CN115149986B - Channel diversity method and device for semantic communication - Google Patents
Channel diversity method and device for semantic communication Download PDFInfo
- Publication number
- CN115149986B CN115149986B CN202210589770.9A CN202210589770A CN115149986B CN 115149986 B CN115149986 B CN 115149986B CN 202210589770 A CN202210589770 A CN 202210589770A CN 115149986 B CN115149986 B CN 115149986B
- Authority
- CN
- China
- Prior art keywords
- semantic
- result
- semantic symbol
- different
- data
- 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
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B7/00—Radio transmission systems, i.e. using radiation field
- H04B7/02—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas
- H04B7/04—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas
- H04B7/0413—MIMO systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B7/00—Radio transmission systems, i.e. using radiation field
- H04B7/02—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas
- H04B7/04—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas
- H04B7/06—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B7/00—Radio transmission systems, i.e. using radiation field
- H04B7/02—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas
- H04B7/04—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas
- H04B7/08—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the receiving station
-
- 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)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Radio Transmission System (AREA)
Abstract
The invention provides a channel diversity method and device for semantic communication, and relates to the fields of communication technology and computer technology. Comprising the following steps: deploying a plurality of different encoding and decoding pairs at a transmitting end and a receiving end; inputting single data to be transmitted into different encoders to obtain a plurality of different semantic symbol vectors; the sending end sends the obtained semantic symbol vectors to the receiving end through a sending device respectively; after physical channel transmission, the receiving end obtains semantic symbol vectors overlapped with noise interference; inputting semantic symbol vectors sent by a certain encoder into a decoder of a pair to obtain a signal recovery or intelligent reasoning result; and combining the results output by the encoders to obtain a final global result. The semantic communication channel diversity method provided by the invention can comprehensively utilize the diversity gain of semantic channels between different encoding and decoding pairs, greatly enhance the reliability of the semantic communication process and improve the quality of a data recovery result or an intelligent reasoning result.
Description
Technical Field
The invention relates to the field of communication technology and computer technology, in particular to a channel diversity method and device for semantic communication.
Background
Channel diversity (or spatial diversity) is a method of enhancing reliability of data transmission in a wireless communication system. Specifically, to obtain diversity gain, the communication system is designed as a multiple-input multiple-output (MIMO) architecture, that is, the transmitting end and the receiving end of the communication system are provided with multiple transceiving antennas. When transmitting data, all antennas of the transmitting end simultaneously transmit the same data, all antennas of the receiving end simultaneously receive the transmitted data, and finally the transmitted data are decoded through a certain combination rule. Since the diversity technique can utilize a plurality of relatively independent wireless transmission channels between the transmitting-receiving antennas, higher transmission reliability can be achieved than one antenna pair. This improvement in performance achieved with different transmission paths is referred to as diversity gain.
Semantic communication is an end-to-end communication technique that utilizes deep learning techniques to replace traditional communications. In semantic communication, the functions of a transmitting end and a receiving end are respectively realized by an encoder and a decoder based on deep learning, wherein the functions of source coding, channel coding, modulation and the like can be realized, and the functions of demodulation, channel decoding, source decoding and the like can be realized by the receiving end. Compared with the traditional communication technology, the semantic communication has stronger data compression capability and noise immunity, and is a key support technology in future 6G and other communication systems.
At present, the technical research on semantic communication is still in a starting stage, and the current semantic communication device can realize a point-to-point communication process, but does not have the capability of recycling semantic channels, so that the diversity gain of the semantic channels cannot be utilized to further enhance the reliability of the communication process.
Disclosure of Invention
Aiming at the problem that the prior art does not have the capability of recycling semantic channels and cannot utilize the diversity gain of the semantic channels to further enhance the reliability of the communication process, the invention provides a channel diversity method and device aiming at semantic communication.
In order to solve the technical problems, the invention provides the following technical scheme:
in one aspect, a method of channel diversity for semantic communications is provided, the method being applied to an electronic device, comprising the steps of:
s1: respectively deploying a plurality of different encoding and decoding pair EDP models at a transmitting end and a receiving end of the semantic communication system;
s2: acquiring single data to be transmitted, inputting the data to be transmitted into different encoders, and acquiring a plurality of different initial semantic symbol vectors;
s3: the plurality of different initial semantic symbol vectors are respectively sent to the receiving end through a sending device of the sending end; the receiving end obtains semantic symbol vectors overlapped with noise interference;
s4: selecting an initial semantic symbol vector in an encoder, and inputting the initial semantic symbol vector into a decoder paired with the encoder to obtain a signal recovery result or an intelligent reasoning result; and combining the results output by the encoders to obtain a final global result.
Optionally, in step S1, a plurality of different codec pair EDP models are deployed at a transmitting end and a receiving end of the semantic communication system, including:
s11: training data are prepared, and N different EDP models are trained at a sending end and a receiving end of a semantic communication system;
s12: simulating the data transmission capacity of each EDP model under different signal-to-noise ratio conditions to obtain the quality of the result obtained by each EDP model under different SNR conditions;
s13: and obtaining an influence relation model of the signal-to-noise ratio SNR of the physical channel on the quality of the EDP output result by fitting or training a prediction model.
Optionally, the EDP model is designed in a targeted manner according to practical application requirements, including: deep learning model, convolutional neural network, recurrent neural network, attention mechanism, fully connected neural network.
Optionally, the evaluation model of the transmission capability of the EDP model under different signal-to-noise ratio SNR conditions can be flexibly designed according to actual application requirements.
Optionally, in step S3, the plurality of different initial semantic symbol vectors are sent to the receiving end through a sending device of the sending end respectively; the receiving end obtains a semantic symbol vector overlapped with noise interference, which comprises the following steps:
transmitting the obtained multiple different initial semantic symbol vectors to a receiving end through a physical channel; the receiving end obtains semantic symbol vectors superimposed with noise interference and records the signal-to-noise ratio SN corresponding to each initial semantic symbol vectorR is represented as gamma 1 ,...,γ N 。
Optionally, according to the semantic multiple access capability of the semantic communication system, the delivering modes of the N initial semantic symbol vectors include: and sending the mixed semantic symbol vectors in a split mode or combining the mixed semantic symbol vectors into a mixed semantic symbol vector and sending the mixed semantic symbol vector.
Optionally, in step S4, an initial semantic symbol vector in an encoder is selected, and the initial semantic symbol vector is input into a decoder paired with the encoder, so as to obtain a signal recovery result or an intelligent reasoning result; combining the results output by the encoders to obtain a final global result, including:
s41: selecting an initial semantic symbol vector in an encoder, and inputting the initial semantic symbol vector into a decoder paired with the encoder; each decoder obtains a data recovery result or an inference result according to the initial semantic symbol vector sent by the paired encoder; after decoding is completed, N results are obtained in total, which are expressed as x 1 ,...,x N ；
S42: predicting the quality of the result obtained by each decoder based on the signal-to-noise ratio SNR, the prediction value of the variance between the result output by the ith decoder and the original data being
Wherein V (·) is a quality prediction function;
s43: all results x are obtained through a preset fusion rule 1 ,...,x N And are fused into a single global result.
Optionally, in step S43, a corresponding fusion method is designed according to actual needs, including: data level fusion, feature level fusion and decision level fusion.
In one aspect, there is provided a channel diversity apparatus for semantic communication, the apparatus being applied to an electronic device, the apparatus comprising:
the model deployment module is used for deploying a plurality of different encoding and decoding pair EDP models at a transmitting end and a receiving end of the semantic communication system respectively;
the initial coding module is used for acquiring single data to be transmitted, inputting the data to be transmitted into different encoders and acquiring a plurality of different initial semantic symbol vectors;
the data transmitting module is used for transmitting the plurality of different initial semantic symbol vectors to the receiving end through a transmitting device of the transmitting end respectively; the receiving end obtains semantic symbol vectors overlapped with noise interference;
the global decoding fusion module is used for selecting an initial semantic symbol vector in one encoder, inputting the initial semantic symbol vector into a decoder paired with the encoder, and obtaining a signal recovery result or an intelligent reasoning result; and combining the results output by the encoders to obtain a final global result.
Optionally, the model deployment module is further configured to:
training data are prepared, and N different EDP models are trained at a sending end and a receiving end of a semantic communication system;
simulating the data transmission capacity of each EDP model under different signal-to-noise ratio conditions to obtain the quality of the result obtained by each EDP model under different SNR conditions;
and obtaining an influence relation model of the signal-to-noise ratio SNR of the physical channel on the quality of the EDP output result by fitting or training a prediction model.
In one aspect, an electronic device is provided that includes a processor and a memory having at least one instruction stored therein, the at least one instruction loaded and executed by the processor to implement a channel diversity method for semantic communications as described above.
In one aspect, a computer-readable storage medium having stored therein at least one instruction that is loaded and executed by a processor to implement a channel diversity method for semantic communications as described above is provided.
The technical scheme provided by the embodiment of the invention has at least the following beneficial effects:
in the scheme, the channel diversity method for semantic communication comprehensively utilizes the diversity gain of semantic channels among different encoding and decoding pairs, thereby greatly enhancing the reliability of the semantic communication process, and improving the quality of data recovery results or intelligent reasoning results, such as image definition, voice definition, text semantic similarity, classification result accuracy, target detection accuracy and the like.
Unlike existing communication diversity techniques, the present invention does not exploit the diversity of multiple physical transmission channels to achieve diversity gain, but rather exploits the diversity of multiple semantic channels. The difference is that the physical channel refers to the physical transmission medium on which the electromagnetic signal propagation in the physical world depends, while the semantic channel includes the data conversion process in the semantic layer and the signal propagation process in the physical layer. The data conversion at the semantic level refers to the conversion and processing process of the semantic communication deep learning model on the data. In this process, the original data is embedded with noise at the semantic level, and even if the physical channel is perfectly noiseless, the semantic channel may have random noise, which reduces the quality of the final transmission result to some extent.
In a specific implementation manner, the conventional communication diversity technology utilizes space-time diversity of physical channels to realize diversity, such as Multi-input Multi-output (MIMO) technology based on multiple physical antennas, data retransmission technology using different transmission time slots, and the like. The invention realizes the diversity of the semantic channel layer by utilizing the difference of the deep learning model, and different transmission branches are realized by using different semantic communication models, and finally, all branch results are combined by utilizing the fusion rule so as to enhance the quality of the data transmission results. In practice, tests show that the physical channel diversity technology has very limited gain for the semantic communication scene, but the invention is more suitable for the semantic communication scene and can effectively enhance the data transmission quality. For image data, the invention can obtain peak signal-to-noise ratio (Peak signal to noise ratio, PSNR) gain exceeding 3dB under the scene of lower signal-to-noise ratio.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a channel diversity method for semantic communications provided by an embodiment of the present invention;
FIG. 2 is a flow chart of a method of channel diversity for semantic communications provided by an embodiment of the present invention;
fig. 3 is a schematic diagram of a transmitting end and a receiving end of a channel diversity method for semantic communication according to an embodiment of the present invention;
FIG. 4 is a block diagram of a channel diversity apparatus for semantic communications provided by an embodiment of the present invention;
fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
In order to make the technical problems, technical solutions and advantages to be solved more apparent, the following detailed description will be given with reference to the accompanying drawings and specific embodiments.
The embodiment of the invention provides a channel diversity method for semantic communication, which can be realized by electronic equipment, wherein the electronic equipment can be a terminal or a server. As shown in fig. 1, a flow chart of a channel diversity method for semantic communication, the processing flow of the method may include the following steps:
s101: respectively deploying a plurality of different EDP (Encoder decoder pair, codec pair) models at a transmitting end and a receiving end of the semantic communication system;
s102: acquiring single data to be transmitted, inputting the data to be transmitted into different encoders to acquire a plurality of different initial SSVs (Semantic symbol vector, semantic symbol vectors);
s103: the plurality of different initial semantic symbol vectors are respectively sent to the receiving end through a sending device of the sending end; the receiving end obtains semantic symbol vectors overlapped with noise interference;
s104: selecting an initial semantic symbol vector in an encoder, and inputting the initial semantic symbol vector into a decoder paired with the encoder to obtain a signal recovery result or an intelligent reasoning result; and combining the results output by the encoders to obtain a final global result.
Optionally, in step S101, a plurality of different codec pair EDP models are deployed at a transmitting end and a receiving end of the semantic communication system, including:
s111: training data are prepared, and N different EDP models are trained at a sending end and a receiving end of a semantic communication system;
s112: simulating the data transmission capacity of each EDP model under different signal-to-noise ratio conditions to obtain the quality of the result obtained by each EDP model under different SNR (Signal to noise ratio, signal-to-noise ratio) conditions;
s113: and obtaining an influence relation model of the signal-to-noise ratio SNR of the physical channel on the quality of the EDP output result by fitting or training a prediction model.
Optionally, the EDP model is designed in a targeted manner according to practical application requirements, including: deep learning model, convolutional neural network, recurrent neural network, attention mechanism, fully connected neural network.
Optionally, the evaluation model of the transmission capability of the EDP model under different signal-to-noise ratio SNR conditions can be flexibly designed according to actual application requirements.
Optionally, in step S103, the plurality of different initial semantic symbol vectors are sent to the receiving end through a sending device of the sending end respectively; the receiving end obtains a semantic symbol vector overlapped with noise interference, which comprises the following steps:
transmitting the obtained multiple different initial semantic symbol vectors to a receiving end through a physical channel; the receiving end obtains semantic symbol vectors superimposed with noise interference, records the signal-to-noise ratio SNR corresponding to each initial semantic symbol vector, and represents gamma 1 ,...,γ N 。
Optionally, according to the semantic multiple access capability of the semantic communication system, the delivering modes of the N initial semantic symbol vectors include: and sending the mixed semantic symbol vectors in a split mode or combining the mixed semantic symbol vectors into a mixed semantic symbol vector and sending the mixed semantic symbol vector.
Optionally, in step S104, an initial semantic symbol vector in an encoder is selected, and the initial semantic symbol vector is input into a decoder paired with the encoder, so as to obtain a signal recovery result or an intelligent reasoning result; combining the results output by the encoders to obtain a final global result, including:
s141: selecting an initial semantic symbol vector in an encoder, and inputting the initial semantic symbol vector into a decoder paired with the encoder; each decoder obtains a data recovery result or an inference result according to the initial semantic symbol vector sent by the paired encoder; after decoding is completed, N results are obtained in total, which are expressed as x 1 ,...,x N ；
S142: predicting the quality of the result obtained by each decoder based on the signal-to-noise ratio SNR, the prediction value of the variance between the result output by the ith decoder and the original data being
Wherein V (·) is a quality prediction function;
s143: all results x are obtained through a preset fusion rule 1 ,...,x N And are fused into a single global result.
Optionally, in step S143, a corresponding fusion method is designed according to actual needs, including: data level fusion, feature level fusion and decision level fusion.
In the embodiment of the invention, the channel diversity method for semantic communication comprehensively utilizes the diversity gain of semantic channels between different encoding and decoding pairs, thereby greatly enhancing the reliability of the semantic communication process and improving the quality of a data recovery result or an intelligent reasoning result. Unlike existing communication diversity techniques, the present invention does not exploit the diversity of multiple physical transmission channels to achieve diversity gain, but rather exploits the diversity of multiple semantic channels. The difference is that the physical channel refers to the physical transmission medium on which the electromagnetic signal propagation in the physical world depends, while the semantic channel includes the data conversion process in the semantic layer and the signal propagation process in the physical layer. The data conversion at the semantic level refers to the conversion and processing process of the semantic communication deep learning model on the data. In this process, the original data is embedded with noise at the semantic level, and even if the physical channel is perfectly noiseless, the semantic channel may have random noise, which reduces the quality of the final transmission result to some extent.
The embodiment of the invention provides a channel diversity method for semantic communication, which can be realized by electronic equipment, wherein the electronic equipment can be a terminal or a server. Taking video transmission or picture transmission as an example, a transmitting end firstly encodes image data into semantic symbol vectors (or semantic symbols) by using an encoder, and then transmits the semantic symbols to a receiving end through a transmitting device. And the receiving end recovers the target image by using a decoder according to the received data, and completes one image transmission task.
As shown in fig. 2, a flow chart of a channel diversity method for semantic communication, the process flow of the method may include the following steps:
s201: preparing training data, such as video, image, voice and the like, and training N different EDP models at a transmitting end and a receiving end of the semantic communication system;
s202: and simulating the data transmission capacity of each EDP model under different signal-to-noise ratio conditions, and obtaining the quality of the result obtained by each EDP model under different SNR conditions.
In a possible implementation, the model capability test: the data transmission capabilities of the various EDPs under different signal-to-noise ratios (Signal to noise ratio, SNR) were simulated and the quality of the results obtained for each EDP under different SNR conditions was obtained.
For example, if the task is to send image data, the variance between the recovered image data and the real image data at the transmitting end can be used as a quality index, denoted as σ 2 V (SNR), where V (SNR) represents a variable that is a function of SNR. The smaller the variance, the higher the image data recovery quality, and vice versa, the worse.
S203: and obtaining an influence relation model of the signal-to-noise ratio SNR of the physical channel on the quality of the EDP output result by fitting or training a prediction model.
In a possible embodiment, a resulting quality model is obtained: and obtaining an influence relation model of the SNR of the physical channel on the quality of the EDP output result by fitting or training a prediction model. For example, for data transmission tasks, when variance data samples are sufficient, then a predictive model of V (SNR) can be obtained by fitting. The quality of the EDP output result can be directly invoked in the subsequent practical process.
In a possible implementation manner, the EDP model is designed in a targeted manner according to practical application requirements, and the EDP model comprises the following components: deep learning model, convolutional neural network, recurrent neural network, attention mechanism, fully connected neural network.
In a feasible implementation mode, an evaluation model of the transmission capacity of the EDP model under different signal-to-noise ratio (SNR) conditions can be flexibly designed according to practical application requirements.
S204: acquiring single data to be transmitted, inputting the data to be transmitted into different encoders, and acquiring a plurality of different initial semantic symbol vectors;
in a possible implementation manner, according to the semantic multiple access capability of the semantic communication system, the delivering modes of the N initial semantic symbol vectors include: and sending the mixed semantic symbol vectors in a split mode or combining the mixed semantic symbol vectors into a mixed semantic symbol vector and sending the mixed semantic symbol vector. But provided that the semantic communication system has semantic multiple access capability.
S205: the plurality of different initial semantic symbol vectors are respectively sent to the receiving end through a sending device of the sending end; the receiving end obtains semantic symbol vectors overlapped with noise interference;
in a possible implementation manner, the obtained multiple different initial semantic symbol vectors are sent to a receiving end through a physical channel; the receiving end obtains semantic symbol vectors superimposed with noise interference, records the signal-to-noise ratio SNR corresponding to each initial semantic symbol vector, and represents gamma 1 ,...,γ N 。
In a possible implementation, if the semantic communication system has multiple access capability, only one hybrid SSV may be sent, so the sender only needs to send data once, where the physical channel SNR of all SSVs is the same.
In a possible implementation, as shown in fig. 3, the present invention requires that the sender and receiver of the semantic communication system have N codec pairs (Encoder decoder pair, EDP) at the same time, and that the decoder in each EDP can only process the semantic symbol vectors (Semantic symbol vector, SSV) sent by its paired encoder. In this scenario, the semantic channel includes the whole process from the data input to the encoder and to the decoder output, mainly including the two parts of the computational reasoning process of the EDP and the transmission process of the physical channel. The core process is as follows: when transmitting single data, the invention requires the transmitting end and the receiving end to repeatedly transmit the same data by using a plurality of different EDPs, and fuses the results obtained by different decoders into a final transmission result. The process utilizes different semantic channels to transmit data, and can obtain diversity gain of the semantic channels.
S206: selecting an initial semantic symbol vector in an encoder, and inputting the initial semantic symbol vector into a decoder paired with the encoder; each decoder obtains a data recovery result or an inference result according to the initial semantic symbol vector sent by the paired encoder; after decoding is completed, N results are obtained in total, which are expressed as x 1 ,...,x N 。
S207: predicting the quality of the result obtained by each decoder based on the signal-to-noise ratio SNR, the prediction value of the variance between the result output by the ith decoder and the original data being
Where V (·) is the quality prediction function.
In a possible embodiment, the quality of the result obtained by each decoder is predicted from the SNR. For example, for a data transfer task, where the decoder outputs a result that is a recovery of the original transmitted data, then the i-th decoder outputs a prediction of the variance between the result and the original dataThe value is
Wherein V (·) is the quality prediction function, obtained from the example given in step four of the model preparation phase.
S208: all results x are obtained through a preset fusion rule 1 ,...,x N And are fused into a single global result.
In one possible implementation, for the data transmission task, when the SNRs are consistent or relatively similar, the average value can be directly calculated, namely
If the difference of SNR of different results is large, a weighted fusion mode can be adopted, namely +.>
Wherein w is i Is x i Weight of (2) is generally +.>
At this time, if a variance prediction result between the output result of the decoder and the original data is available, it is expressed as +.>
Then a minimum variance fusion rule is used, and the weight can be calculated by +.>
The fused result is the final result.
In a possible implementation manner, the fusion method given by the foregoing embodiment is only an alternative solution for a data transmission task, and a corresponding fusion method may be designed according to actual needs, where the selecting includes: data level fusion, feature level fusion and decision level fusion.
In the embodiment of the invention, the channel diversity method for semantic communication comprehensively utilizes the diversity gain of semantic channels between different encoding and decoding pairs, thereby greatly enhancing the reliability of the semantic communication process and improving the quality of a data recovery result or an intelligent reasoning result. For example, when the data transmission task is video or picture transmission, the present invention can effectively enhance the sharpness of the received image, especially in scenes where the physical channel conditions are relatively poor.
Unlike existing communication diversity techniques, the present invention does not exploit the diversity of multiple physical transmission channels to achieve diversity gain, but rather exploits the diversity of multiple semantic channels. The difference is that the physical channel refers to the physical transmission medium on which the electromagnetic signal propagation in the physical world depends, while the semantic channel includes the data conversion process in the semantic layer and the signal propagation process in the physical layer. The data conversion at the semantic level refers to the conversion and processing process of the semantic communication deep learning model on the data. In this process, the original data is embedded with noise at the semantic level, and even if the physical channel is perfectly noiseless, the semantic channel may have random noise, which reduces the quality of the final transmission result to some extent.
In a specific implementation manner, the conventional communication diversity technology utilizes space-time diversity of physical channels to realize diversity, such as Multi-input Multi-output (MIMO) technology based on multiple physical antennas, data retransmission technology using different transmission time slots, and the like. The invention realizes the diversity of the semantic channel layer by utilizing the difference of the deep learning model, and different transmission branches are realized by using different semantic communication models, and finally, all branch results are combined by utilizing the fusion rule so as to enhance the quality of the data transmission results. In practice, tests show that the physical channel diversity technology has very limited gain for the semantic communication scene, but the invention is more suitable for the semantic communication scene and can effectively enhance the data transmission quality. For image data, the invention can obtain peak signal-to-noise ratio (Peak signal to noise ratio, PSNR) gain exceeding 3dB under the scene of lower signal-to-noise ratio.
Fig. 4 is a block diagram of a channel diversity apparatus for semantic communications according to an example embodiment. Referring to fig. 4, the apparatus 300 includes:
the model deployment module 310 is configured to deploy a plurality of different codec pair EDP models at a transmitting end and a receiving end of the semantic communication system respectively;
the initial encoding module 320 is configured to obtain a single data to be sent, input the data to be sent into different encoders, and obtain a plurality of different initial semantic symbol vectors;
a data sending module 330, configured to send the plurality of different initial semantic symbol vectors to the receiving end through a sending device of the sending end; the receiving end obtains semantic symbol vectors overlapped with noise interference;
the global decoding fusion module 340 is configured to select an initial semantic symbol vector in an encoder, input the initial semantic symbol vector into a decoder paired with the encoder, and obtain a signal recovery result or an intelligent reasoning result; and combining the results output by the encoders to obtain a final global result.
Optionally, the model deployment module 310 is further configured to: training data are prepared, and N different EDP models are trained at a sending end and a receiving end of a semantic communication system;
simulating the data transmission capacity of each EDP model under different signal-to-noise ratio conditions to obtain the quality of the result obtained by each EDP model under different SNR conditions;
and obtaining an influence relation model of the signal-to-noise ratio SNR of the physical channel on the quality of the EDP output result by fitting or training a prediction model.
Optionally, the EDP model is designed in a targeted manner according to practical application requirements, including: deep learning model, convolutional neural network, recurrent neural network, attention mechanism, fully connected neural network.
Optionally, the evaluation model of the transmission capability of the EDP model under different signal-to-noise ratio SNR conditions can be flexibly designed according to actual application requirements.
Optionally, the data sending module 330 is further configured to send the obtained multiple different initial semantic symbol vectors to the receiving end through a physical channel; the receiving end obtains semantic symbol vectors superimposed with noise interference, records the signal-to-noise ratio SNR corresponding to each initial semantic symbol vector, and represents gamma 1 ,...,γ N 。
Optionally, according to the semantic multiple access capability of the semantic communication system, the delivering modes of the N initial semantic symbol vectors include: and sending the mixed semantic symbol vectors in a split mode or combining the mixed semantic symbol vectors into a mixed semantic symbol vector and sending the mixed semantic symbol vector.
Optionally, the step global decoding fusion module 340 is further configured to select an initial semantic symbol vector in an encoder, and input the initial semantic symbol vector into a decoder paired with the encoder; each decoder obtains a data recovery result or an inference result according to the initial semantic symbol vector sent by the paired encoder; after decoding is completed, N results are obtained in total, which are expressed as x 1 ,...,x N ；
Predicting the quality of the result obtained by each decoder based on the signal-to-noise ratio SNR, the prediction value of the variance between the result output by the ith decoder and the original data being
Wherein V (·) is a quality prediction function;
all results x are obtained through a preset fusion rule 1 ,...,x N And are fused into a single global result.
Optionally, the step global decoding fusion module 340 is further configured to design a corresponding fusion method according to actual needs, including: data level fusion, feature level fusion and decision level fusion.
In the embodiment of the invention, the channel diversity method for semantic communication comprehensively utilizes the diversity gain of semantic channels between different encoding and decoding pairs, thereby greatly enhancing the reliability of the semantic communication process and improving the quality of a data recovery result or an intelligent reasoning result.
Fig. 5 is a schematic structural diagram of an electronic device 400 according to an embodiment of the present invention, where the electronic device 400 may have a relatively large difference due to different configurations or performances, and may include one or more processors (central processing units, CPU) 401 and one or more memories 402, where at least one instruction is stored in the memories 402, and the at least one instruction is loaded and executed by the processors 401 to implement the following steps of a channel diversity method for semantic communication:
s1: respectively deploying a plurality of different encoding and decoding pair EDP models at a transmitting end and a receiving end of the semantic communication system;
s2: acquiring single data to be transmitted, inputting the data to be transmitted into different encoders, and acquiring a plurality of different initial semantic symbol vectors;
s3: the plurality of different initial semantic symbol vectors are respectively sent to the receiving end through a sending device of the sending end; the receiving end obtains semantic symbol vectors overlapped with noise interference;
s4: selecting an initial semantic symbol vector in an encoder, and inputting the initial semantic symbol vector into a decoder paired with the encoder to obtain a signal recovery result or an intelligent reasoning result; and combining the results output by the encoders to obtain a final global result.
In an exemplary embodiment, a computer readable storage medium, e.g., a memory comprising instructions executable by a processor in a terminal to perform the above-described channel diversity method for semantic communications is also provided. For example, the computer readable storage medium may be ROM, random Access Memory (RAM), CD-ROM, magnetic tape, floppy disk, optical data storage device, etc.
It will be understood by those skilled in the art that all or part of the steps for implementing the above embodiments may be implemented by hardware, or may be implemented by a program for instructing relevant hardware, where the program may be stored in a computer readable storage medium, and the storage medium may be a read-only memory, a magnetic disk or an optical disk, etc.
The foregoing description of the preferred embodiments of the invention is not intended to limit the invention to the precise form disclosed, and any such modifications, equivalents, and alternatives falling within the spirit and scope of the invention are intended to be included within the scope of the invention.
Claims (7)
1. A channel diversity method for semantic communications, comprising the steps of:
s1: respectively deploying a plurality of different encoding and decoding pair EDP models at a transmitting end and a receiving end of the semantic communication system;
in the step S1, a plurality of different codec pair EDP models are deployed at a transmitting end and a receiving end of the semantic communication system, respectively, including:
s11: training data are prepared, and N different EDP models are trained at a sending end and a receiving end of a semantic communication system;
s12: simulating the data transmission capacity of each EDP model under different signal-to-noise ratio (SNR) conditions to obtain the quality of a result obtained by each EDP model under different SNR conditions;
s13: obtaining an influence relation model of the signal-to-noise ratio SNR of the physical channel on the quality of the EDP output result by fitting or training a prediction model;
s2: acquiring single data to be transmitted, inputting the data to be transmitted into different encoders, and acquiring a plurality of different initial semantic symbol vectors;
s3: the plurality of different initial semantic symbol vectors are respectively sent to the receiving end through a sending device of the sending end; the receiving end obtains semantic symbol vectors overlapped with noise interference;
s4: selecting an initial semantic symbol vector in an encoder, and inputting the initial semantic symbol vector into a decoder paired with the encoder to obtain a signal recovery result or an intelligent reasoning result; combining the results output by the decoders to obtain a final global result;
in the step S4, an initial semantic symbol vector in an encoder is selected, and the initial semantic symbol vector is input into a decoder paired with the encoder to obtain a signal recovery result or an intelligent reasoning result; combining the results output by the decoders to obtain a final global result, including:
s41: selecting an initial semantic symbol vector in an encoder, and inputting the initial semantic symbol vector into a decoder paired with the encoder; each decoder obtains a data recovery result or an inference result according to the initial semantic symbol vector sent by the paired encoder; after decoding is completed, N results are obtained in total, which are expressed as x 1 ,...,x N ；
S42: predicting the quality of the result obtained by each decoder according to the signal-to-noise ratio SNR, wherein the predicted value of the variance between the result output by the ith decoder and the data to be transmitted at the transmitting end is
Wherein V (·) is a quality prediction function; i is a positive integer between 1 and N; the signal-to-noise ratio SNR corresponding to each initial semantic symbol vector is denoted as gamma 1 ,...,γ N ；
S43: all results x are obtained through a preset fusion rule 1 ,...,x N And are fused into a single global result.
2. The method of claim 1, wherein the EDP model comprises: deep learning model, convolutional neural network, cyclic neural network, attention mechanism, fully connected neural network.
3. The method of claim 1, wherein different EDP model transmission capability assessment models are designed based on different signal-to-noise ratio SNR conditions.
4. The method according to claim 3, wherein in S3, the plurality of different initial semantic symbol vectors are sent to the receiving end by a sending device of the sending end, respectively; the receiving end obtains a semantic symbol vector overlapped with noise interference, which comprises the following steps:
transmitting the obtained multiple different initial semantic symbol vectors to a receiving end through a physical channel; the receiving end obtains semantic symbol vectors superimposed with noise interference, records the signal-to-noise ratio SNR corresponding to each initial semantic symbol vector, and represents gamma 1 ,...,γ N 。
5. The method of claim 4, wherein the transmitting means of the N initial semantic symbol vectors according to semantic multiple access capability of the semantic communication system comprises: and sending the mixed semantic symbol vectors in a split mode or combining the mixed semantic symbol vectors into a mixed semantic symbol vector and sending the mixed semantic symbol vector.
6. The method according to claim 1, wherein in step S43, a corresponding fusion rule is designed, and the fusion rule includes: data level fusion, feature level fusion and decision level fusion.
7. A channel diversity apparatus for semantic communication, characterized in that the apparatus is adapted for the method of any of the previous claims 1-6, the apparatus comprising:
the model deployment module is used for deploying a plurality of different encoding and decoding pair EDP models at a transmitting end and a receiving end of the semantic communication system respectively;
the model deployment module is further configured to:
training data are prepared, and N different EDP models are trained at a sending end and a receiving end of a semantic communication system;
simulating the data transmission capacity of each EDP model under different signal-to-noise ratio (SNR) conditions to obtain the quality of the result obtained by each EDP model under different SNR conditions;
obtaining an influence relation model of the signal-to-noise ratio SNR of the physical channel on the quality of the EDP output result by fitting or training a prediction model;
the initial coding module is used for acquiring single data to be transmitted, inputting the data to be transmitted into different encoders and acquiring a plurality of different initial semantic symbol vectors;
the data transmitting module is used for transmitting the plurality of different initial semantic symbol vectors to the receiving end through a transmitting device of the transmitting end respectively; the receiving end obtains semantic symbol vectors overlapped with noise interference;
the global decoding fusion module is used for selecting an initial semantic symbol vector in one encoder, inputting the initial semantic symbol vector into a decoder paired with the encoder, and obtaining a signal recovery result or an intelligent reasoning result; combining the results output by the decoders to obtain a final global result;
the global decoding fusion module is also used for selecting an initial semantic symbol vector in one encoder and inputting the initial semantic symbol vector into a decoder paired with the encoder; each decoder obtains a data recovery result or an inference result according to the initial semantic symbol vector sent by the paired encoder; after decoding is completed, N results are obtained in total, which are expressed as x 1 ,...,x N ；
Predicting the quality of the result obtained by each decoder according to the signal-to-noise ratio SNR, wherein the predicted value of the variance between the result output by the ith decoder and the data to be transmitted at the transmitting end is
Wherein V (·) is a quality prediction function, i is a positive integer between 1 and N; the signal-to-noise ratio SNR corresponding to each initial semantic symbol vector is denoted as gamma 1 ,...,γ N ；
All results x are obtained through a preset fusion rule 1 ,...,x N Merging into a single global result;
the global decoding fusion module is further used for designing corresponding fusion rules, wherein the fusion rules comprise: data level fusion, feature level fusion and decision level fusion.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210589770.9A CN115149986B (en) | 2022-05-27 | 2022-05-27 | Channel diversity method and device for semantic communication |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210589770.9A CN115149986B (en) | 2022-05-27 | 2022-05-27 | Channel diversity method and device for semantic communication |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115149986A CN115149986A (en) | 2022-10-04 |
| CN115149986B true CN115149986B (en) | 2023-05-23 |
Family
ID=83405740
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210589770.9A Active CN115149986B (en) | 2022-05-27 | 2022-05-27 | Channel diversity method and device for semantic communication |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115149986B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115470799B (en) * | 2022-11-03 | 2023-04-07 | 电子科技大学 | Text transmission and semantic understanding integrated method for network edge equipment |
| CN116209069B (en) * | 2023-04-25 | 2023-07-21 | 北京邮电大学 | Multiple access method based on semantic domain and related equipment |
| CN116543769B (en) * | 2023-06-08 | 2023-12-29 | 清华大学 | MIMO voice transmission method and system based on semantic perception network |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104113398A (en) * | 2014-07-08 | 2014-10-22 | 西安电子科技大学 | MIMO blind channel estimation fuzziness removal method based on orthogonal space-time block codes |
| CN112019245A (en) * | 2020-08-26 | 2020-12-01 | 上海科技大学 | Method, apparatus, device and medium for predicting and measuring channel information in real time |
| WO2021255605A1 (en) * | 2020-06-19 | 2021-12-23 | Nokia Technologies Oy | Apparatus, method and computer program product for optimizing parameters of a compressed representation of a neural network |
| CN113890795A (en) * | 2021-09-09 | 2022-01-04 | 广州杰赛科技股份有限公司 | Method, device and medium for constructing large-scale MIMO channel estimation model |
| CN114143040A (en) * | 2021-11-08 | 2022-03-04 | 浙江工业大学 | Confrontation signal detection method based on multi-channel feature reconstruction |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10347271B2 (en) * | 2015-12-04 | 2019-07-09 | Synaptics Incorporated | Semi-supervised system for multichannel source enhancement through configurable unsupervised adaptive transformations and supervised deep neural network |
| US10560303B2 (en) * | 2017-07-17 | 2020-02-11 | Cox Communications, Inc. | System and methods for a smart adaptive communication channel |
| WO2021041551A2 (en) * | 2019-08-26 | 2021-03-04 | Board Of Regents, The University Of Texas System | Autoencoder-based error correction coding for low-resolution communication |
| CN110807332B (en) * | 2019-10-30 | 2024-02-27 | 腾讯科技（深圳）有限公司 | Training method, semantic processing method, device and storage medium for semantic understanding model |
| CN112583458B (en) * | 2020-12-08 | 2022-03-01 | 齐鲁工业大学 | MIMO end-to-end transmission system based on deep learning and wireless transformation network |
| CN114448563B (en) * | 2021-12-13 | 2024-03-08 | 北京邮电大学 | Semantic code transmission method and electronic equipment |
| CN114490920B (en) * | 2022-01-21 | 2022-10-14 | 华中科技大学 | Automatic semantic knowledge base construction and updating method, semantic encoder and communication architecture |
-
2022
- 2022-05-27 CN CN202210589770.9A patent/CN115149986B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104113398A (en) * | 2014-07-08 | 2014-10-22 | 西安电子科技大学 | MIMO blind channel estimation fuzziness removal method based on orthogonal space-time block codes |
| WO2021255605A1 (en) * | 2020-06-19 | 2021-12-23 | Nokia Technologies Oy | Apparatus, method and computer program product for optimizing parameters of a compressed representation of a neural network |
| CN112019245A (en) * | 2020-08-26 | 2020-12-01 | 上海科技大学 | Method, apparatus, device and medium for predicting and measuring channel information in real time |
| CN113890795A (en) * | 2021-09-09 | 2022-01-04 | 广州杰赛科技股份有限公司 | Method, device and medium for constructing large-scale MIMO channel estimation model |
| CN114143040A (en) * | 2021-11-08 | 2022-03-04 | 浙江工业大学 | Confrontation signal detection method based on multi-channel feature reconstruction |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115149986A (en) | 2022-10-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115149986B (en) | Channel diversity method and device for semantic communication | |
| Roh et al. | Efficient feedback methods for MIMO channels based on parameterization | |
| Zhang et al. | Multi-user semantic communications for cooperative object identification | |
| CN102388544A (en) | Wireless transmission of layered signals in a relay network | |
| US8831128B2 (en) | MIMO communication system signal detection method | |
| Elbir et al. | Federated learning for physical layer design | |
| CN102144256A (en) | Method and apparatus for fast nearestneighbor search for vector quantizers | |
| CN114765568B (en) | Multiple access method and device for semantic communication system | |
| CN110380762A (en) | A kind of extensive cut-in method that calculating is merged with communication | |
| CN115146125A (en) | Receiving end data filtering method and device under semantic communication multi-address access scene | |
| Tang et al. | Contrastive learning based semantic communications | |
| Zou et al. | Generative adversarial network for wireless communication: Principle, application, and trends | |
| Ye et al. | Bilinear convolutional auto-encoder based pilot-free end-to-end communication systems | |
| CN115695810A (en) | Low bit rate image compression coding method based on semantic communication | |
| CN112802485B (en) | Voice data processing method and device, computer equipment and storage medium | |
| Anjum et al. | Acoustic channel-aware autoencoder-based compression for underwater image transmission | |
| Njoku et al. | BLER performance evaluation of an enhanced channel autoencoder | |
| Qin et al. | Mem-DeepSC: A Semantic Communication System with Memory | |
| CN116543769B (en) | MIMO voice transmission method and system based on semantic perception network | |
| Hu et al. | SemHARQ: Semantic-Aware HARQ for Multi-task Semantic Communications | |
| WO2023283785A1 (en) | Method for processing signal, and receiver | |
| Shao et al. | A DenseNet-Based Learning Framework Toward Maritime End-to-End Autoencoder Communication Systems | |
| US20230078979A1 (en) | A Receiver for a Communication System | |
| Ren et al. | Asymmetric Semantic Communication System Based on Diffusion Model in IoT | |
| CN116938401A (en) | Data transmission method and communication device |
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 |