Data communication method and communication equipment

Abstract

The invention discloses a data communication method and a communication device, and relates to the technical field of artificial intelligence and communication, the disclosed communication device comprises a first communication device and a second communication device, the first communication device comprises a first transceiver and a first processor, the first processor is coupled to the first transceiver, and the second transceiver is coupled to the second processor. Wherein the first processor is configured to: determine a gating parameter matrix and coding and decoding parameters required for transmitting data; transmitting the text encoded data over the communication channel; the text coding data is generated by performing semantic and part-of-speech-based coding processing on a target text to be sent by a semantic communication task on the basis of a first gating matrix and coding parameters in the coding and decoding parameters; the gating parameter matrix comprises the first gating matrix used for auxiliary coding and the second gating matrix used for auxiliary decoding.

Description

Technical Field

[0001] The present application relates to the fields of artificial intelligence and communication technologies, and particularly to a data communication method and a communication device. Background Art

[0002] Future 6G (6th Generation Mobile Communication Technology) will integrate artificial intelligence with traditional communication to achieve the vision of extending from the real world to the virtual world. With the continuous enhancement of the transmission capacity of communication systems, the system complexity gradually increases, and traditional communication can no longer meet the growing business needs. Intelligent and simplified communication is established on the basis of traditional communication. Among them, the nodes in the network will become new nodes with intelligent functions, and the protocol structure of the network itself will tend to be extremely simple. This way promotes the further evolution of the sixth-generation mobile communication system towards a wireless communication network with endogenous wisdom and native simplicity.

[0003] Semantic communication is gradually becoming one of the key technologies of the sixth-generation mobile communication technology. The goal of semantic communication is to ensure that the receiver accurately extracts and interprets the meaning of information, emphasizing the content and significance of information. Semantic communication can transmit only key information and ignore redundant parts to compress the amount of transmitted data, thereby saving bandwidth resources. Among them, natural language processing based on deep learning has promoted further research on text semantic communication. Text semantic communication aims to maximize the system capacity and reduce semantic errors rather than bit or symbol errors in traditional communication by restoring the meaning of sentences in the text.

[0004] However, the text semantic communication in the known technologies still has the defects of poor restoration quality of the transmitted text and high semantic errors. Summary of the Invention

[0005] For this reason, the present application discloses the following technical solutions:

[0006] A first communication device, comprising:

[0007] A first transceiver; and

[0008] A first processor coupled to the first transceiver, wherein the first processor is configured to:

[0009] Determine a gating parameter matrix and encoding and decoding parameters required for transmitting data;

[0010] Transmit text encoded data on a communication channel;

[0011] Wherein, the text encoding data is generated by performing encoding processing based on semantics and part of speech on the target text to be sent in a semantic communication task according to the first gating matrix and the encoding parameters in the encoding and decoding parameters; the gating parameter matrix includes the first gating matrix for assisting encoding and the second gating matrix for assisting decoding.

[0012] Optionally, after determining the gating parameter matrix and the encoding and decoding parameters required for transmitting data, the first processor is further configured to:

[0013] Send the gating parameter matrix and the encoding and decoding parameters to a second communication device;

[0014] Or, receive the gating parameter matrix and the encoding and decoding parameters sent by the second communication device;

[0015] Wherein, the gating parameter matrix and the encoding and decoding parameters are respectively provided by the gating model and the encoding and decoding model generated by the first communication device through joint training, or are respectively provided by the gating model and the encoding and decoding model generated by the second communication device through joint training.

[0016] Optionally, the first processor is further configured to:

[0017] Determine the semantic features and part-of-speech features of each word in the target text, as well as the semantic features of the context information corresponding to each word;

[0018] Map the semantic features and part-of-speech features of each word to the same feature space based on the first gating matrix to obtain a mapping result;

[0019] Perform feature fusion processing on the mapping result corresponding to each word and the semantic features of the corresponding context information based on the attention mechanism to obtain the semantic encoding data of the target text;

[0020] Perform channel encoding on the semantic encoding data to obtain the text encoding data.

[0021] Optionally, after mapping the semantic features and part-of-speech features of each word to the same feature space based on the first gating matrix to obtain a mapping result, the first processor is further configured to:

[0022] Map the semantic vector and the part-of-speech vector of each word to the same feature space through a projection matrix to obtain first attention information; the first attention information includes a first query, a first key, and a first key value obtained by mapping the semantic vector and the part-of-speech vector to the same feature space;

[0023] The first processor performs feature fusion processing on the semantic features of the mapping result corresponding to each word and the corresponding context information based on an attention mechanism to obtain semantic encoding data of the target text, and is further configured to:

[0024] Based on a first attention mechanism and corresponding encoding parameters, fuse the first attention information of each word to obtain a fusion feature of each word;

[0025] Based on a second attention mechanism and corresponding encoding parameters, fuse the fusion feature of each word and the semantic features of the corresponding context information to obtain semantic encoding data of each word, so as to obtain semantic encoding data of the target text based on the semantic encoding data of each word.

[0026] Optionally, the first processor is further configured to:

[0027] Report first capability information to a second communication device;

[0028] Or, receive second capability information reported by the second communication device;

[0029] Wherein, the first capability information indicates that the first communication device supports semantic encoding and decoding based on a gating mechanism; the second capability information indicates that the second communication device supports semantic encoding and decoding based on a gating mechanism.

[0030] A second communication device, comprising:

[0031] A second transceiver; and

[0032] A second processor, which is coupled to the second transceiver, wherein the second processor is configured to:

[0033] Determine a gating parameter matrix and encoding and decoding parameters required for transmitting data;

[0034] Receive text encoding data on a communication channel; the text encoding data is generated by a first communication device through encoding processing based on semantics and part of speech for a target text to be transmitted in a semantic communication task;

[0035] Perform decoding processing based on a gating mechanism on the text encoding data through a second gating matrix and decoding parameters in the encoding and decoding parameters to recover the target text;

[0036] Wherein, the gating parameter matrix includes a first gating matrix for assisting encoding and the second gating matrix for assisting decoding.

[0037] Optionally, when the second processor determines the gating parameter matrix and encoding and decoding parameters required for transmitting data, it is further configured to:

[0038] Receive the gating parameter matrix and the encoding / decoding parameters sent by the first communication device;

[0039] Or, send the gating parameter matrix and the encoding / decoding parameters to the first communication device;

[0040] Wherein, the gating parameter matrix and the encoding / decoding parameters are respectively provided by the gating model and the encoding / decoding model generated by the first communication device through joint training, or are respectively provided by the gating model and the encoding / decoding model generated by the second communication device through joint training.

[0041] Optionally, the second processor performs gating mechanism-based decoding processing on the text-encoded data through the second gating matrix and the decoding parameter in the encoding / decoding parameters, and is further configured to:

[0042] Determine the semantic importance of the sub-encoded data corresponding to each word in the text-encoded data through the second gating matrix, and perform enhancement processing on each sub-encoded data that matches the corresponding semantic importance to obtain enhanced text-encoded data;

[0043] Perform channel decoding on the enhanced text-encoded data to obtain channel-decoded data;

[0044] Perform semantic decoding on the channel-decoded data based on the decoding parameter to recover the target text.

[0045] Optionally, when the second processor determines the semantic importance of the sub-encoded data corresponding to each word in the text-encoded data through the second gating matrix, it is further configured to:

[0046] Map each sub-encoded data to a corresponding third attention information based on a linear transformation matrix; the third attention information includes a third query and a third key obtained by linearly transforming the corresponding sub-encoded data based on the corresponding linear transformation matrix;

[0047] Determine the attention score corresponding to the sub-encoded data based on the third query and the third key corresponding to each sub-encoded data;

[0048] Map the attention score corresponding to each sub-encoded data to a semantic importance score.

[0049] Optionally, when the second processor performs enhancement processing on each sub-encoded data that matches the corresponding semantic importance, it is further configured to:

[0050] Generate a semantic graph; the node information in the semantic graph includes sub-encoded data, and the edge information between nodes is generated based on the semantic importance of the sub-encoded data in the two end nodes corresponding to the edge.

[0051] Based on the corresponding edge information, perform weighted propagation and update on the node information in the semantic graph through a graph convolutional neural network, so as to perform enhancement processing on the sub-encoded data in each node information that matches the corresponding semantic importance.

[0052] Optionally, the second processor is further configured to:

[0053] Report second capability information to the first communication device;

[0054] Or, receive the first capability information reported by the first communication device;

[0055] Wherein, the first capability information indicates that the first communication device supports semantic encoding and decoding based on a gating mechanism; the second capability information indicates that the second communication device supports semantic encoding and decoding based on a gating mechanism.

[0056] A data communication method, including:

[0057] Determine the gating parameter matrix and encoding and decoding parameters required for transmitting data;

[0058] Send text-encoded data on a communication channel;

[0059] Wherein, the text-encoded data is generated by performing encoding processing based on semantics and part of speech on the target text to be sent in a semantic communication task based on the first gating matrix and the encoding parameters in the encoding and decoding parameters; the gating parameter matrix includes the first gating matrix for assisting encoding and the second gating matrix for assisting decoding.

[0060] A data communication method, including:

[0061] Determine the gating parameter matrix and encoding and decoding parameters required for transmitting data;

[0062] Receive text-encoded data on a communication channel; the text-encoded data is generated by the first communication device by performing encoding processing based on semantics and part of speech on the target text to be sent in a semantic communication task;

[0063] Perform decoding processing based on the gating mechanism on the text-encoded data through the second gating matrix and the decoding parameters in the encoding and decoding parameters to recover the target text;

[0064] Wherein, the gating parameter matrix includes the first gating matrix for assisting encoding and the second gating matrix for assisting decoding. Description of the Drawings

[0065] To more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments or related technologies. Obviously, the drawings in the following description are only the embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on the provided drawings.

[0066] Figure 1 is the composition structure diagram of the first communication device provided by the present application;

[0067] Figure 2 is the schematic diagram of the encoding processing process of the first processor provided by the present application based on semantics and part of speech;

[0068] Figure 3 is the schematic diagram of the semantic encoding and decoding process in the text semantic communication provided by the present application;

[0069] Figure 4 is the composition structure diagram of the second communication device provided by the present application;

[0070] Figure 5 is the schematic diagram of the decoding processing process of the second processor provided by the present application based on the gating mechanism;

[0071] Figure 6 is the structure diagram of the receiving end gating mechanism provided by the present application;

[0072] Figure 7 is the semantic graph example provided by the present application;

[0073] Figure 8 and Figure 9 are respectively the implementation flowcharts of text semantic communication in different examples provided by the present application;

[0074] Figure 10 is the schematic diagram of the simulation result of the text semantic communication provided by the present application;

[0075] Figure 11 is the flowchart of the data communication method applied to the first communication device provided by the present application;

[0076] Figure 12 is the flowchart of the data communication method applied to the second communication device provided by the present application. Detailed implementation manners

[0077] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0078] The embodiments of the present application provide a data communication method and a communication device, which are used to overcome the defects of poor restoration quality of the transmitted text and high semantic errors in text semantic communication.

[0079] Among them, semantic communication is a brand-new communication paradigm, and its core lies in improving the accuracy and efficiency of communication by transmitting and understanding the semantic information of data. Text semantic communication is a type of semantic communication, aiming to maximize the system capacity and reduce semantic errors rather than bit or symbol errors in traditional communication by restoring the meaning of sentences in the text.

[0080] The communication device provided by the embodiments of the present application includes a first communication device and a second communication device. Among them, the first communication device and the second communication device are respectively the sending-end communication device and the receiving-end communication device in text semantic communication. Exemplarily, for downlink communication, the first communication device and the second communication device may be a base station and a mobile station respectively; for uplink communication, the first communication device and the second communication device may be a mobile station and a base station respectively. Optionally, the mobile station may specifically be a user equipment (UE) such as a mobile user's mobile phone.

[0081] See Figure 1 , which shows the composition structure diagram of the first communication device. The first communication device provided by the embodiments of the present application includes: a first transceiver 11 and a first processor 12.

[0082] The first transceiver is used for data transceiver, for example, sending or receiving communication data in text semantic communication.

[0083] The first transceiver may specifically include a first transmitter for sending data and a first receiver for receiving data.

[0084] The first processor is coupled to the first transceiver.

[0085] Among them, the first processor is configured to:

[0086] Determine the gating parameter matrix and encoding / decoding parameters required for transmitting data;

[0087] Send text-encoded data on the communication channel;

[0088] The text encoding data is generated by encoding the target text to be sent in the semantic communication task based on semantics and part of speech, based on the first gating matrix in the gating parameter matrix and the encoding parameter in the encoding and decoding parameters.

[0089] The semantic communication task can be an uplink text semantic communication task or a downlink text semantic communication task for text semantic communication in a semantic communication scenario. For example, an uplink text semantic communication task initiated by a user equipment for requesting to upload text data, or a downlink text semantic communication task for requesting to download text data, etc.

[0090] After obtaining the semantic communication task, the first communication device can obtain the text encoding data by encoding the target text to be sent in the semantic communication task, so as to send the text encoding data on a communication channel and implement semantic communication of the target text.

[0091] When performing text encoding, the semantic communication encoding method in the prior art usually only focuses on the word semantics in the text and ignores the part-of-speech features. However, the applicant finds that the part of speech also contains corresponding semantic information, which can further assist in understanding the text meaning. Based on this, the first communication device at the sending end of this application jointly encodes the semantic and part-of-speech information of the words in the text to improve the understanding ability of the overall characteristics of the text, and further improve the recovery quality of the received text at the receiving end.

[0092] Moreover, the applicant finds that it is difficult to achieve a good enhancement effect on semantic reasoning in text semantic communication by simply connecting the semantic features and part-of-speech features (such as word semantic vectors and part-of-speech vectors) of the words in the text. In view of this situation, the embodiment of this application proposes to jointly encode the semantic features and part-of-speech features of the words in the text in a new collaborative learning manner at the sending end to improve the understanding ability of the model for the overall characteristics of the text; at the same time, considering that when the noise intensity in the text transmission signal exceeds the signal intensity and the communication quality is poor, the fidelity of text semantic information extraction and processing will decrease, the embodiment of this application also designs a gating mechanism at the receiving end. Optionally, the gating mechanism is specifically a gating mechanism based on Graph Convolutional Networks (GCN). The designed gating mechanism is used to dynamically adjust the weights of different information in the text transmission signal according to the different importance of different information in the text transmission signal to improve the noise resistance and ensure the fidelity of text semantic information extraction and processing.

[0093] Based on the above co - design of the sender and the receiver, for the text semantic communication task, the first processor in the first communication device can first determine the gating parameter matrix and the encoding - decoding parameters required for data transmission. Among them, the gating parameter matrix is used to support the gating mechanism at the receiver, such as the GCN - based gating mechanism at the receiver. The gating parameter matrix includes a first gating matrix for assisting encoding and a second gating matrix for assisting decoding, and the encoding - decoding parameters include encoding parameters for encoding and decoding parameters for decoding.

[0094] In implementation, during the model training phase, the first communication device at the sender or the second communication device at the receiver can perform joint training of the gating model and the encoding - decoding model in advance. The gating parameter matrix and the encoding - decoding parameters can be provided by the gating model and the encoding - decoding model generated by the first communication device through joint training respectively, or provided by the gating model and the encoding - decoding model generated by the second communication device through joint training respectively.

[0095] The trained encoding - decoding model can be used for semantic encoding - decoding and channel encoding - decoding of text data.

[0096] Correspondingly, after the first processor determines the gating parameter matrix and the encoding - decoding parameters required for data transmission, it is further configured to perform one of the following:

[0097] 11) Send the gating parameter matrix and the encoding - decoding parameters to the second communication device.

[0098] This implementation 11) corresponds to the case where the first communication device at the sender performs joint training of the gating model and the encoding - decoding model.

[0099] 12) Receive the gating parameter matrix and the encoding - decoding parameters sent by the second communication device.

[0100] This implementation 12) corresponds to the case where the second communication device at the receiver performs joint training of the gating model and the encoding - decoding model.

[0101] When the first processor sends the gating parameter matrix and the encoding - decoding parameters to the second communication device or receives the gating parameter matrix and the encoding - decoding parameters sent by the second communication device, it can specifically send some or all of the gating parameter matrix and the encoding - decoding parameters to the second communication device, or receive some or all of the gating parameter matrix and the encoding - decoding parameters sent by the second communication device, and there is no limitation on this.

[0102] Among them, if the first processor sends some of the gating parameter matrix and the encoding and decoding parameters to the second communication device, at least the second gating matrix for assisting decoding in the gating parameter matrix and the decoding parameters for decoding in the encoding and decoding parameters should be sent to the second communication device to ensure that the second communication device can perform gating mechanism-based decoding processing on the received text encoding data through the second gating matrix and the decoding parameters.

[0103] Similarly, if the first processor receives some of the gating parameter matrix and the encoding and decoding parameters sent by the second communication device, at least the first gating matrix for assisting encoding in the gating parameter matrix and the encoding parameters for encoding in the encoding and decoding parameters should be received to ensure that the first communication device can perform corresponding encoding processing on the target text to be sent through the first gating matrix and the encoding parameters.

[0104] In a specific application scenario, the same electronic device, such as a first electronic device like a base station or a second electronic device like a mobile station, can send or receive data in semantic communication according to actual needs, that is, the same electronic device can act as both a data sender and a data receiver depending on its actual communication requirements. Therefore, preferably, the first processor can send all of the gating parameter matrix and the encoding and decoding parameters to the second communication device, or receive all of the gating parameter matrix and the encoding and decoding parameters sent by the second communication device, so that the second communication device or the first communication device has the text encoding and decoding functions based on the gating parameter matrix and the encoding and decoding parameters, thereby supporting it to perform data sending and receiving as needed in text semantic communication.

[0105] On the basis of determining the gating parameter matrix and the encoding and decoding parameters required for transmitting data, for the target text to be sent in the semantic communication task, after the first processor completes the encoding processing of the target text based on semantics and part of speech, it can send the text encoding data corresponding to the target text to the second electronic device through the communication channel to achieve semantic communication of the target text.

[0106] In summary, for text semantic communication, through the dual-end collaborative design of semantic encoding based on part-of-speech enhancement at the sending end and decoding based on the GCN gating mechanism at the receiving end in the embodiments of this application, a closed-loop optimization process can be formed based on part-of-speech enhancement and the dynamic gating mechanism at the sending end and the receiving end, so as to improve the accuracy and robustness of semantic representation through dual-end cooperation, and enhance the flexibility and noise resistance of signals in text semantic communication, thereby ensuring the understanding ability and recovery quality of the overall characteristics of the text.

[0107] In an alternative embodiment, the first processor in the first communication device is further configured to perform semantic and part-of-speech based encoding processing on the target text based on the first gating matrix in the gating parameter matrix and the encoding parameter in the encoding and decoding parameters, so as to generate text encoding data corresponding to the target text.

[0108] See Figure 2 the schematic diagram of the encoding process shown. Optionally, the process of the first processor performing semantic and part-of-speech based encoding processing on the target text based on the first gating matrix in the gating parameter matrix and the encoding parameter in the encoding and decoding parameters can be implemented as follows:

[0109] Step 201: Determine the semantic features and part-of-speech features of each word in the target text, and the semantic features of the context information corresponding to each word.

[0110] Among them, the first processor can first perform word segmentation on the target text to obtain each word of the target text. Each word of the target text can be a single character, a word, or a phrase, and there is no limitation on this, which can be determined according to the actual situation.

[0111] After that, the semantic vector of each word and the part-of-speech vector of the part of speech corresponding to each word can be determined, and the semantic vector and part-of-speech vector of each word are respectively used as the semantic feature and part-of-speech feature of each word.

[0112] Similarly, the semantic vector of the context information corresponding to each word can be determined, and this semantic vector is used as the semantic feature of the context information corresponding to the word.

[0113] Step 202: Map the semantic features and part-of-speech features of each word to the same feature space based on the first gating matrix to obtain a mapping result.

[0114] The first gating matrix includes three groups of learnable projection matrices for feature mapping. In this step, specifically, through this projection matrix, the semantic vector and part-of-speech vector of each word are mapped to the same feature space to obtain the first attention information.

[0115] This first attention information is the mapping result.

[0116] Among them, the first attention information includes the first query, the first key, and the first key value obtained after mapping the semantic vector and part-of-speech vector of the word to the same feature space.

[0117] Assume that the semantic vector and part-of-speech vector of the word are respectively represented as W sem and W pos , and the three groups of projection matrices included in the first gating matrix are respectively represented as A q , A k , Av , specifically, through the following exemplary mapping formula, the semantic vector W sem and the part-of-speech vector W pos can be mapped to the same feature space:

[0118] Q = W sem ·A q ;

[0119] K = W pos ·A k ;

[0120] V = W pos ·A v .

[0121] Among them, Q, K, and V here respectively represent the first query, the first key, and the first key value.

[0122] Step 203: Based on the attention mechanism, perform feature fusion processing on the mapping results corresponding to each word and the semantic features of the corresponding context information to obtain the semantic encoding data of the target text.

[0123] According to the foregoing steps, it can be known that the mapping result includes the first attention information. In this step, correspondingly, based on the attention mechanism, feature fusion processing can be performed on the first attention information corresponding to each word and the semantic features of the corresponding context information. This process can be further implemented as the following steps 21)-22):

[0124] 21) Based on the first attention mechanism and the corresponding encoding parameters, fuse the first attention information of each word to obtain the fusion feature of each word.

[0125] Optionally, the first attention mechanism is the dot product attention mechanism.

[0126] In this step, the first processor can specifically fuse the first query, the first key, and the first key value of each word based on the dot product attention mechanism and the corresponding encoding parameters to obtain the fusion feature of each word. The exemplary dot product attention formula is as follows:

[0127]

[0128] Among them, the operation result, that is, Attention(Q, K, V), is a set of fused embeddings F sem+pos , representing the fusion feature of the word, d represents the vector dimension of the word semantic vector or the part-of-speech vector. The dimensions of these two vectors are the same and remain unchanged in the subsequent processing process, and T represents the matrix transpose.

[0129] 22) Based on the second attention mechanism and corresponding encoding parameters, the fusion features of each word and the semantic features of the corresponding context information are fused to obtain the semantic encoding data of each word, so as to obtain the semantic encoding data of the target text based on the semantic encoding data of each word.

[0130] Optionally, the second attention mechanism is a multi-head attention mechanism.

[0131] After getting the fusion feature F of the word sem+pos After that, the fusion feature F of the word can be further integrated through the multi-head attention mechanism and the corresponding encoding parameters. sem+pos The semantic feature H corresponding to the context information of the word context The fusion is performed and the fusion result is used as the semantic encoding data of the word. Among them, the semantic feature H of the context information corresponding to the word context It can be output through the hidden layer of the encoder-decoder model.

[0132] The fusion feature F of the word is sem+pos The semantic feature H corresponding to the context information of the word context When fusion is performed, the semantic features H corresponding to the context information of the word can be context As the query, the fused features F of the word sem+pos and feature values ​​as keys and key values, and the semantic features H of the context information context And the fusion feature F sem+pos Dynamically extract semantic information using the multi-head attention mechanism in semantic encoding final , and finally the extracted semantic information F final As the semantic encoding data of the word. The semantic encoding data of each word in the target text forms the semantic encoding data F of the target text.

[0133] The encoding parameters may include but are not limited to the number of transformer layers in the encoder / encoding module of the encoding / decoding model, the embedding dimension, the number of attention heads, the dropout parameter size, etc. In step 21) and step 22), corresponding feature fusion processing may be performed based on the corresponding attention mechanism and the corresponding encoding parameters according to actual needs.

[0134] See also Figure 3 The semantic encoding and decoding process shown, steps 201 to 203, is essentially an implementation process of jointly encoding the semantic vector and part-of-speech vector of a word based on the semantic encoding module in the encoding and decoding model provided by the present application.

[0135] Figure 3The input text therein represents the target text to be sent for the semantic communication task at the sending end, and the output text represents the text recovered by the receiving end through decoding processing based on the gating mechanism.

[0136] Step 204: Perform channel coding on the semantic coding data of the target text to obtain the text coding data.

[0137] Based on step 203, with reference to Figure 3 , the channel coding module in the codec model can further perform channel coding on the semantic coding data F of the target text to obtain the text coding data x that can be used for transmission, and then send the text coding data x over the communication channel.

[0138] This embodiment realizes the joint coding of the semantic features and part-of-speech features of words in the text in a new collaborative learning manner at the sending end, which can effectively improve the ability to understand the overall characteristics of the text. In addition, through the codec processing combined with the corresponding gating parameters, a closed-loop optimization process can be formed between the sending end and the receiving end based on part-of-speech enhancement and the dynamic gating mechanism, so as to support the improvement of the flexibility and noise resistance of signals in text semantic communication through two-end cooperation, effectively ensuring the ability to understand the overall characteristics of the text and the recovery quality.

[0139] In an alternative embodiment, the first processor in the first communication device is further configured to:

[0140] Report first capability information to the second communication device; or, receive second capability information reported by the second communication device.

[0141] Wherein, the first capability information indicates that the first communication device supports semantic coding and decoding based on the gating mechanism; the second capability information indicates that the second communication device supports semantic coding and decoding based on the gating mechanism.

[0142] Specifically, the first processor may report the first capability information to the second communication device or receive the second capability information reported by the second communication device before determining the gating parameter matrix and coding and decoding parameters required for transmitting data.

[0143] Wherein, if the first communication device is a mobile station such as a user's mobile phone, the first processor in the first communication device may report the first capability information to the second communication device such as a base station to inform the second communication device that the first communication device supports semantic coding and decoding based on the gating mechanism, thus facilitating subsequent semantic communication processing based on semantic coding and decoding with the gating mechanism between the first communication device and the second communication device.

[0144] If the first communication device is a base station, the first processor in the first communication device can receive the second capability information reported by a second communication device such as a mobile station like a user's mobile phone, so as to know that the second communication device supports semantic encoding and decoding based on a gating mechanism, thereby facilitating subsequent semantic communication processing based on the gating mechanism between the first communication device and the second communication device.

[0145] In this embodiment, before semantic communication, first, the mobile station reports its communication capabilities to the base station, which can facilitate the base station to understand whether the mobile station to be communicated supports semantic encoding and decoding based on the gating mechanism. Furthermore, it is convenient for the base station to make an effective decision on whether to perform text semantic communication based on the gating mechanism with the mobile station to be communicated according to the actual communication capabilities of the mobile station to be communicated, so as to avoid ineffective communication in the case where the mobile station does not have the ability of semantic encoding and decoding based on the gating mechanism.

[0146] See Figure 4 Referring to the structural composition diagram of the second communication device shown, the second communication device provided in the embodiment of the present application includes a second transceiver 21 and a second processor 22.

[0147] Among them, the second transceiver is used for data transmission and reception, such as sending or receiving transmission data in text semantic communication, etc.

[0148] Specifically, the second transceiver may include a second transmitter for sending data and a second receiver for receiving data.

[0149] The second processor is coupled to the second transceiver.

[0150] Among them, the second processor is configured to:

[0151] Determine the gating parameter matrix and encoding and decoding parameters required for transmitting data;

[0152] Receive text-encoded data on the communication channel; the text-encoded data is generated by the first communication device through encoding processing of the target text to be sent in the semantic communication task based on semantics and part of speech;

[0153] Perform decoding processing on the text-encoded data based on the gating mechanism through the second gating matrix and the decoding parameters in the encoding and decoding parameters to recover the target text;

[0154] Among them, as described above, the gating parameter matrix includes a first gating matrix for assisting encoding and the second gating matrix for assisting decoding.

[0155] For the text semantic communication task, similar to the function of the first processor in the first communication device, the second processor in the second communication device also first determines the gating parameter matrix and encoding and decoding parameters required for transmitting data.

[0156] As described above, in implementation, during the model training phase, the joint training of the gating model and the encoding / decoding model can be pre - performed by the first communication device at the sending end or the second communication device at the receiving end. The gating parameter matrix and the encoding / decoding parameters can be provided by the gating model and the encoding / decoding model generated by the first communication device through joint training respectively, or provided by the gating model and the encoding / decoding model generated by the second communication device through joint training respectively.

[0157] Correspondingly, when the second processor determines the gating parameter matrix and the encoding / decoding parameters required for data transmission, it is further configured to perform one of the following:

[0158] 31) Receive the gating parameter matrix and the encoding / decoding parameters sent by the first communication device.

[0159] This implementation 31) corresponds to the case where the first communication device at the sending end performs the joint training of the gating model and the encoding / decoding model.

[0160] 32) Send the gating parameter matrix and the encoding / decoding parameters to the first communication device.

[0161] This implementation 32) corresponds to the case where the second communication device at the receiving end performs the joint training of the gating model and the encoding / decoding model.

[0162] When the second processor receives the gating parameter matrix and the encoding / decoding parameters sent by the first communication device or sends the gating parameter matrix and the encoding / decoding parameters to the first communication device, specifically, it can receive some or all of the gating parameter matrix and the encoding / decoding parameters sent by the first communication device, or send some or all of the gating parameter matrix and the encoding / decoding parameters to the first communication device, and there is no limitation on this.

[0163] Among them, if the second processor receives some of the gating parameter matrix and the encoding / decoding parameters sent by the first communication device, the second processor should at least receive the second gating matrix for assisting decoding in the gating parameter matrix and the decoding parameters for decoding in the encoding / decoding parameters, to ensure that the second communication device can perform gating - mechanism - based decoding processing on the received text - encoded data through the second gating matrix and the decoding parameters.

[0164] Similarly, if the second processor sends some of the gating parameter matrix and the encoding / decoding parameters to the first communication device, the second processor should at least send the first gating matrix for assisting encoding in the gating parameter matrix and the encoding parameters for encoding in the encoding / decoding parameters to the first communication device, to ensure that the first communication device can perform corresponding encoding processing on the target text to be sent through the first gating matrix and the encoding parameters.

[0165] As described above, in a specific application scenario, the same electronic device, such as a first electronic device like a base station or a second electronic device like a mobile station, can send or receive data in semantic communication according to actual needs. That is, the same electronic device can act as both a data sender and a data receiver depending on its actual communication requirements. Therefore, preferably, the second processor can send all the parameters of the gating parameter matrix and the encoding and decoding parameters to the first communication device, or receive all the parameters of the gating parameter matrix and the encoding and decoding parameters sent by the first communication device, so that the first communication device or the second communication device has the text encoding and decoding functions based on the gating parameter matrix and the encoding and decoding parameters, thereby supporting it to send and receive data as needed in text semantic communication.

[0166] Based on determining the gating parameter matrix and the encoding and decoding parameters required for transmitting data, for text semantic communication, the second processor can receive the text-encoded data of the target text on the communication channel, and perform gating mechanism-based decoding processing on the text-encoded data through the second gating matrix and the decoding parameters in the encoding and decoding parameters to recover the target text.

[0167] The process of the second processor performing gating mechanism-based decoding processing on the received text-encoded data will be described in detail in the following embodiments.

[0168] In summary, in this embodiment, by receiving at the receiving end the text-encoded data generated by the sending end through encoding processing of the target text based on semantics and part of speech, and using the second gating matrix and the decoding parameters to perform gating mechanism-based decoding processing on the text-encoded data, a closed-loop optimization process is formed at the sending end and the receiving end based on part-of-speech enhancement and a dynamic gating mechanism. Thus, the flexibility and noise resistance of the signal in text semantic communication can be improved based on two-end cooperation, and the understanding ability and recovery quality of the overall characteristics of the text can be enhanced.

[0169] In an alternative embodiment, when the second processor performs gating mechanism-based decoding processing on the text-encoded data through the second gating matrix and the decoding parameters in the encoding and decoding parameters, it is further configured to execute Figure 5 the decoding processing flow shown as follows:

[0170] Step 501: Determine the semantic importance of the sub-encoded data corresponding to each word in the text-encoded data through the second gating matrix, and perform enhancement processing on each sub-encoded data that matches the corresponding semantic importance to obtain the enhanced text-encoded data.

[0171] Among them, the second gating matrix includes three linear transformation matrices for linear transformation.

[0172] The second processor determines the semantic importance of the sub-encoded data corresponding to each word in the text-encoded data through the second gating matrix, and is further configured to extract the semantic importance score of the sub-encoded data from the word semantic information and part-of-speech corresponding to the sub-encoded data by using a shallow network based on the attention mechanism. This process can be implemented as the following steps 41-43):

[0173] 41) Map each of the sub-encoded data to corresponding third attention information based on a linear transformation matrix; the third attention information includes a third query and a third key obtained by linearly transforming the corresponding sub-encoded data based on the corresponding linear transformation matrix.

[0174] In this embodiment, the text-encoded data received by the second processor on the communication channel is represented as y. Compared with the text-encoded data x sent by the first communication device on the communication channel, there may be noise or loss in the text-encoded data y.

[0175] See also in conjunction with Figure 6 In the structure diagram of the receiving-end gating mechanism shown, in step 41), the second processor can map the received text-encoded data y to a third query Q, a third key K, and a third key value V by performing three sets of linear transformations based on three sets of linear transformation matrices.

[0176] Where, Q = y · W q , K = y · W k , V = y · W v , W q , W k and W v respectively represent three sets of trainable linear transformation matrices.

[0177] 42) Determine the attention score corresponding to the sub-encoded data based on the third query and the third key corresponding to each sub-encoded data.

[0178] Optionally, the second processor can specifically calculate the attention score corresponding to the sub-encoded data according to the third query and the third key corresponding to each sub-encoded data through the following exemplary attention score calculation formula:

[0179]

[0180] Where, Q and K here respectively represent the third query and the third key, d represents the vector dimension of the word semantic vector or part-of-speech vector, and T represents matrix transpose.

[0181] 43) Map the attention score corresponding to each sub-encoded data to a semantic importance score.

[0182] After obtaining the attention score A corresponding to the sub-encoded data, with reference to Figure 6 the second processor can map the attention score A to a semantic importance score S through a linear transformation and an activation function, so as to characterize the semantic importance of the sub-encoded data based on the semantic importance score S corresponding to the sub-encoded data.

[0183] After that, the second processor can continue to perform enhancement processing on each sub-encoded data in the text-encoded data y that matches the corresponding semantic importance.

[0184] When the second processor performs enhancement processing on each of the sub-encoded data that matches the corresponding semantic importance, it is further configured to perform the following steps 51)-52):

[0185] 51) Generate a semantic graph; the node information in the semantic graph includes sub-encoded data, and the edge information between nodes is generated based on the semantic importance of the sub-encoded data at both ends of the edge corresponding to the edge.

[0186] Optionally, the edge information corresponding to the edge between nodes can be an edge weight.

[0187] Specifically, the edge information / edge weight can be generated by performing a weighted operation on the semantic importance of the sub-encoded data at both ends of the edge corresponding to the edge, and is used to characterize the correlation between the different single-word semantic characteristics represented by the sub-encoded data of different words at both ends of the edge corresponding to the edge.

[0188] With reference to Figure 7 shows an example of a semantic graph, where the node information on each node is respectively the sub-encoded data corresponding to each word in the target text "The cat is sitting on the book" in the text-encoded data of the target text. This sub-encoded data is essentially a semantic feature vector containing single-word semantics, part of speech, and semantic information of the context corresponding to the word. The edge information / edge weight between nodes can be generated by performing a weighted operation on the semantic importance scores of the sub-encoded data at both ends of the edge corresponding to the edge.

[0189] 52) Based on the corresponding edge information, use a graph convolutional neural network to perform weighted propagation and update on the node information in the semantic graph, so as to perform enhancement processing on the sub-encoded data in each node information that matches the corresponding semantic importance.

[0190] With reference to Figure 6 a graph convolutional neural network GCN can, based on the corresponding edge information, perform weighted propagation and update on the node information in the semantic graph through graph convolutional operations and adaptive adjustment of the adjacency matrix, so as to achieve enhancement processing on the sub-encoded data in each node information that matches the corresponding semantic importance.

[0191] Among them, the node information update process for each node on the l-th layer can be expressed as: to effectively capture the global and local relationships in the semantic graph. In this formula represents the normalized symmetric adjacency matrix, represents the convolution weight of the l-th layer, and σ(·) represents the activation function.

[0192] Step 502: Perform channel decoding on the enhanced text encoding data to obtain channel decoding data.

[0193] Referring to Figure 3 the semantic encoding and decoding process shown, after obtaining the enhanced text encoding data, the enhanced text encoding data can be channel decoded through the channel decoding module in the encoding and decoding model, so as to obtain the channel decoding data.

[0194] Step 503: Perform semantic decoding on the channel decoding data based on the decoding parameters to recover the target text.

[0195] After obtaining the channel decoding data, referring to Figure 3 it is further possible to perform semantic decoding on the channel decoding data through the semantic decoding module in the encoding and decoding model based on the corresponding decoding parameters to recover the target text.

[0196] Among them, the decoding parameters may include but are not limited to the number of transformer layers, embedding dimension, number of attention heads, dropout parameter size, etc. in the decoder / decoding module of the encoding and decoding model.

[0197] This embodiment realizes decoding processing based on the gating mechanism for the received text encoding data at the receiving end. By combining the semantic features and part-of-speech features of words in the text in a new collaborative learning manner at the sending end, a closed-loop optimization process can be formed between the sending end and the receiving end based on part-of-speech enhancement and the dynamic gating mechanism, supporting the improvement of the flexibility and noise resistance of signals in text semantic communication through two-end cooperation, and ensuring the understanding ability and recovery quality of the overall characteristics of the text.

[0198] In an optional embodiment, the second processor in the second communication device is further configured to:

[0199] report second capability information to the first communication device; or, receive the first capability information reported by the first communication device;

[0200] Among them, the first capability information characterizes that the first communication device supports semantic encoding and decoding based on the gating mechanism; the second capability information characterizes that the second communication device supports semantic encoding and decoding based on the gating mechanism.

[0201] Specifically, the second processor may report second capability information to the first communication device before determining the gating parameter matrix and encoding / decoding parameters required for data transmission; or receive the first capability information reported by the first communication device.

[0202] Among them, if the second communication device is a mobile station such as a user's mobile phone, the second processor in the second communication device may report the second capability information to the first communication device such as a base station, so as to inform the first communication device that the second communication device supports semantic encoding / decoding based on the gating mechanism, thereby facilitating subsequent semantic communication processing based on the gating mechanism between the first communication device and the second communication device.

[0203] If the second communication device is a base station, the second processor in the second communication device may receive the first capability information reported by the first communication device such as a mobile station like a user's mobile phone, so as to know that the first communication device supports semantic encoding / decoding based on the gating mechanism, thereby facilitating subsequent semantic communication processing based on the gating mechanism between the first communication device and the second communication device.

[0204] In this embodiment, by first reporting the communication capability from the mobile station to the base station before semantic communication, it is convenient for the base station to know whether the mobile station to be communicated supports semantic encoding / decoding based on the gating mechanism, and then it is convenient for the base station to make an effective decision on whether to perform text semantic communication based on the gating mechanism with the mobile station to be communicated according to the actual communication capability of the mobile station to be communicated, so as to avoid ineffective communication when the mobile station does not have the ability of semantic encoding / decoding based on the gating mechanism.

[0205] The following provides an application example of the method of this application.

[0206] In this example, the first communication device at the sending end is a base station, and the second communication device at the receiving end is a mobile station. In this example, the joint training of the gating model and the encoding / decoding model is pre - carried out at the base station or the mobile station, so as to obtain the gating parameter matrix and the encoding / decoding parameters. The training dataset can be, but is not limited to, the European Parliament dataset.

[0207] See Figure 8, which shows the implementation flowchart of text semantic communication in the case of joint training of the gating model and the encoding and decoding model at the base station. In this example, the mobile station first reports its capability information to the base station to inform the base station that it (the mobile station) has the semantic encoding and decoding communication capability based on the gating mechanism. After that, the base station conducts joint training of the gating model and the encoding and decoding model to obtain the gating parameter matrix (which can be simply referred to as the "gating matrix") and the encoding and decoding parameters, and sends the corresponding gating matrix and decoding parameters to the mobile station. On this basis, for the target text to be transmitted in the semantic communication task, after the base station performs semantic and part-of-speech based encoding processing on the target text based on the first gating parameter in the gating parameter matrix and the encoding parameter in the encoding and decoding parameters, it sends the obtained text encoding data to the mobile station, so that the mobile station can perform decoding processing on the text encoding data based on the second gating parameter in the gating parameter matrix and the decoding parameter in the encoding and decoding parameters, thereby restoring the target text.

[0208] See Figure 9 , which shows the implementation flowchart of text semantic communication in the case of joint training of the gating model and the encoding and decoding model at the mobile station. In this example, the mobile station also first reports its capability information to the base station to inform the base station that it (the mobile station) has the semantic encoding and decoding communication capability based on the gating mechanism. After that, the mobile station conducts joint training of the gating model and the encoding and decoding model to obtain the gating parameter matrix (which can be simply referred to as the "gating matrix") and the encoding and decoding parameters, and sends the corresponding gating matrix and encoding parameters to the base station. On this basis, for the target text to be transmitted in the semantic communication task, after the base station performs semantic and part-of-speech based encoding processing on the target text based on the first gating parameter in the gating parameter matrix and the encoding parameter, it sends the obtained text encoding data to the mobile station, so that the mobile station can perform decoding processing on the text encoding data based on the second gating parameter in the gating parameter matrix and the decoding parameter in the encoding and decoding parameters, thereby restoring the target text.

[0209] In practical applications, the above example can be implemented by building a simulation environment and performing simulation processing. Optionally, this example uses the text semantic recovery evaluation metric BLEU (Bilingual Evaluation Understudy) to measure the performance of the method of this application in text semantic communication.

[0210] Among them, BLEU is an evaluation metric for text semantic recovery, which is used to evaluate the difference between the predicted text (the text decoded and restored) and the actual text. The BLEU score ranges within the interval [0, 1]. The higher the BLEU score, the higher the semantic similarity between the predicted text and the actual text. If the BLEU score is equal to 1, the predicted text is exactly the same as the actual text. If the BLEU score is equal to 0, the predicted text is completely different from the actual text.

[0211] See Figure 10 the schematic diagram of the simulation results of text semantic communication shown in Figure 10 . In this example, within the SNR (Signal-to-Noise Ratio) range of 0 - 18 dB, BLEU(1-gram) remains stable at a relatively high value with an increase in SNR, and the score is around 0.93. Therefore, based on the method of the embodiments of the present application, the recovery quality of the text in text semantic communication can be improved, and the effect of reducing text semantic errors can be achieved.

[0212] Corresponding to the above-mentioned first communication device, an embodiment of the present application further provides a data communication method applied to the first communication device. See Figure 11 , the data communication method applied to the first communication device includes:

[0213] Step 1101: Determine the gating parameter matrix and encoding / decoding parameters required for transmitting data;

[0214] Step 1102: Transmit the text-encoded data over the communication channel;

[0215] Wherein, the text-encoded data is generated by performing encoding processing based on semantics and part-of-speech on the target text to be transmitted for the semantic communication task based on the first gating matrix and the encoding parameters in the encoding / decoding parameters; the gating parameter matrix includes the first gating matrix for assisting encoding and the second gating matrix for assisting decoding.

[0216] In an alternative embodiment, the determining the gating parameter matrix and encoding / decoding parameters required for transmitting data includes:

[0217] Transmit the gating parameter matrix and the encoding / decoding parameters to the second communication device;

[0218] Or, receive the gating parameter matrix and the encoding / decoding parameters transmitted by the second communication device;

[0219] Wherein, the gating parameter matrix and the encoding / decoding parameters are respectively provided by the gating model and the encoding / decoding model jointly trained by the first communication device, or are respectively provided by the gating model and the encoding / decoding model jointly trained by the second communication device.

[0220] In an alternative embodiment, before transmitting the text-encoded data over the communication channel, the method further includes:

[0221] Determine the semantic features and part-of-speech features of each word in the target text, as well as the semantic features of the context information corresponding to each word;

[0222] Map the semantic features and part-of-speech features of each word to the same feature space based on the first gating matrix to obtain a mapping result;

[0223] Feature fusion processing is performed on the semantic features of the mapping result corresponding to each word and the corresponding context information based on the attention mechanism to obtain the semantic encoding data of the target text;

[0224] Channel encoding is performed on the semantic encoding data to obtain the text encoding data.

[0225] In an alternative embodiment, the mapping the semantic features and part-of-speech features of each word to the same feature space based on the first gating matrix to obtain a mapping result includes:

[0226] Mapping the semantic vector and part-of-speech vector of each word to the same feature space through a projection matrix to obtain first attention information; the first attention information includes a first query, a first key, and a first key value obtained by mapping the semantic vector and the part-of-speech vector to the same feature space;

[0227] The feature fusion processing of the mapping result corresponding to each word and the semantic features of the corresponding context information based on the attention mechanism to obtain the semantic encoding data of the target text includes:

[0228] Based on the first attention mechanism and corresponding encoding parameters, fusing the first attention information of each word to obtain the fusion feature of each word;

[0229] Based on the second attention mechanism and corresponding encoding parameters, fusing the fusion feature of each word and the semantic features of the corresponding context information to obtain the semantic encoding data of each word, so as to obtain the semantic encoding data of the target text based on the semantic encoding data of each word.

[0230] In an alternative embodiment, before the method determines the gating parameter matrix and encoding / decoding parameters required for transmitting data, it further includes:

[0231] Reporting first capability information to a second communication device;

[0232] Or, receiving second capability information reported by the second communication device;

[0233] Wherein, the first capability information indicates that the first communication device supports semantic encoding and decoding based on the gating mechanism; the second capability information indicates that the second communication device supports semantic encoding and decoding based on the gating mechanism.

[0234] For the data communication method disclosed in this embodiment, since it corresponds to the first communication device disclosed in the corresponding embodiment above, the description is relatively simple. For relevant similarities, please refer to the description of the first communication device in the corresponding embodiment above, and details are not described here again.

[0235] Corresponding to the above-mentioned second communication device, an embodiment of the present application further provides a data communication method applied to the second communication device. Refer to Figure 12 The data communication method applied to the second communication device includes:

[0236] Step 1201: Determine the gating parameter matrix and encoding / decoding parameters required for transmitting data;

[0237] Step 1202: Receive text-encoded data on the communication channel; the text-encoded data is generated by the first communication device through encoding processing of the target text to be sent for the semantic communication task based on semantics and part of speech;

[0238] Step 1203: Perform decoding processing based on the gating mechanism on the text-encoded data through the second gating matrix and the decoding parameter in the encoding / decoding parameters to recover the target text;

[0239] Among them, the gating parameter matrix includes a first gating matrix for assisting encoding and the second gating matrix for assisting decoding.

[0240] In an alternative embodiment, the determining the gating parameter matrix and encoding / decoding parameters required for transmitting data includes:

[0241] Receiving the gating parameter matrix and the encoding / decoding parameters sent by the first communication device;

[0242] Or, sending the gating parameter matrix and the encoding / decoding parameters to the first communication device;

[0243] Among them, the gating parameter matrix and the encoding / decoding parameters are respectively provided by the gating model and the encoding / decoding model generated by the first communication device through joint training, or are respectively provided by the gating model and the encoding / decoding model generated by the second communication device through joint training.

[0244] In an alternative embodiment, performing decoding processing based on the gating mechanism on the text-encoded data through the second gating matrix and the decoding parameter in the encoding / decoding parameters includes:

[0245] Determine the semantic importance of the sub-encoded data corresponding to each word in the text-encoded data through the second gating matrix, perform enhancement processing on each sub-encoded data that matches the corresponding semantic importance, and obtain enhanced text-encoded data;

[0246] Perform channel decoding on the enhanced text-encoded data to obtain channel decoding data;

[0247] Perform semantic decoding on the channel decoding data based on the decoding parameter to recover the target text.

[0248] In an alternative embodiment, determining the semantic importance of the sub-encoded data corresponding to each word in the text-encoded data through the second gating matrix includes:

[0249] Mapping each of the sub-encoded data to corresponding third attention information based on a linear transformation matrix; the third attention information includes a third query and a third key obtained by linearly transforming the corresponding sub-encoded data based on the corresponding linear transformation matrix;

[0250] Determining the attention score corresponding to the sub-encoded data based on the third query and the third key corresponding to each of the sub-encoded data;

[0251] Mapping the attention score corresponding to each of the sub-encoded data to a semantic importance score.

[0252] In an alternative embodiment, performing enhancement processing on each of the sub-encoded data that matches the corresponding semantic importance includes:

[0253] Generating a semantic graph; the node information in the semantic graph includes sub-encoded data, and the edge information between nodes is generated based on the semantic importance of the sub-encoded data at both ends of the edge corresponding to the edge;

[0254] Performing weighted propagation and update on the node information in the semantic graph through a graph convolutional neural network based on the corresponding edge information, so as to perform enhancement processing on the sub-encoded data in each node information that matches the corresponding semantic importance.

[0255] In an alternative embodiment, before determining the gating parameter matrix and the encoding / decoding parameters required for transmitting data, the above method further includes:

[0256] Reporting second capability information to the first communication device;

[0257] Or, receiving the first capability information reported by the first communication device;

[0258] Wherein, the first capability information indicates that the first communication device supports semantic encoding / decoding based on a gating mechanism; the second capability information indicates that the second communication device supports semantic encoding / decoding based on a gating mechanism.

[0259] For the data communication method disclosed in this embodiment, since it corresponds to the second communication device disclosed in the corresponding embodiment above, the description is relatively simple. For relevant similarities, please refer to the description of the second communication device in the corresponding embodiment above, and details are not described here again.

[0260] It should be noted that the various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other.

[0261] For the convenience of description, when describing the above system or device, it is divided into various modules or units according to functions for separate description. Of course, when implementing the present application, the functions of each unit can be realized in the same or multiple software and / or hardware.

[0262] From the description of the above embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the technical solution of the present application, in essence, or the part that makes a creative contribution can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disc, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present application.

[0263] Finally, it should also be noted that in this text, relational terms such as first, second, third, and fourth are used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of another identical element in the process, method, article or device including the said element.

[0264] The above are only the preferred embodiments of the present application. It should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. A first communication device, comprising: a first transceiver; and a first processor coupled to the first transceiver, wherein the first processor is configured to: Determine the gating parameter matrix and encoding and decoding parameters required for transmitting data; sending text encoded data over a communication channel; Among them, the text encoding data is generated by encoding the target text to be sent for the semantic communication task based on semantics and part of speech based on the first gating matrix and the encoding parameters in the encoding and decoding parameters; the gating parameter matrix includes the first gating matrix for auxiliary encoding and the second gating matrix for auxiliary decoding.

2. The first communication device according to claim 1, wherein the first processor is further configured to: Sending the gating parameter matrix and the encoding and decoding parameters to a second communication device; or, receiving the gating parameter matrix and the encoding and decoding parameters sent by the second communication device; in, The gating parameter matrix and the encoding and decoding parameters are respectively provided by the gating model and the encoding and decoding model generated by the first communication device through joint training, or are respectively provided by the gating model and the encoding and decoding model generated by the second communication device through joint training.

3. The first communication device according to claim 1, wherein the first processor is further configured to: Determine the semantic features and part-of-speech features of each word in the target text, as well as the semantic features of the context information corresponding to each word; Mapping the semantic features and part-of-speech features of each word to the same feature space based on the first gating matrix to obtain a mapping result; Based on the attention mechanism, feature fusion processing is performed on the mapping result corresponding to each word and the semantic features of the corresponding context information to obtain the semantic encoding data of the target text; Channel coding is performed on the semantically coded data to obtain the text coded data.

4. The first communication device according to claim 1, wherein the first processor is further configured to: Reporting the first capability information to the second communication device; or, receiving second capability information reported by a second communication device; in, The first capability information indicates that the first communication device supports semantic coding and decoding based on a gating mechanism; the second capability information indicates that the second communication device supports semantic coding and decoding based on a gating mechanism.

5. A second communication device, comprising: a second transceiver; and a second processor coupled to the second transceiver, wherein the second processor is configured to: Determine the gating parameter matrix and encoding and decoding parameters required for transmitting data; Receiving text encoding data in a communication channel; the text encoding data is generated by the first communication device through encoding processing based on semantics and part of speech on a target text to be sent in a semantic communication task; By using a second gating matrix and decoding parameters in the encoding and decoding parameters, the text encoding data is subjected to a decoding process based on a gating mechanism to restore the target text; The gating parameter matrix includes a first gating matrix for auxiliary encoding and a second gating matrix for auxiliary decoding.

6. The second communication device according to claim 5, wherein the second processor is further configured to: Receiving the gating parameter matrix and the encoding and decoding parameters sent by the first communication device; or, sending the gating parameter matrix and the encoding and decoding parameters to the first communication device; in, The gating parameter matrix and the encoding and decoding parameters are respectively provided by the gating model and the encoding and decoding model generated by the first communication device through joint training, or are respectively provided by the gating model and the encoding and decoding model generated by the second communication device through joint training.

7. The second communication device according to claim 5, wherein the second processor performs a decoding process based on a gating mechanism on the text encoded data by using a second gating matrix and a decoding parameter in the encoding and decoding parameters, and is further configured to: Determining the semantic importance of the sub-encoded data corresponding to each word in the text encoded data by using the second gating matrix, performing enhancement processing matching the corresponding semantic importance on each of the sub-encoded data, and obtaining enhanced text encoded data; Channel-decoding the enhanced text-encoded data to obtain channel-decoded data; The channel-decoded data is semantically decoded based on the decoding parameters to restore the target text.

8. The second communication device according to claim 5, wherein the second processor is further configured to: Reporting second capability information to the first communication device; or, receiving first capability information reported by the first communication device; in, The first capability information indicates that the first communication device supports semantic coding and decoding based on a gating mechanism; the second capability information indicates that the second communication device supports semantic coding and decoding based on a gating mechanism.

9. A data communication method, comprising: Determine the gating parameter matrix and encoding and decoding parameters required for transmitting data; sending text encoded data over a communication channel; Among them, the text encoding data is generated by encoding the target text to be sent for the semantic communication task based on semantics and part of speech based on the first gating matrix and the encoding parameters in the encoding and decoding parameters; the gating parameter matrix includes the first gating matrix for auxiliary encoding and the second gating matrix for auxiliary decoding.

10. A data communication method, comprising: Determine the gating parameter matrix and encoding and decoding parameters required for transmitting data; receiving text encoded data on a communication channel; The text encoding data is generated by the first communication device by performing encoding processing based on semantics and part of speech on the target text to be sent in the semantic communication task; By using a second gating matrix and decoding parameters in the encoding and decoding parameters, the text encoding data is subjected to a decoding process based on a gating mechanism to restore the target text; The gating parameter matrix includes a first gating matrix for auxiliary encoding and a second gating matrix for auxiliary decoding.

Images