

[0142] According to the present embodiment, a corresponding decoding sub-model can be selected for data reconstruction based on the classification result. Since the decoding sub-model is dedicated to data reconstruction of a specific type of information, the model parameters thereof are more applicable to the specific type of information than a decoding sub-model for data reconstruction of a non-specific type of information, which is conducive to more accurate data reconstruction (e.g., the reconstruction result is closer to the semantics of the first information).

[0143] In the scenario of semantic communication, in addition to data reconstruction, semantic communication can also be used as a paradigm conversion, which is conducive to preserving the inherent meaning of the information transmitted to the receiving end. The training loss of the encoding model (e.g., a sub-encoder) can include a semantic loss in addition to a data reconstruction loss, which facilitates the preservation of meaning in the information transmission process based on the semantic loss, and ensures that the transmitted information retains its intended meaning.

[0144] The encoder (e.g., the above-mentioned encoding model) is deployed at the sending end to process source coding, channel coding, and modulation. Two decoders can be deployed at the receiving end, e.g., decoder #A is the above-mentioned classification sub-model, and decoder #B is the above-mentioned decoding sub-model. Decoder #B can implement data reconstruction by performing joint demodulation, channel decoding, and source decoding; decoder #A can perform task classification and also evaluate semantic information preservation, effectively performing machine learning tasks such as image classification.

[0145] The above-mentioned encoding model, classification sub-model, and decoding sub-model can be determined by jointly training three DNNs (one encoder and two decoders) in MTL, combining reconstruction loss with semantic loss. This process captures the latent feature space, promotes reliable transmission of compressed feature vectors with minimal channel usage, while maintaining a low semantic loss.

[0146] For example, when the above-mentioned encoding model, classification sub-model, and decoding sub-model are used in a semantic communication scenario, the goal of accidental semantic communication is to maintain the integrity of meaning in different forms of information, and seamlessly integrate with mobile communication networks (e.g., 5G, 6G, and subsequent versions of the network), which will significantly benefit from semantic preservation, enrich user experience, and enable more context-aware communication.

[0147] In some embodiments, the first initial model and the second initial model are trained based on the training sample set, including:

[0148] The first initial model, the second initial model, and the third initial model are trained based on the training sample set.

[0149] The input of the third initial model includes sample perception information after the sample encoding feature is reflected by the sample reflection device, and the output of the third initial model includes a perception result of the sample reflection device. The perception model is obtained after the third initial model is trained.

[0150] In some embodiments, the information sending method further includes:

[0151] Receiving perception information after the encoding feature is reflected by the target device;

[0152] Inputting the perception information into the perception model to obtain a perception result of the target device.

[0153] In this embodiment, sensing can be taken as a task in semantic communication MTL, so that the tasks in MTL are extended from saving semantic information and data reconstruction to environmental sensing, such as detecting potential target devices. On this basis, the communication between the sending end and the receiving end can include Integrated Sensing and Communications (ISAC). ISAC can utilize the dual use (communication and sensing) of communication signals for sensing, eliminating the need for additional probe signals, which is beneficial for saving bandwidth, reducing delay, and improving energy efficiency.

[0154] For example, in a vehicle to everything (V2X) network, the dual functions of communication and sensing in ISAC can enable vehicles to exchange critical information about their positions and intentions, while sensing and monitoring potential dangers around them, identifying changing road conditions, and detecting other vehicles, pedestrians, and objects such as traffic signs.

[0155] FIG. 5 is a schematic diagram of a sensing scenario according to an embodiment of the present disclosure.

[0156] In some embodiments, as shown in FIG. 5, the sending end can send the input signal in the form of a sensing waveform through the channel. After the signal is shot at the target device, it is reflected by the target device and received by the sending end or other nodes. The sending end or other nodes can be deployed with sensing algorithms. Through the sensing algorithms, the received reflected signal can be calculated to determine the perception result, such as the position, angle, speed, etc. of the target device.

[0157] FIG. 6 is a schematic diagram illustrating a communication and perception integrated scenario according to an embodiment of the present disclosure.

[0158] In some embodiments, based on the embodiment shown in FIG. 5, the sending end can also send the signal to the receiving end through the channel, and the receiving end can perform source decoding, channel decoding, demodulation, etc. on the received signal to obtain the reconstructed data.

[0159] FIG. 7 is a schematic diagram illustrating a semantic communication and perception integrated scenario according to an embodiment of the present disclosure.

[0160] In some embodiments, as shown in FIG. 7, based on the trained encoding model, classification submodel, and decoding submodel, a perception model can also be trained and obtained, which can be deployed in a node (such as the sending end or other devices) capable of receiving the reflected signal of the target device. For the sake of example, the present disclosure illustrates the deployment of the perception model in the sending end.

[0161] For example, in the training process, in addition to training the first initial model and the second initial model (which can include the initial classification submodel and the initial decoding submodel), the third initial model can also be trained. Through training, the encoding model, the classification submodel, the decoding submodel, and the perception model can be obtained.

[0162] For example, the encoding model and the perception model can be deployed in the sending end, and the classification submodel and the decoding submodel can be deployed in the receiving end.

[0163] The sending end can input the first information into the encoding model to obtain the encoding feature, and then send the encoding feature to the target device through the channel on the one hand, and send the encoding feature to the receiving end through the channel on the other hand.

[0164] The target device can reflect the signal containing the first encoding feature to the sending end or other nodes. Taking the reflection to the sending end as an example, the sending end can input the received reflected first encoding feature into the perception model, and the perception model can output the perception result of the target device, such as the position, angle, speed, etc. of the target device.

[0165] The receiving end can input the first encoding feature into the classification submodel and the decoding submodel respectively, the classification submodel can output the classification result of the first encoding feature, such as the label, and the decoding submodel can output the data reconstruction result of the first encoding feature.

[0166] To optimize the overall performance, the DNN (e.g., encoding model) for information transmission and sensing shown in FIG. 7 can be considered to drive the communication waveform design. With the encoder (e.g., encoding model), the transmitter performs joint operations including source coding, channel coding, and modulation to produce a signal (e.g., containing encoded features). The receiver employs another DNN as a decoder (e.g., decoding model) to perform joint operations such as demodulation, channel decoding, source decoding, etc., to achieve data reconstruction.

[0167] Meanwhile, the signal transmitted by the transmitter has a dual role, on the one hand, it is used for communication with the receiver, and on the other hand, it has the ability to sense. In the presence of a target device, the signal is reflected from the target device and received by the transmitter or other nodes, there can be two possible sensing cases. For these cases, a decoder (e.g., sensing model) can be deployed at the transmitter or other nodes for sensing. The task of this decoder includes detecting the presence of the target and determining its distance.

[0168] Embodiments of the present disclosure consider all involved data, channels, and potential target features through MTL (e.g., including classification tasks, data reconstruction tasks, sensing tasks, semantic preservation tasks), and employ DNNs for joint training. In MTL, two encoders (e.g., classification sub-model, decoding sub-model) can be deployed at the receiver, one encoder operates as a task classifier, and the other classifier is used for data reconstruction. Among them, the task classifier can also be used to evaluate the fidelity of task completion (e.g., semantic preservation task), such as image classification based on the received signal, which enhances the combination of semantics and ISAC.

[0169] FIG. 8A is a schematic diagram of a deep neural network according to an embodiment of the present disclosure.

[0170] In some embodiments, the DNN as an encoder (e.g., encoding model) can be as shown in FIG. 8A, including convolutional layers (Conv2D), max-pooling layers (MaxPool), flattening layers (Flatten), fully connected layers (Dense), linear layers (Linear), dropout layers (Dropout), etc. The dimension of the input vector can be 32x32x3, and the dimension of the output vector can be 1x n c The activation function includes but is not limited to ReLU (Rectified Linear Unit), and the last fully connected layer can be denoted as a linear layer.

[0171] wherein n c The specific value of n is not limited by the present disclosure, and the number of each layer can be one or more, and the number and connection relationship of the layers can not be limited to the case shown in FIG. 8A.

[0172] For example, the encoding model is trained based on the first initial model, which can be a DNN. During the training process, the number and size of layers in the DNN can be adjusted according to the loss function. For example, the loss function includes the classification cross-entropy. Based on the loss function, the classification accuracy can be determined, and the number and size of layers in the DNN can be adjusted until the classification accuracy cannot be further improved.

[0173] FIG. 8B is a schematic diagram of another deep neural network according to an embodiment of the present disclosure.

[0174] In some embodiments, the DNN as a decoder (e.g., a classification submodel) can be as shown in FIG. 8B, including a fully connected layer (Dense). The dimension of the input vector can be 1 x n c , and the dimension of the output vector can be 1 x 10. The activation function includes but is not limited to ReLU and Softmax.

[0175] wherein n c is not limited by the present disclosure, and the number of fully connected layers can not be limited to the case shown in FIG. 8B.

[0176] FIG. 8C is a schematic diagram of another deep neural network according to an embodiment of the present disclosure.

[0177] In some embodiments, the DNN as a decoder (e.g., a decoding submodel) can be as shown in FIG. 8C, including a fully connected layer (Dense). The dimension of the input vector can be 1 x n c , and the dimension of the output vector can be 32 x 32 x 3. The activation function includes but is not limited to ReLU, and the last fully connected layer can be denoted as a linear layer.

[0178] wherein n c is not limited by the present disclosure, and the number of fully connected layers can not be limited to the case shown in FIG. 8C.

[0179] FIG. 8D is a schematic diagram of another deep neural network according to an embodiment of the present disclosure.

[0180] In some embodiments, the DNN as a decoder (e.g., a perception model) can be as shown in FIG. 8D, including a fully connected layer (Dense). The dimension of the input vector can be 1 x n c , and the dimension of the output vector can be 1 x 2. The activation function includes but is not limited to ReLU and Softmax.

[0181] wherein n c is not limited by the present disclosure, and the number of fully connected layers can not be limited to the case shown in FIG. 8B.

[0182] The communication method related to the embodiments of the present disclosure can include at least one of steps S201-S202. For example, step S201 can be implemented as an independent embodiment, step S202 can be implemented as an independent embodiment, steps S201+S202 can be implemented as an independent embodiment, but are not limited thereto.

[0183] In some embodiments, steps S201 and S202 can be exchanged in order or executed simultaneously.

[0184] In some embodiments, step S201 is optional, and one or more of the steps can be omitted or replaced in different embodiments.

[0185] In some embodiments, step S202 is optional, and one or more of the steps can be omitted or replaced in different embodiments.

[0186] In some embodiments, other optional implementations described before or after the description corresponding to FIG. 2 can be referred to.

[0187] In a first aspect, the embodiments of the present disclosure propose a model training method. FIG. 9 is a schematic flowchart of a model training method according to an embodiment of the present disclosure. The model training method shown in the present embodiment can be executed by a terminal.

[0188] As shown in FIG. 9, the model training method can include the following steps:

[0189] In step S901, a training sample set is constructed, wherein the samples in the training sample set include sample information.

[0190] In step S902, the first initial model and the second initial model are trained based on the training sample set; wherein the input of the first initial model includes the sample information, the output of the first initial model includes the sample encoding features extracted from the sample information, the input of the second initial model includes the sample encoding features, and the output of the second initial model includes the sample reconstruction data obtained by decoding the sample encoding features; the first initial model is trained to obtain an encoding model, and the second initial model is trained to obtain a decoding model.

[0191] It should be noted that the embodiment shown in FIG. 9 can be independently implemented, or can be implemented in combination with at least one other embodiment of the present disclosure. The specific implementation can be selected as needed, and the present disclosure is not limited.

[0192] In some embodiments, the second initial model comprises an initial classification sub-model and at least one initial decoding sub-model; wherein the input of the initial classification sub-model comprises the sample encoded feature, and the output of the initial classification sub-model comprises a sample classification result of the sample encoded feature; the input of the initial decoding sub-model comprises the sample encoded feature, and the output of the initial decoding sub-model comprises sample reconstruction data obtained by decoding the sample encoded feature; the initial classification sub-model is trained to obtain a classification sub-model, and the initial decoding sub-model is trained to obtain a decoding sub-model.

[0193] In some embodiments, the loss function for training the first initial model and the second initial model based on the training sample set comprises a classification cross-entropy.

[0194] In some embodiments, the training of the first initial model and the second initial model based on the training sample set comprises: training the first initial model, the second initial model, and a third initial model based on the training sample set; wherein the input of the third initial model comprises sample perception information after the sample encoded feature is reflected by a sample reflection device, and the output of the third initial model comprises a perception result of the sample reflection device; and the third initial model is trained to obtain a perception model.

[0195] In some embodiments, the first initial model and / or the second initial model comprises a deep neural network (DNN).

[0196] It should be noted that the embodiment shown in FIG. 9 can be independently implemented, or can be combined with at least one other embodiment of the present disclosure for implementation. The specific implementation can be selected as needed, and the present disclosure does not limit it.

[0197] The optional implementation of the first aspect and the optional implementation of the optional embodiment of the first aspect can refer to the optional implementation of the embodiment shown in FIG. 2 and other related parts in the embodiment related to FIG. 2, which will not be repeated here.

[0198] In the second aspect, the embodiments of the present disclosure provide an information sending method. FIG. 10 is a schematic flowchart of an information sending method according to an embodiment of the present disclosure. The information sending method shown in the present embodiment can be executed by a network device.

[0199] As shown in FIG. 10, the information sending method can comprise the following steps:

[0200] In step S1001, input the first information to be transmitted into the encoding model of any of the foregoing embodiments to obtain an encoded feature;

[0201] In step S1002, the encoded feature is sent to a receiving end.

[0202] It should be noted that the embodiment shown in FIG. 10 can be independently implemented, or can be implemented in combination with at least one other embodiment of the present disclosure. The specific implementation can be selected as needed, and the present disclosure does not limit.

[0203] In some embodiments, the training of the first initial model and the second initial model based on the training sample set comprises: training the first initial model, the second initial model, and a third initial model based on the training sample set; wherein the input of the third initial model comprises sample perception information after the sample encoding feature is reflected by a sample reflection device, and the output of the third initial model comprises a perception result of the sample reflection device; and the third initial model is trained to obtain a perception model.

[0204] The information sending method further comprises: receiving perception information after the encoding feature is reflected by the target device; inputting the perception information into the perception model to obtain a perception result of the target device.

[0205] The optional implementation of the second aspect and the optional implementation of the optional embodiment of the second aspect can refer to the optional implementation of the embodiment shown in FIG. 2 and other related parts of the embodiment involved in FIG. 2, which will not be repeated here.

[0206] In a third aspect, embodiments of the present disclosure provide an information receiving method. FIG. 11 is a schematic flowchart of an information receiving method according to an embodiment of the present disclosure. The information receiving method shown in the present embodiment can be executed by a network device.

[0207] As shown in FIG. 11, the information receiving method can include the following steps:

[0208] In step S1101, the encoding feature transmitted by the sending end is received, wherein the encoding feature is output obtained by inputting the first information to be transmitted into the encoding model of any of the foregoing embodiments by the sending end;

[0209] In step S1102, the encoding feature is input into the decoding model to obtain the first reconstruction data obtained by decoding the encoding feature.

[0210] It should be noted that the embodiment shown in FIG. 11 can be independently implemented, or can be implemented in combination with at least one other embodiment of the present disclosure. The specific implementation can be selected as needed, and the present disclosure does not limit.

[0211] In some embodiments, the second initial model comprises an initial classification submodel and at least one initial decoding submodel; wherein an input of the initial classification model comprises the sample encoded feature, and an output of the initial classification model comprises a sample classification result of the sample encoded feature; an input of the initial decoding submodel comprises the sample encoded feature, and an output of the initial decoding submodel comprises sample reconstruction data obtained by decoding the sample encoded feature; the initial classification submodel is trained to obtain a classification submodel, and the initial decoding submodel is trained to obtain a decoding submodel.

[0212] The encoded feature is input into the decoding model, comprising: inputting the encoded feature into the classification submodel to obtain a first classification result of the encoded feature; determining a decoding submodel corresponding to the first classification result in the at least one decoding submodel, and inputting the encoded feature into the decoding submodel.

[0213] The third aspect and the optional implementation of the optional embodiment of the third aspect can refer to the optional implementation of the embodiment shown in FIG. 2 and other related parts of the embodiment involved in FIG. 2, which will not be repeated here.

[0214] In some embodiments, the name of information and the like is not limited to the name recorded in the embodiment, and the terms such as "information", "message", "signal", "signaling", "report", "configuration", "indication", "instruction", "command", "channel", "parameter", "domain", "field", "symbol", "codebook", "codeword", "codepoint", "bit", "data", "program", "chip", and the like can be replaced with each other.

[0215] In some embodiments, "obtaining", "getting", "getting", "receiving", "transmitting", "bidirectional transmission", "sending and / or receiving" can be replaced with each other, which can be interpreted as receiving from other subjects, obtaining from protocols, obtaining from high layers, processing to obtain, independently implementing, and other meanings.

[0216] In some embodiments, the terms such as "sending", "transmitting", "reporting", "issuing", "transmitting", "bidirectional transmission", "sending and / or receiving" can be replaced with each other.

[0217] Corresponding to the embodiments of the model training method, the information sending method, and the information receiving method described above, the disclosure also provides embodiments of a model training device, an information sending device, and an information receiving device.

[0218] FIG. 12 is a schematic block diagram of a model training device according to an embodiment of the disclosure. As shown in FIG. 12, the model training device includes a sample set construction module 1201 and a training module 1202.

[0219] In some embodiments, the sample set construction module is configured to construct a training sample set, wherein a sample in the training sample set includes sample information.

[0220] The training module is configured to train a first initial model and a second initial model based on the training sample set; wherein an input of the first initial model includes the sample information, an output of the first initial model includes sample encoding features extracted from the sample information, an input of the second initial model includes the sample encoding features, and an output of the second initial model includes sample reconstruction data obtained by decoding the sample encoding features; an encoding model is obtained after training the first initial model, and a decoding model is obtained after training the second initial model.

[0221] In some embodiments, the second initial model includes an initial classification sub-model and at least one initial decoding sub-model; wherein an input of the initial classification sub-model includes the sample encoding features, and an output of the initial classification sub-model includes a sample classification result of the sample encoding features; an input of the initial decoding sub-model includes the sample encoding features, and an output of the initial decoding sub-model includes sample reconstruction data obtained by decoding the sample encoding features; a classification sub-model is obtained after training the initial classification sub-model, and a decoding sub-model is obtained after training the initial decoding sub-model.

[0222] In some embodiments, a loss function for training the first initial model and the second initial model based on the training sample set includes a classification cross-entropy.

[0223] In some embodiments, the training module is configured to train the first initial model, the second initial model, and a third initial model based on the training sample set; wherein an input of the third initial model includes sample perception information obtained by reflecting the sample encoding features by a sample reflection device, and an output of the third initial model includes a perception result of the sample reflection device; and a perception model is obtained after training the third initial model.

[0224] In some embodiments, the first initial model and / or the second initial model includes a deep neural network (DNN).

[0225] FIG. 13 is a schematic block diagram of an information sending device according to an embodiment of the present disclosure. As shown in FIG. 13, the information sending device includes a processing module 1301, a sending module 1302, and a receiving module 1303.

[0226] In some embodiments, the processing module is configured to input the first information to be transmitted into the encoding model in any of the above embodiments to obtain an encoded feature.

[0227] The sending module is configured to send the encoded feature to a receiving end.

[0228] In some embodiments, the processing module is configured to train the first initial model, the second initial model, and a third initial model based on the training sample set.

[0229] The input of the third initial model includes sample perceptual information after the sample encoded feature is reflected by a sample reflection device, and the output of the third initial model includes a perception result of the sample reflection device. The third initial model is trained to obtain a perception model.

[0230] The receiving module is configured to receive perceptual information after the encoded feature is reflected by a target device.

[0231] The processing module is further configured to input the perceptual information into the perception model to obtain a perception result of the target device.

[0232] FIG. 14 is a schematic block diagram of an information receiving device according to an embodiment of the present disclosure. As shown in FIG. 14, the information receiving device includes a receiving module 1401 and a processing module 1402.

[0233] In some embodiments, the receiving module is configured to receive an encoded feature transmitted by a sending end, where the encoded feature is output of the sending end after inputting first information to be transmitted into the encoding model in any of the above embodiments.

[0234] The processing module is configured to input the encoded feature into the decoding model to obtain first reconstruction data after decoding the encoded feature.

[0235] In some embodiments, the second initial model comprises an initial classification sub-model and at least one initial decoding sub-model; wherein an input of the initial classification model comprises the sample encoding feature, and an output of the initial classification model comprises a sample classification result of the sample encoding feature; an input of the initial decoding sub-model comprises the sample encoding feature, and an output of the initial decoding sub-model comprises sample reconstruction data obtained by decoding the sample encoding feature; a classification sub-model is obtained after training the initial classification sub-model, and a decoding sub-model is obtained after training the initial decoding sub-model;

[0236] The processing module is configured to input the encoding feature into the classification sub-model to obtain a first classification result of the encoding feature, and determine a decoding sub-model corresponding to the first classification result in the at least one decoding sub-model and input the encoding feature into the decoding sub-model.

[0237] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts are described in the part of the method embodiments. The device embodiments described above are merely illustrative, wherein the modules described as separate components can or can not be physically separated, and the components displayed as modules can or can not be physical modules, i.e., they can be located in one place or distributed on multiple network modules. According to actual needs, some or all of the modules can be selected to achieve the purpose of the present embodiment. Those skilled in the art can understand and implement without creative labor.

[0238] The embodiments of the present disclosure also propose a device for implementing any of the above methods, for example, a device comprising units or modules for implementing each step performed by a terminal in any of the above methods. For another example, another device is proposed, comprising units or modules for implementing each step performed by a network device (such as an access network device, a core network function node, a core network device, etc.) in any of the above methods.

[0239] 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, the processor is connected with a memory, the memory stores instructions, and the processor calls the instructions stored in the memory to realize any of the above methods or realize the functions of each unit or module of the above apparatus, wherein the processor is 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 units or modules are realized by the design of the logical relationship of elements in the circuit; for another 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 all implemented in the form of processor calling software, or all implemented in the form of hardware circuit, or part implemented in the form of processor calling software and the remaining part implemented in the form of hardware circuit.

[0240] 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 (CPU), 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 can be reconfigured. For example, the processor is a hardware circuit implemented by an application-specific integrated circuit (ASIC) or a programmable logic device (PLD), 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 instructions to implement the functions of the above part or all units or modules. In addition, it can also be a hardware circuit 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.

[0241] FIG. 15A is a structural schematic diagram of a communication device 15100 according to an embodiment of the present disclosure. The communication device 15100 can be a network device (for example, an access network device, a core network device, and the like), can be a terminal (for example, a user equipment, and the like), can be a chip, a chip system, or a processor supporting the network device to implement any of the above methods, and can also be a chip, a chip system, or a processor supporting the terminal to implement any of the above methods. The communication device 15100 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.

[0242] As shown in FIG. 15A, the communication device 15100 includes one or more processors 15101. The processor 15101 can be a general processor or a special-purpose processor, etc., for example, a baseband processor or a central processing unit. The baseband processor can be used to process communication protocols and communication data, the central processing unit can be used to control a communication apparatus (e.g., 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 15100 is configured to perform any of the above methods. Optionally, the one or more processors 15101 are configured to invoke instructions to cause the communication device 15100 to perform any of the above methods.

[0243] In some embodiments, the communication device 15100 further includes one or more transceivers 15102. When the communication device 15100 includes one or more transceivers 15102, the transceiver 15102 performs at least one of the communication steps (e.g., steps S201, S202, but not limited to) in the above methods, and the processor 15101 performs at least one of the other steps (e.g., steps S201, S202, 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 by each other, and the terms transmitter, transmitting unit, transmitter, transmitting circuit, etc. can be replaced by each other, and the terms receiver, receiving unit, receiver, receiving circuit, etc. can be replaced by each other.

[0244] In some embodiments, the communication device 15100 further includes one or more memories 15103 for storing data. Optionally, all or part of the memory 15103 can also be outside the communication device 15100. In optional embodiments, the communication device 15100 can include one or more interface circuits 15104. Optionally, the interface circuit 15104 is connected to the memory 15102, and the interface circuit 15104 can be used to receive data from the memory 15102 or other devices, and can be used to send data to the memory 15102 or other devices. For example, the interface circuit 15104 can read data stored in the memory 15102 and send the data to the processor 15101.

[0245] The communication device 15100 described in the above embodiments can be a network device or a terminal, but the scope of the communication device 15100 described in the present disclosure is not limited thereto, and the structure of the communication device 15100 can not be limited by FIG. 15A. 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 also include storage components 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, and the like; (6) other devices, and the like.

[0246] FIG. 15B is a structural diagram of a chip 15200 according to an embodiment of the present disclosure. For the case where the communication device 15100 is a chip or a chip system, the structural diagram of the chip 15200 shown in FIG. 15B can be referred to, but is not limited thereto.

[0247] The chip 15200 includes one or more processors 15201. The chip 15200 is configured to perform any of the above methods.

[0248] In some embodiments, the chip 15200 further includes one or more interface circuits 15202. Optionally, the terms interface circuit, interface, transceiver pin, and the like can be replaced with each other. In some embodiments, the chip 15200 further includes one or more memories 15203 for storing data. Optionally, all or part of the memory 15203 can be outside the chip 15200. Optionally, the interface circuit 15202 is connected to the memory 15203, and the interface circuit 15202 can be configured to receive data from the memory 15203 or other devices, and the interface circuit 15202 can be configured to send data to the memory 15203 or other devices. For example, the interface circuit 15202 can read data stored in the memory 15203 and send the data to the processor 15201.

[0249] In some embodiments, the interface circuit 15202 performs at least one of the communication steps (such as steps S201, S202, but not limited thereto) of the above methods, such as transmitting and / or receiving. The interface circuit 15202 performing the communication steps such as transmitting and / or receiving in the above methods means that the interface circuit 15202 performs data interaction between the processor 15201, the chip 15200, the memory 15203, or a transceiver device. In some embodiments, the processor 15201 performs at least one of the other steps (such as steps S201, S202, but not limited thereto).

[0250] The modules and / or devices described in various embodiments of the virtual device, the physical device, the chip, etc. can be combined or separated according to actual conditions. Alternatively, part or all of the steps can also be executed by a plurality of modules and / or devices in cooperation, which is not limited here.

[0251] The disclosure also provides a storage medium, and the storage medium stores instructions. When the instructions run on the communication device 15100, the communication device 15100 executes 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. It 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. It can also be a transitory storage medium.

[0252] The disclosure also provides a program product, and the program product is executed by the communication device 15100, so that the communication device 15100 executes any one of the above methods. Alternatively, the program product is a computer program product.

[0253] The disclosure also provides a computer program, which makes the computer execute any one of the above methods when it runs on the computer.

Claims

1. A model training method, characterized in that, The method comprises: constructing a training sample set, wherein the samples in the training sample set comprise sample information; training a first initial model and a second initial model based on the training sample set; wherein the input of the first initial model comprises the sample information, the output of the first initial model comprises sample encoding features extracted from the sample information, the input of the second initial model comprises the sample encoding features, and the output of the second initial model comprises sample reconstruction data decoded from the sample encoding features; the first initial model is trained to obtain an encoding model, and the second initial model is trained to obtain a decoding model.

2. The method of claim 1, wherein, The second initial model comprises an initial classification sub-model and at least one initial decoding sub-model; wherein the input of the initial classification sub-model comprises the sample encoding features, the output of the initial classification sub-model comprises a sample classification result of the sample encoding features, the input of the initial decoding sub-model comprises the sample encoding features, and the output of the initial decoding sub-model comprises sample reconstruction data decoded from the sample encoding features; the initial classification sub-model is trained to obtain a classification sub-model, and the initial decoding sub-model is trained to obtain a decoding sub-model.

3. The method of claim 2, wherein, The loss function for training the first initial model and the second initial model based on the training sample set comprises a classification cross-entropy.

4. The method according to any one of claims 1 to 3, characterized in that, The training of the first initial model, the second initial model, and a third initial model based on the training sample set comprises: training the first initial model, the second initial model, and the third initial model based on the training sample set; wherein the input of the third initial model comprises sample perception information of the sample encoding features after being reflected by a sample reflection device, the output of the third initial model comprises a perception result of the sample reflection device, and the third initial model is trained to obtain a perception model.

5. The method according to any one of claims 1 to 4, characterized in that, The first initial model and / or the second initial model comprise a deep neural network (DNN).

6. An information transmission method characterized by comprising: The method is performed by a sending end and comprises: inputting first information to be transmitted into the encoding model of any one of claims 1 to 5 to obtain encoding features; sending the encoding features to a receiving end.

7. The method of claim 6, wherein, The training of the first initial model and the second initial model based on the training sample set comprises: training the first initial model, the second initial model, and the third initial model based on the training sample set; wherein the input of the third initial model comprises sample perception information of the sample encoding features after being reflected by a sample reflection device, the output of the third initial model comprises a perception result of the sample reflection device, and the third initial model is trained to obtain a perception model. The information sending method further comprises: receiving perception information of the encoding features after being reflected by a target device; inputting the perception information into the perception model to obtain a perception result of the target device.

8. An information receiving method characterized by comprising: The method is performed by a receiving end and comprises: receiving encoding features transmitted by a sending end, wherein the encoding features are output of the sending end after inputting first information to be transmitted into the encoding model of any one of claims 1 to 5; input the encoding feature into the decoding model to obtain first reconstruction data obtained by decoding the encoding feature.

9. The method of claim 8, wherein, The second initial model comprises an initial classification sub-model and at least one initial decoding sub-model. The input of the initial classification model comprises the sample encoding feature, and the output of the initial classification model comprises a sample classification result of the sample encoding feature; the input of the initial decoding sub-model comprises the sample encoding feature, and the output of the initial decoding sub-model comprises sample reconstruction data obtained by decoding the sample encoding feature; the initial classification sub-model is trained to obtain a classification sub-model, and the initial decoding sub-model is trained to obtain a decoding sub-model. The inputting of the encoding feature into the decoding model comprises: inputting the encoding feature into the classification sub-model to obtain a first classification result of the encoding feature; determining a decoding sub-model corresponding to the first classification result in at least one decoding sub-model, and inputting the encoding feature into the decoding sub-model.

10. A model training apparatus, comprising: The apparatus comprises: a sample set construction module configured to construct a training sample set, wherein a sample in the training sample set comprises sample information; a training module configured to train a first initial model and a second initial model based on the training sample set; The input of the first initial model comprises the sample information, the output of the first initial model comprises sample encoding feature extracted from the sample information, the input of the second initial model comprises the sample encoding feature, and the output of the second initial model comprises sample reconstruction data obtained by decoding the sample encoding feature; the first initial model is trained to obtain an encoding model, and the second initial model is trained to obtain a decoding model.

11. An information transmitting apparatus characterized by comprising: The apparatus comprises: a processing module configured to input first information to be transmitted into the encoding model in any one of claims 1 to 5 to obtain an encoding feature; a sending module configured to send the encoding feature to a receiving end.

12. An information receiving apparatus characterized by comprising: The apparatus comprises: a receiving module configured to receive the encoding feature transmitted by a sending end, wherein the encoding feature is output obtained by inputting first information to be transmitted by the sending end into the encoding model in any one of claims 1 to 5; a processing module configured to input the encoding feature into the decoding model to obtain first reconstruction data obtained by decoding the encoding feature.

13. A communication device, characterized by comprise: one or more processors; The communication device is configured to perform the method in any one of claims 1 to 9.

14. A communication system, characterized by The apparatus comprises a sending end and a receiving end, wherein the sending end is configured to implement the information sending method in any one of claims 6 to 7, and the receiving end is configured to implement the information receiving method in any one of claims 8 to 9.

15. A storage medium, the storage medium storing instructions, wherein, When the instructions run on the communication device, the communication device is caused to perform the method in any one of claims 1 to 9.

16. A program product, characterized by The program product is executed by the communication device, and the communication device is caused to perform the method in any one of claims 1 to 9.

Images