[go: up one dir, main page]

CN114564963B - A retransmission method for semantic communication - Google Patents

A retransmission method for semantic communication Download PDF

Info

Publication number
CN114564963B
CN114564963B CN202210063290.9A CN202210063290A CN114564963B CN 114564963 B CN114564963 B CN 114564963B CN 202210063290 A CN202210063290 A CN 202210063290A CN 114564963 B CN114564963 B CN 114564963B
Authority
CN
China
Prior art keywords
vector
decoder
semantic
retransmission
layer
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202210063290.9A
Other languages
Chinese (zh)
Other versions
CN114564963A (en
Inventor
周清炀
李荣鹏
赵志峰
张宏纲
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University ZJU
Original Assignee
Zhejiang University ZJU
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zhejiang University ZJU filed Critical Zhejiang University ZJU
Priority to CN202210063290.9A priority Critical patent/CN114564963B/en
Publication of CN114564963A publication Critical patent/CN114564963A/en
Application granted granted Critical
Publication of CN114564963B publication Critical patent/CN114564963B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/30—Semantic analysis
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/10—Text processing
    • G06F40/12—Use of codes for handling textual entities
    • G06F40/126—Character encoding
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/237—Lexical tools
    • G06F40/242—Dictionaries
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/279—Recognition of textual entities
    • G06F40/284—Lexical analysis, e.g. tokenisation or collocates
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/279—Recognition of textual entities
    • G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D30/00—Reducing energy consumption in communication networks
    • Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Detection And Prevention Of Errors In Transmission (AREA)

Abstract

本发明提出了一种语义通信的重传方法，提出了基于语义通信的重传方式，相较于遇到差错时，直接重传信息，该方法有效的利用了之前传输的信息，通过将重传的信息与最初传输的信息的整合，有校的提高了重传的准确率。提出了两种具体的重传方案，一种方案是多解码器方式，采用不同输入维度的解码器，在重传时候，采用更大输入维度的解码器整合多次的传输信息，进行一个信息解码。另一种方案是单解码器方式，直接采用一个大输入维度的解码器，在信息未重传时，对剩余的空维度采取一个自动补零的方式，节约了接收方的内存使用。

The present invention proposes a retransmission method for semantic communication and a retransmission method based on semantic communication. Compared with directly retransmitting information when an error occurs, this method effectively utilizes the previously transmitted information and effectively improves the accuracy of retransmission by integrating the retransmitted information with the initially transmitted information. Two specific retransmission schemes are proposed. One scheme is a multi-decoder method, which uses decoders with different input dimensions. When retransmitting, a decoder with a larger input dimension is used to integrate multiple transmissions and perform one information decoding. Another scheme is a single decoder method, which directly uses a decoder with a large input dimension. When the information is not retransmitted, an automatic zero-filling method is used for the remaining empty dimensions, thereby saving the memory usage of the receiver.

Description

Retransmission method for semantic communication
Technical Field
The invention relates to the technical field of semantic communication, in particular to a retransmission method of semantic communication.
Background
According to shannon and wever proposed information elements, communications can be divided into three layers. The first layer is a transmission problem, which mainly researches how to accurately transmit the communication symbols, the second layer is a semantic problem, which mainly researches how to accurately transmit the semantics in the communication symbols, and the third layer is a utility problem, which mainly solves the problem how to effectively influence the behaviors of the received semantics according to the expected mode. Because of the time limitation, in seventy years since shannon established the information theory, a great number of scholars have made a great deal of attempts to approach shannon's limit, but these efforts have focused mainly on the first level of communication, i.e. how to accurately transmit communication symbols. In recent years, the development of related technologies such as artificial intelligence and natural language processing has provided possibilities for exploring the second level of semantic communication, and semantic communication will gradually become a research trend in the communication field.
In the communication process, when facing different complex signal-to-noise conditions, transmission errors are unavoidable, when encountering transmission errors, the error conditions encountered in the transmission process can be avoided by continuously retransmitting information, the characteristics of different coding modes can be utilized in the traditional communication, the HARQ mode is adopted, the information is retransmitted, and meanwhile, the information is combined with the information transmitted before, so that channel resources are saved, and meanwhile, the accuracy of transmission is improved. However, in terms of semantic communication, aiming at the problem of information retransmission, no good solution is provided, so a semantic communication retransmission method based on a transducer is provided, the strong computing capacity of a neural network is fully exerted, the information transmitted before is fully utilized after the information is retransmitted, and the two are combined, thereby improving the accuracy of retransmission.
Disclosure of Invention
The invention aims to provide a retransmission method of semantic communication, which overcomes the defects in the prior art.
In order to achieve the above purpose, the present invention provides the following technical solutions:
the application discloses a retransmission method of semantic communication, which comprises the following steps:
S1, training a retransmission mode, wherein the retransmission mode comprises a multi-decoder mode and a single-decoder mode;
s2, numbering all words possibly used in the transmission process of the two parties according to the shared knowledge of the two parties, and creating a dictionary;
S3, word embedding is carried out on the input sentences by utilizing an input sentence embedding module, and position vectors are added; word embedding is carried out on the target sentences by utilizing a target sentence word embedding module, and position vectors are added; obtaining a word vector with a position vector corresponding to an input sentence and a word vector with a position vector corresponding to a target sentence;
S4, semantic coding: word vectors with position vectors corresponding to input sentences pass through a semantic coding layer, semantic coding is completed through a Encoder layer of Transformrer, and semantic coding vectors are obtained;
s5, passing the semantic coding vector through a wireless channel;
S6, semantic decoding: semantic coding vectors passing through a wireless channel are subjected to semantic decoding through a decoding layer of a transducer, and decoded semantic text is output after probability logistic regression processing;
S7, the receiver performs retransmission judgment according to the received semantic text; if retransmission is needed, a retransmission instruction is sent to a sender, the sender selects a retransmission mode, and the semantic code vector in the step S4 is sent to a wireless channel again to obtain a retransmitted semantic code vector; if no retransmission is needed, ending the transmission;
S8, combining the retransmitted semantic code vector with the previously received semantic code vector, and aggregating through a dimension integration module to obtain an aggregated semantic code vector; carrying out semantic decoding on the aggregated semantic coding vectors through a decoding layer of a transducer, outputting decoded semantic text after probability logistic regression processing, and returning to the step S7;
preferably, the multi-decoder mode includes the following training process:
A11, training an encoder and a decoder which can normally transmit under each signal-to-noise ratio;
a12, fixing the encoder in the step A11, and newly creating a second decoder, wherein the input of the second decoder is twice as large as that of the decoder in the step S1, and a dimension integration module is arranged in the second decoder; the dimension integration module is composed of a full-connection layer;
A13, transmitting the coding results of the twice coder, splicing the two coding results after the two coding results pass through a wireless channel, sending the spliced results into a second decoder, integrating the two coding results by utilizing a dimension integration module in the second decoder, and performing corresponding decoding operation on the integrated information by the second decoder;
a14, newly creating an N decoder according to the step A12 and the step A13, wherein N is a natural number greater than 2, and the input of the N decoder is N times of the decoder in the step A11;
and A15, finishing training, and sequentially starting the decoders according to the serial numbers of the decoders according to the number of retransmission required in the transmission process.
Preferably, the single decoder mode includes the following training process:
B11, establishing an encoder and a decoder, wherein a dimension integration module is arranged in the decoder, testing performance of the decoder under different retransmission times according to a channel to be transmitted and a text for training, determining the allowed maximum retransmission times N according to the performance, and determining the input dimension of the dimension integration module in the decoder according to the maximum retransmission times N; the input dimension of the dimension integrating module is N+1 times of the output dimension of the encoder;
B12, randomly determining the number of times of retransmission required in the training process, wherein the number of times of retransmission is smaller than the maximum number of times of retransmission, splicing the retransmitted information with the original information, taking the information as an input dimension of a dimension integrating module, and carrying out zero padding operation on the rest empty dimension;
B13, repeating the step B12 to obtain a single decoder which can be used for multiple retransmission.
Preferably, the step S2 specifically includes the following steps:
S21, reading the whole text file for transmission;
s22, word segmentation is carried out on the whole text, the use times of each word in the text are counted, each word is numbered, and words with too low use times are removed;
s23, adding the beginning or ending characters into the whole dictionary;
s24, outputting a dictionary.
Preferably, the specific steps in the step S3 are as follows:
s31, creating an embedding layer, sending sentences to be transmitted into the embedding layer, and converting the sentences into word vectors with mapping dimensions;
s32, calculating and adding a position vector;
s33, adding the word vector and the position vector to obtain the word vector with the position information.
Preferably, the specific steps in the step S4 are as follows:
S41, defining three matrixes, and carrying out three-time linear change on word vectors with position vectors corresponding to input sentences according to the three matrixes to obtain query vectors, key vectors and value vectors;
S42, self-Attention calculation is carried out on the query vector, the key vector and the value vector, so that an Attention vector is obtained;
S43, carrying out residual connection, adding the attention vector and the word vector with the position vector corresponding to the input sentence, and carrying out layer normalization operation on the obtained result to obtain a residual vector;
s44, performing feedforward transmission operation on the obtained residual vector, and activating the residual vector by using an activation function ReLU through two layers of linear mapping to obtain a semantic coding vector.
Preferably, the specific steps in the step S6 are as follows:
s61, inputting word vectors with position vectors corresponding to the obtained target sentences in the S3 into a multi-head self-attention layer for decoding;
S62, inputting the coded information received through the wireless channel and the output information of the multi-head self-attention layer in the last step into the multi-head attention layer for decoding;
S63, after the target sentence passes through the multi-head self-attention layer and the multi-head attention layer, a semantic decoding vector is obtained through a feedforward transmission layer;
S64, carrying out probability logistic regression through a Softmax function, and outputting sentences.
The invention has the beneficial effects that:
1. compared with the method for directly retransmitting information when encountering errors, the method effectively utilizes the information transmitted before, and improves the accuracy of retransmission by integrating the retransmitted information with the information transmitted initially.
2. Two specific retransmission schemes are provided, wherein one scheme is a multi-decoder mode, decoders with different input dimensions are adopted, and during retransmission, decoders with larger input dimensions are adopted to integrate multiple times of transmission information so as to decode one information; the other scheme is a single decoder mode, a decoder with a large input dimension is directly adopted, when information is not retransmitted, an automatic zero filling mode is adopted for the rest empty dimension, and the memory use of a receiver is saved in both modes.
The features and advantages of the present invention will be described in detail by way of example with reference to the accompanying drawings.
Drawings
FIG. 1 is a schematic diagram of a framework of a retransmission method of semantic communications according to the present invention;
FIG. 2 is a schematic diagram of a multi-decoder mode according to an embodiment of the present invention;
fig. 3 is a schematic diagram of a single decoder mode according to an embodiment of the present invention.
Detailed Description
The present invention will be further described in detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the detailed description and specific examples, while indicating the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention. In addition, in the following description, descriptions of well-known structures and techniques are omitted so as not to unnecessarily obscure the present invention.
Referring to fig. 1, an embodiment of the present invention provides a retransmission method of semantic communications,
Step one: numbering all words possibly used in the transmission process of the two parties according to the shared knowledge of the two parties, and creating a dictionary;
The dictionary is created by the following steps:
step 1, reading in the whole text file for transmission;
step 2, word segmentation is carried out on the whole text, the use times of each word in the text are counted, each word is numbered, and words with too low use times are removed;
Step 3, adding characters with special meanings such as start or stop in the whole dictionary;
and step 4, outputting a dictionary.
Step two: word embedding is carried out on the input sentences by utilizing an input sentence embedding module, and position vectors are added; word embedding is carried out on the target sentences by utilizing a target sentence word embedding module, and position vectors are added; obtaining a word vector with a position vector corresponding to an input sentence and a word vector with a position vector corresponding to a target sentence;
word embedding and position coding, which comprises the following specific processes:
Step 1, creating an embedding layer Embedding, sending a statement to be transmitted into the embedding layer Embedding, and converting the statement into a word vector with the mapping dimension;
step 2, adding a position vector, wherein the formula is as follows:
Wherein pos refers to the position of a word in a sentence, the value range is [0, L), L is the length of the sentence, i refers to the dimension number of the subvector, the value range is [0, embedding_dimension/2), embedding _dimension is the embedded dimension, and d model refers to the value of the embedded dimension embedding _dimension of Embedding layers;
Step 3, after obtaining the word vector of each word and the corresponding position vector thereof, adding the word vector and the corresponding position vector to obtain the word vector X Embedding with the position information.
Step three, semantic coding, namely finishing semantic coding according to semantic information of sentences through Encoder layers of Transformrer after the transmitted sentences pass through the semantic coding layers; the specific process is as follows:
Step 1, defining three matrixes W Q,WK,WV, and carrying out three linear changes on the word vector X Embedding obtained in the previous step according to the three matrixes to obtain a query vector Q, a key vector K and a value vector V;
step 2, calculating Self-Attention of the three vectors;
resulting in vector X attention, where d k represents the dimension of the vector;
Step 3, carrying out residual connection, adding the X attention obtained in the previous step with the X Embedding obtained in the second step, and carrying out layer normalization operation on the obtained result to obtain a residual vector X 'attention, wherein X' attention＝layernorm(X+Xattention);
step 4, performing feedforward transmission operation on the obtained residual vector X' attention, and obtaining the vector X by performing two-layer linear mapping and activating with an activating function ReLU hidden
Xhidden＝Linear(ReLU(Linear(Xattention)))；
Step four, sentences with the semantic codes completed are transmitted through a wireless channel;
Step five, semantic decoding: carrying out semantic decoding on statement texts with two denoising modes and a decoding layer of a general transducer, and outputting the statement texts with the decoded statement texts after probability logistic regression processing; the specific process is as follows:
step 1, inputting word embedded vectors of the target sentences obtained in the step two into a multi-head self-attention layer for decoding;
step 2, inputting the coding information received through the wireless channel and the output information of the multi-head self-attention layer in the previous step into the multi-head attention layer for decoding;
Step 3, after the target sentence passes through the multi-head self-attention layer and the multi-head attention layer, a semantic decoding vector is obtained through a feedforward transmission layer;
and 4, carrying out probability logistic regression through a Softmax function, and outputting sentences.
Step six, the receiving party judges whether the received sentence needs to be retransmitted, if so, retransmission information is sent to the sending party;
Step seven, the sender repeats the operation of the step four, and the sentences with the semantic codes completed in the step three are sent into the wireless channel again;
step eight, the receiver combines the retransmission information with the information received before, aggregates the information through a full connection layer, and then repeats the step five until the transmission is successful or the maximum retransmission times are reached;
Specific: step 1, repeating the operation of the step four, and sending the information which is subjected to semantic coding before into a channel again;
step 2, the receiver splices the retransmitted information with the information received before, and sends the information to a full-connection layer for aggregation of the information after the splicing is completed;
and step 3, sending the vector with the aggregated information obtained in the previous step into a decoder, and repeating the operation of the step five.
Referring to fig. 2, a multi-decoder scheme is used for retransmission, and the training process is as follows:
A11, training an encoder and a decoder which can normally transmit under each signal-to-noise ratio;
a12, fixing the encoder in the step A11, and newly creating a second decoder, wherein the input of the second decoder is twice as large as that of the decoder in the step S1, and a dimension integration module is arranged in the second decoder; the dimension integration module is composed of a full-connection layer;
A13, transmitting the coding results of the twice coder, splicing the two coding results after the two coding results pass through a wireless channel, sending the spliced results into a second decoder, integrating the two coding results by utilizing a dimension integration module in the second decoder, and performing corresponding decoding operation on the integrated information by the second decoder;
a14, newly creating an N decoder according to the step A12 and the step A13, wherein N is a natural number greater than 2, and the input of the N decoder is N times of the decoder in the step A11;
A15, training is completed, and in the transmission process, the decoders are started in sequence according to the serial numbers of the decoders according to the number of retransmission required;
Referring to fig. 3, a single decoder mode is adopted for retransmission, and the training process is as follows:
B11, establishing an encoder and a decoder, wherein a dimension integration module is arranged in the decoder, testing performance of the decoder under different retransmission times according to a channel to be transmitted and a text for training, determining the allowed maximum retransmission times N according to the performance, and determining the input dimension of the dimension integration module in the decoder according to the maximum retransmission times N; the input dimension of the dimension integrating module is N+1 times of the output dimension of the encoder;
B12, randomly determining the number of times of retransmission required in the training process, wherein the number of times of retransmission is smaller than the maximum number of times of retransmission, splicing the retransmitted information with the original information, taking the information as an input dimension of a dimension integrating module, and carrying out zero padding operation on the rest empty dimension;
B13, repeating the step B12 to obtain a single decoder which can be used for multiple retransmission.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, or alternatives falling within the spirit and principles of the invention.

Claims (3)

1.一种语义通信的重传方法，其特征在于，包括如下步骤：1. A retransmission method for semantic communication, characterized in that it comprises the following steps: S1、训练重传方式，所述重传方式包括多解码器方式和单解码器方式；S1, training retransmission mode, the retransmission mode includes a multi-decoder mode and a single decoder mode; S2、根据接收双方的共有知识，将双方传输过程中所有可能会用的单词进行编号，创建一个词典；S2. Based on the common knowledge of the receiving parties, all the words that may be used in the transmission process between the two parties are numbered to create a dictionary; S3、利用输入语句嵌入模块对输入语句进行词嵌入，并添加位置向量；利用目标语句词嵌入模块对目标语句进行词嵌入，并添加位置向量；得到输入语句对应的带有位置向量的词向量和目标语句对应的带有位置向量的词向量；S3, using the input sentence embedding module to embed the input sentence and add the position vector; using the target sentence word embedding module to embed the target sentence and add the position vector; obtaining the word vector with the position vector corresponding to the input sentence and the word vector with the position vector corresponding to the target sentence; S4、语义编码：输入语句对应的带有位置向量的词向量经过语义编码层，通过Transformrer的Encoder层完成语义编码，得到语义编码向量；S4, semantic encoding: The word vector with the position vector corresponding to the input sentence passes through the semantic encoding layer, and the semantic encoding is completed through the Encoder layer of Transformrer to obtain the semantic encoding vector; 所述步骤S4中的具体步骤如下：The specific steps in step S4 are as follows: S41、定义三个矩阵，根据三个矩阵，对输入语句对应的带有位置向量的词向量进行三次线性变化，得到查询向量、键向量和值向量；S41, define three matrices, and perform three linear changes on the word vector with the position vector corresponding to the input sentence according to the three matrices to obtain a query vector, a key vector, and a value vector; S42、对查询向量、键向量和值向量进行自注意力Self-Attention的计算，得到注意力向量；S42, performing self-attention calculation on the query vector, the key vector and the value vector to obtain an attention vector; S43、进行残差连接，将注意力向量与输入语句对应的带有位置向量的词向量相加，并将得到的结果进行层归一化操作得到残差向量；S43, performing a residual connection, adding the attention vector to the word vector with the position vector corresponding to the input sentence, and performing a layer normalization operation on the obtained result to obtain a residual vector; S44、将得到的残差向量进行前馈传输操作，通过两层线性映射并用激活函数ReLU进行激活得到语义编码向量；S44, performing a feed-forward transmission operation on the obtained residual vector, and obtaining a semantic coding vector through two-layer linear mapping and activation with an activation function ReLU; S5、将语义编码向量通过无线信道；S5, passing the semantic coding vector through a wireless channel; S6、语义解码：将通过无线信道的语义编码向量，通过Transformer的解码层，进行语义解码，通过概率逻辑回归处理后输出解码后的语义文本；S6, semantic decoding: The semantic coding vector passing through the wireless channel is passed through the decoding layer of Transformer for semantic decoding, and the decoded semantic text is output after probabilistic logistic regression processing; 所述步骤S6中的具体步骤如下：The specific steps in step S6 are as follows: S61、将S3的得到的目标语句对应的带有位置向量的词向量输入多头自注意层进行解码；S61, input the word vector with position vector corresponding to the target sentence obtained in S3 into the multi-head self-attention layer for decoding; S62、将通过无线信道接收的编码信息和上一步中多头自注意层的输出信息输入多头注意层进行解码；S62, inputting the coded information received through the wireless channel and the output information of the multi-head self-attention layer in the previous step into the multi-head attention layer for decoding; S63、在目标语句通过多头自注意层与多头注意层之后通过前馈传输层，得到语义解码向量；S63, after the target sentence passes through the multi-head self-attention layer and the multi-head attention layer, it passes through the feed-forward transmission layer to obtain a semantic decoding vector; S64、通过Softmax函数进行概率逻辑回归，输出语句；S64, perform probabilistic logistic regression through the Softmax function and output the sentence; S7、接收方根据接收到的语义文本进行重传判断；若需要重传，则给发送方发送重传指令，发送方选择重传方式，将步骤S4中的语义编码向量再次送入无线信道，得到重传后的语义编码向量；若无需重传，则传输结束；S7, the receiving party makes a retransmission decision based on the received semantic text; if retransmission is required, a retransmission instruction is sent to the sending party, and the sending party selects a retransmission mode, and sends the semantic coding vector in step S4 to the wireless channel again to obtain the retransmitted semantic coding vector; if retransmission is not required, the transmission ends; S8、将重传后的语义编码向量以及之前接收到的语义编码向量相结合，通过维度整合模块进行聚合得到聚合后的语义编码向量；将聚合后的语义编码向量通过Transformer的解码层，进行语义解码，通过概率逻辑回归处理后输出解码后的语义文本，返回步骤S7；S8, combining the retransmitted semantic coding vector and the previously received semantic coding vector, and aggregating them through the dimension integration module to obtain an aggregated semantic coding vector; passing the aggregated semantic coding vector through the decoding layer of the Transformer for semantic decoding, and outputting the decoded semantic text after probabilistic logistic regression processing, and returning to step S7; 所述多解码器方式包括如下训练过程：The multi-decoder method includes the following training process: A11、训练一个能够在各个信噪比下正常传输的编码器与解码器；A11. Train an encoder and decoder that can transmit normally under various signal-to-noise ratios; A12、固定步骤A11中的编码器，新建第二解码器，所述第二解码器的输入为步骤S1中的解码器的两倍，所述第二解码器中设有维度整合模块；所述维度整合模块由全连接层构成；A12, fix the encoder in step A11, create a second decoder, the input of the second decoder is twice that of the decoder in step S1, and the second decoder is provided with a dimension integration module; the dimension integration module is composed of a fully connected layer; A13、传输两次编码器的编码结果，在通过无线信道后，将两者进行拼接，将拼接后的结果送入第二解码器中，利用第二解码器中的维度整合模块，进行整合，第二解码器对整合后的信息进行相应的解码操作；A13, transmitting the encoding results of the encoder twice, splicing the two after passing through the wireless channel, sending the spliced result to the second decoder, integrating it using the dimension integration module in the second decoder, and the second decoder performing corresponding decoding operations on the integrated information; A14、根据步骤A12与步骤A13，新建第N解码器，所述N为大于2的自然数，所述第N解码器的输入为步骤A11中的解码器的N倍；A14. According to step A12 and step A13, a new Nth decoder is created, wherein N is a natural number greater than 2, and the input of the Nth decoder is N times the input of the decoder in step A11; A15、完成训练，在传输过程中，根据需要重传的次数按照解码器的序号依次启动解码器；A15, after training is completed, during the transmission process, the decoders are started in sequence according to the sequence number of the decoders according to the number of retransmissions required; 所述单解码器方式包括如下训练过程：The single decoder method includes the following training process: B11、建立一个编码器和解码器，所述解码器中设有维度整合模块，根据需要传输的信道以及训练用到的文本，测试解码器在不同重传次数下性能的表现，根据性能表现，决定所允许的最大重传次数N，根据最大重传次数N，决定解码器中维度整合模块的输入维度；所述维度整合模块的输入维度为编码器输出维度的N+1倍；B11. Establish an encoder and a decoder, wherein the decoder is provided with a dimension integration module, and the performance of the decoder under different retransmission times is tested according to the transmission channel required and the text used for training, and the maximum number of retransmissions N allowed is determined according to the performance, and the input dimension of the dimension integration module in the decoder is determined according to the maximum number of retransmissions N; the input dimension of the dimension integration module is N+1 times the output dimension of the encoder; B12、在训练过程中，随机决定需要重传次数，所述重传次数小于最大重传次数，并将重传的信息与原始信息进行拼接，作为维度整合模块的输入维度，并将余下的空维度，进行一个补零的操作；B12. During the training process, randomly determine the number of retransmissions required, which is less than the maximum number of retransmissions, and concatenate the retransmitted information with the original information as the input dimension of the dimension integration module, and perform a zero padding operation on the remaining empty dimensions; B13、重复步骤B12，得到能够用于多次重传的单解码器。B13. Repeat step B12 to obtain a single decoder that can be used for multiple retransmissions. 2.如权利要求1所述的一种语义通信的重传方法，其特征在于，所述步骤S2具体包括如下：2. A semantic communication retransmission method according to claim 1, characterized in that step S2 specifically comprises the following: S21、读入整个用于传输的文本文件；S21, read in the entire text file for transmission; S22、对整个文本进行分词处理，统计文本中每个单词的使用次数，给每个单词进行编号，移除那些使用次数过低的单词；S22, performing word segmentation processing on the entire text, counting the number of times each word in the text is used, numbering each word, and removing words with too low a number of times; S23、在整个字典中添加开始或终止的字符；S23, adding a start or end character to the entire dictionary; S24、输出词典。S24. Output dictionary. 3.如权利要求1所述的一种语义通信的重传方法，其特征在于，所述步骤S3中的具体步骤如下：3. A semantic communication retransmission method according to claim 1, characterized in that the specific steps in step S3 are as follows: S31、创建一个嵌入层，将需要传输的语句送入嵌入层，将其转化成为相映维度的词向量；S31. Create an embedding layer, send the sentence to be transmitted into the embedding layer, and convert it into a word vector of corresponding dimension; S32、计算并添加位置向量；S32, calculate and add the position vector; S33、将词向量与位置向量相加，得到拥有位置信息的词向量。S33. Add the word vector to the position vector to obtain a word vector with position information.
CN202210063290.9A 2022-01-20 2022-01-20 A retransmission method for semantic communication Active CN114564963B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210063290.9A CN114564963B (en) 2022-01-20 2022-01-20 A retransmission method for semantic communication

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210063290.9A CN114564963B (en) 2022-01-20 2022-01-20 A retransmission method for semantic communication

Publications (2)

Publication Number Publication Date
CN114564963A CN114564963A (en) 2022-05-31
CN114564963B true CN114564963B (en) 2024-08-30

Family

ID=81712182

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210063290.9A Active CN114564963B (en) 2022-01-20 2022-01-20 A retransmission method for semantic communication

Country Status (1)

Country Link
CN (1) CN114564963B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2026014876A1 (en) * 2024-07-09 2026-01-15 엘지전자 주식회사 Semantic communication

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR101599875B1 (en) * 2008-04-17 2016-03-14 삼성전자주식회사 Method and apparatus for multimedia encoding based on attribute of multimedia content, method and apparatus for multimedia decoding based on attributes of multimedia content
US10097240B2 (en) * 2013-02-19 2018-10-09 Astrolink International, Llc System and method for inferring schematic and topological properties of an electrical distribution grid
CN112511172B (en) * 2020-11-11 2023-03-24 山东云海国创云计算装备产业创新中心有限公司 Decoding method, device, equipment and storage medium

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
面向语义通信系统的自适应编码与传输方法研究;周清炀;中国优秀硕士学位论文全文数据库信息科技辑;20240615(第6期);I136-95 *

Also Published As

Publication number Publication date
CN114564963A (en) 2022-05-31

Similar Documents

Publication Publication Date Title
Jiang et al. Deep source-channel coding for sentence semantic transmission with HARQ
CN109522403B (en) A Method of Abstract Text Generation Based on Fusion Coding
EP3602415B1 (en) Stop code tolerant image compression neural networks
CN110278002A (en) Polar Code Belief Propagation List Decoding Method Based on Bit Flip
CN114817565A (en) Semantic communication method and system based on knowledge graph
CN109842418A (en) A kind of polarization code belief propagation interpretation method based on bit reversal
CN113609157B (en) Language conversion model training, language conversion method, device, equipment and medium
CN115761758A (en) Semantic communication method, device and system for image classification task
CN114461816A (en) Realization method of information supplement semantic communication system based on knowledge graph
CN114154508B (en) A semantic communication system and method based on Transformer
CN108540267A (en) A kind of multi-user data information detecting method and device based on deep learning
CN114564963B (en) A retransmission method for semantic communication
CN114564962A (en) Semantic communication code rate control method based on Transformer
CN1312937A (en) Method and device for masking errors
CN112953565B (en) Return-to-zero convolutional code decoding method and system based on convolutional neural network
CN118333011A (en) Semantic communication method and system based on semantic base
CN111464267B (en) Communication data checking method and device, computer equipment and storage medium
CN112953552A (en) Q-learning assisted successive cancellation rollover decoder and decoding method thereof
CN114548073B (en) A denoising method based on semantic communication system
CN112036122A (en) Text recognition method, electronic device and computer readable medium
CN119210664B (en) Semantic awareness HARQ communication method, semantic awareness HARQ communication device, electronic equipment and storage medium
CN118153678B (en) Semantic communication method, system, equipment and medium based on graphic-text associated knowledge network
CN119449236A (en) Semantic communication system, method and related equipment
CN115085780B (en) Offshore high-reliability end-to-end communication self-encoder design method based on channel feedback
CN108199807B (en) Polarization code reliability estimation method

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant