CN116390134B - Semantic communication transmission method based on non-orthogonal multiple access - Google Patents
Semantic communication transmission method based on non-orthogonal multiple accessInfo
- Publication number
- CN116390134B CN116390134B CN202310419000.4A CN202310419000A CN116390134B CN 116390134 B CN116390134 B CN 116390134B CN 202310419000 A CN202310419000 A CN 202310419000A CN 116390134 B CN116390134 B CN 116390134B
- Authority
- CN
- China
- Prior art keywords
- user
- signal
- semantic
- base station
- signals
- 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
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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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
- H03—ELECTRONIC CIRCUITRY
- H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
- H03M7/00—Conversion of a code where information is represented by a given sequence or number of digits to a code where the same, similar or subset of information is represented by a different sequence or number of digits
- H03M7/30—Compression; Expansion; Suppression of unnecessary data, e.g. redundancy reduction
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W24/00—Supervisory, monitoring or testing arrangements
- H04W24/02—Arrangements for optimising operational condition
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W28/00—Network traffic management; Network resource management
- H04W28/02—Traffic management, e.g. flow control or congestion control
- H04W28/06—Optimizing the usage of the radio link, e.g. header compression, information sizing, discarding information
- H04W28/065—Optimizing the usage of the radio link, e.g. header compression, information sizing, discarding information using assembly or disassembly of packets
-
- 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)
- Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Computer Networks & Wireless Communication (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Signal Processing (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
本发明公开了一种基于非正交多址接入的语义通信传输方法，基站基于文本数据集构建语义背景知识库，并通过广播分享给网络中的多个用户；在基站和用户侧分别设计基于Transformer结构的发射器和接收器，实现用户端语义信息提取与基站端的语义信息恢复；设计非正交多址接入语义通信系统，利用语义辅助执行串行干扰消除；收集用户的发送信号和基站的接收信号，根据交叉熵损失训练网络模型；基于训练的模型参数，结合贪婪算法和语义解码器恢复用户数据，实现语义信息传输。该方法以语义信息辅助消除干扰实现精确重构用户信号，减少误差传播，并更新基站侧的用户解码器网络参数，保证每名用户的传输准确性，同时也能减少基站的处理量和消耗。
The present invention discloses a semantic communication transmission method based on non-orthogonal multiple access. The base station constructs a semantic background knowledge base based on a text dataset and shares it with multiple users in the network via broadcast. Transformer-based transmitters and receivers are designed on the base station and user sides, respectively, to extract semantic information on the user side and recover it on the base station side. A non-orthogonal multiple access semantic communication system is designed, which uses semantics to assist in performing serial interference cancellation. The user's transmitted signals and the base station's received signals are collected, and a network model is trained using cross-entropy loss. Based on the trained model parameters, a greedy algorithm and a semantic decoder are combined to recover user data and achieve semantic information transmission. This method uses semantic information to assist in interference cancellation, accurately reconstructs user signals, reduces error propagation, and updates the user decoder network parameters on the base station side, ensuring transmission accuracy for each user while also reducing the base station's processing load and consumption.
Description
Technical Field
The invention belongs to the technical field of communication, and particularly relates to a semantic communication transmission method based on non-orthogonal multiple access.
Background
In the last few decades, communication has mainly focused on how accurately signals in bits are transmitted. In such systems, bit error rates are mainly used to measure system performance. With the continuous development of communication technology, the current transmission rate has increased by tens of thousands times, and the system capacity is approaching shannon limit gradually. Conventional modes of communication are extremely challenging and it is difficult to meet the needs of many emerging applications. Semantic communication gradually comes into the field of view of people, and improvement of communication efficiency is expected by virtue of the semantic communication. The purpose of the semantic communication system is to transmit the meaning of sentences instead of bit transmission in traditional communications. Only information important, relevant and useful to the user can be extracted from the bulk of the data and transmitted. By doing so, the data traffic will be greatly reduced, which may significantly improve transmission efficiency and reliability, which may be particularly useful when bandwidth is limited and SNR is low in a typical communication system.
Designing a suitable multiple access technique is one of the most important ways to increase the system capacity. Multiple access techniques can be broadly divided into two different approaches, orthogonal Multiple Access (OMA) and non-orthogonal multiple access (NOMA). NOMA can serve multiple users at the same time using a single frequency resource, and is a powerful candidate for the next generation of radio access networks. Compared to the well-known high-capacity OMA technique Orthogonal Frequency Division Multiple Access (OFDMA), NOMA can allocate one frequency channel to multiple users simultaneously within the same cell and has many advantages, including improved Spectral Efficiency (SE), higher cell edge throughput, relaxed channel feedback, and low transmission delay. At present, the NOMA research is focused on aspects of power distribution, system capacity and the like, and mainly solves the application problem by considering the shannon capacity theorem.
The introduction of non-orthogonal multiple access systems can further improve spectral efficiency in combination with the concept of semantic communication. Currently, research into semantic communication systems is mainly focused on the traditional communication framework, and is not combined with NOMA. At present, most semantic communication researches do not get rid of thought and framework of end-to-end communication, and the advantages of semantic transmission are not fully exerted. Unlike the modulation and demodulation of signals in conventional communication methods, the structure of the encoder and decoder for semantic communication is inconsistent, and input and output cannot be simply docked. And the processing sequence of the data is also different.
Disclosure of Invention
The invention aims to provide a semantic communication transmission method based on non-orthogonal multiple access. The method can realize the transmission of the non-orthogonal multiple access system in a semantic communication framework, introduce semantic information assistance in the serial interference elimination and realize the decoding of user information, improve the spectrum efficiency, overcome the problem of a large amount of information redundancy in the traditional communication, and optimize the operation time sequence of a base station for realizing the signal detection and the decoding of user data.
The semantic communication transmission method based on non-orthogonal multiple access comprises the following steps:
s1, a semantic background knowledge base is built at a base station based on a text data set, the semantic background knowledge base is shared to each user in a network through broadcasting, and the user uses the semantic background knowledge base to transcode the sent information;
S2, respectively designing a transmitter and a receiver based on a transducer structure at a base station side and a user side to realize semantic information extraction and compression at the user side and semantic information recovery at the base station side;
S3, constructing an uplink non-orthogonal multiple access semantic communication system, reserving semantic information to realize signal reconstruction, and assisting in executing serial interference elimination;
S4, designing a user signal unified decoding network at a base station side to acquire received data, collecting the transmitted data of a user, calculating a cross entropy loss function, and training a depth network model of a receiving end and a transmitting end of the non-orthogonal multiple access semantic communication system;
s5, recovering the user data received by the base station based on the trained model parameters by combining a greedy algorithm and a semantic decoder, and realizing semantic information transmission.
In step S1, a semantic background knowledge base is built based on a text data set at a base station, and is shared to a plurality of users in a network through broadcasting, and the users utilize the semantic knowledge base to perform transcoding processing on the transmitted information. In order to enable the transceiver to understand the semantic information, the base station broadcasts a semantic knowledge base to each user in the network, and a user transmitting end transcodes the user information according to the knowledge base and converts the transmitted sentences into complex signal streams. Wherein the signals of each user do not affect each other and follow a consistent transcoding rule. The data sent by user i is noted as a set of sentences S i＝[s1,s2,…,sN. Where a single sentence s n＝[w1,w2,…,wL],n＝[1,2,…,N],wl represents the first word in the sentence. The user signals are transmitted to the base station through respective channel superposition. After receiving the user superposition signals, the base station executes serial interference elimination to detect signals of different users, restores the signals to the original data of each user according to a knowledge base after decoding and restoring, and restores the user data to complete sentences according to dictionary indexes.
Further, in step S2, transmitters and receivers based on a transducer structure are respectively designed at a base station and a user side to realize semantic information extraction and compression at the user side and semantic information recovery at the base station side, specifically, a semantic encoder and a semantic decoder are respectively composed of 4 transducer encoder layers and 4 transducer decoder layers. At the user transmitting end, the user data is firstly encoded by a semantic encoder, semantic information is extracted and compressed, and then channel encoding is carried out, so that the semantic information is ensured to be successfully transmitted on a physical channel. The signal sent by the user may be denoted as X i＝Cα(Eβ (S)), where C α denotes a channel encoder with parameter α and E β denotes a semantic encoder network with parameter β. After the base station receives the signals and completes signal detection and serial interference elimination, the signals of each user are decoded and recovered independently. The receiver comprises a signal reconstruction network and a decoding recovery network, wherein the signal reconstruction network implements user signal reconstruction and removal from the superimposed signal when performing serial interference cancellation. In the signal reconstruction network, we divide the reconstruction mode into three types of channel reconstruction, semantic reconstruction and direct reconstruction. When performing semantic reconstruction, the base station needs to introduce a semantic encoder based on a transducer after finishing decoding the user signal with strong receiving power. Since the signal dimension changes after decoding, we eliminate embedding layers in the encoder and meet the re-encoding requirement by supplementing the constants of specific dimensions. In the decoding recovery network, after the serial interference elimination is completed, the base station detects signals from different users, recovers transmission signals through channel decoding, and then combines a background knowledge base to recover the signals into data sent by the users through a semantic decoder. The recovered data is represented as: Wherein the method comprises the steps of A channel decoder with a parameter delta is indicated,A semantic decoder representing a parameter χ.
Further, in step S3, an uplink non-orthogonal multiple access semantic communication system is built and designed, semantic information is reserved to realize signal reconstruction, serial interference elimination is assisted, and specifically, 2 users are served by one base station in consideration of a semantic communication scheme based on uplink non-orthogonal multiple access. Both the user and the base station are equipped with a single antenna. To ensure the difference in the received power at the base station, a user channel gain state |h 1|2>|h2|2 is set, and the two users are respectively marked as near users and far users. The users transmit signals to the base station at respective transmit powers. The transmission power P i of the two users is the same and P 1＝P2 =1, and the signal received by the base station is the superposition of the signals of the two users. After signal detection, semantic information is reserved, signal reconstruction is realized by means of a channel codec, channel interference is eliminated, serial interference elimination is assisted, and therefore user data is recovered. The base station realizes the processing procedure of the serial interference elimination, namely, after receiving the superposition signals from two users, the base station detects the signals of different users through the serial interference elimination and decodes the signals. The signal transmitted by the user i is recorded as X i, and the signal received by the base station is recorded asWherein H i is the Rayleigh fading channel between the user and the base station, and Z is the varianceIs a free-standing zero-mean additive white gaussian noise. The user signal detection is realized by using a minimum mean square error algorithm, and the formula is as follows:
Firstly, signal detection is carried out on a near user, a far user signal is regarded as interference, semantic information assistance is introduced, the influence of a channel is eliminated, and the detected near user signal is subjected to The reconstruction is specifically that after the near user signal is detected by using the minimum mean square error, the signal needs to be processed, so as to eliminate transmission errors and channel interference and assist the signal detection of the far user. Defining the processing network as R (,), the processing signal is expressed asIn conventional communications, when reconstructing a received signal, it is necessary to demodulate the signal, then perform channel decoding, and then re-encode and modulate the signal. Based on this idea, a semantic reconstruction scheme is proposed. Modulation in the traditional scheme is converted into semantic coding under the semantic communication framework. After the near-user signal channel is decoded, semantic decoding is carried out to replace demodulation in traditional communication, and then encoding is carried out again. Since the dimensionality of the signal after semantic decoding changes, a constant with a specific dimensionality needs to be added to meet the coding requirement, and coding errors are reduced through training. And when recoding is carried out, embedding layers in the semantic encoder are removed, and the consistency of the recovery signal and the transmission signal is ensured. However, under the framework of semantic communication, if the semantic information is decoded and re-extracted, the extracted and compressed semantic information in the current signal is lost. On the basis, a channel reconstruction scheme is further proposed, namely semantic information in the signal is reserved, and the influence caused by channel interference and signal reconstruction is reduced through channel decoding. In order to show the accuracy of channel reconstruction, we also use the direct reconstruction scheme as a contrast scheme, i.e. without any signal processing, directly using the detected signal for reconstruction, i.e.,The method for setting perfect channel state information of the receiving end and reconstructing the channel state information comprises the following steps:
After the near-user signal reconstruction is completed, the user signal is removed from the total received signal, i.e., Y 2＝Y-Y1. After near-user interference is removed, the residual signal Y 2 contains only far-user signal and gaussian white noise interference Z in the channel. Signal detection of remote user from signal Y 2 Thereby separating out the signals of different users. Therefore, the near user suffers more interference than the far user, while the accuracy of the near user decoding reconstruction has a larger impact on the far user.
Further, in step S4, the user decoding network parameters are jointly trained, the transmitted signal and the received signal of the communication are collected, and the network model is trained according to the cross entropy loss, specifically: in the local training stage, the transmitting signals and the receiving signals of each user are collected, the cross entropy loss is used as an optimization function, the transmission loss values of the two users are obtained respectively, and the depth network parameters of the transmitting and receiving ends of the non-orthogonal multiple access semantic communication system are updated in real time according to the average value of the signal loss. The base station is configured with a receiver to decode and recover the two detected user signals respectively, and distributes the same signal decoding network structure for different received user signals, if the decoding networks are updated independently, namely, the decoding networks are trained independently for different user signals, the remote users are severely interfered and cannot transmit signals correctly, and the average loss value of the two users is difficult to drop correctly. Therefore, the parameters of different decoding networks are uniformly trained, and the data of each user can be correctly transmitted.
Further, in step S5, based on the trained model parameters, user data is recovered by combining a greedy algorithm and a semantic decoder, so as to realize semantic information transmission, which specifically includes: in the training stage, the signals recovered by the base station are used for a cross entropy loss function, a training result is obtained by calculating the difference between the predicted probability distribution and the real probability distribution, and the signals cannot be recovered into sentences directly corresponding to the background semantic knowledge base. In practical application, the trained semantic decoder network is combined with a greedy algorithm, the result of the maximum probability at each word position in the decoded recovery signal is taken, and the rest results are discarded and recombined into a symbol stream. In combination with the semantic knowledge base, the base station can recycle the symbol stream back to the text information that it wants to send. And comparing the text recovered by the receiving end with the text originally sent by the user to obtain BLEU scores and semantic similarity for evaluating the accuracy of the recovered sentences. Wherein, for a transmitted sentence s with a length of l, the receiving end recovers the lengthIs a sentence of (a)The calculation mode of the BLEU score is as follows:
where u n is the weight of the n-grams and p n is the fraction of the n-grams, expressed as: C k (. Cndot.) is the frequency count function of the kth element in the nth gram. Because the BLEU score cannot understand semantic information, the transmission accuracy can only be judged according to the sentence length and the word accuracy, so that the semantic similarity is added as a scoring standard of a semantic layer, and the calculation mode is as follows:
Wherein B φ (·) is a pre-trained bert model containing billions of parameter information for extracting semantics. The signals of each user adopt the same flow to restore sentence information. In addition, the trained decoder model is adopted, so that the signal can be directly restored into a symbol stream when the near-user signal is reconstructed. At this time, the semantic encoder consistent with the transmitting end is used, so that the requirement of signal reconstruction can be met, and the change of signal dimension does not need to be further considered.
Compared with the prior art, the invention has the following remarkable advantages:
(1) The serial interference elimination method of the non-orthogonal multiple access system based on semantic information assistance combines a semantic communication framework, breaks through the limitation of insufficient frequency spectrum resources in the traditional scheme by extracting useful semantic information and filtering irrelevant redundant information, and simultaneously combines the non-orthogonal multiple access system to further improve the frequency spectrum efficiency.
(2) According to the serial interference elimination method for the non-orthogonal multiple access system based on semantic information assistance, three types of signal reconstruction modes are compared and analyzed, the semantic information assistance is used for eliminating interference during the serial interference elimination, the semantic information extracted from the signals is reserved, accurate reconstruction of user signals is achieved, error propagation is reduced, and interference to other users is reduced.
(3) The serial interference elimination method of the non-orthogonal multiple access system based on semantic information assistance provides decoder network parameters for jointly training different users, and the auxiliary training among the users can reduce the interference of near users to far users, ensure the transmission precision of the two users, simultaneously reduce the processing capacity and consumption during training and reduce the processing complexity of a receiver.
Drawings
Fig. 1 is a flow chart of a serial interference cancellation method of a non-orthogonal multiple access system based on semantic information assistance according to an embodiment of the present invention.
Fig. 2 is a diagram of an uplink non-orthogonal multiple access communication network architecture for two users served by a single base station according to an embodiment.
Fig. 3 is a schematic diagram of a network structure of three user signal reconstruction schemes according to an embodiment.
FIG. 4 is a graph of average scores for user BLUE under joint design user receiver parameters in an embodiment;
FIG. 5 is a graph of semantic similarity average scores under joint design user receiver parameters in an embodiment;
fig. 6 shows the average score of user data when jointly designing a user receiver and separately designing receiver parameters in an embodiment, wherein the near-user reconstruction is channel reconstruction.
Detailed Description
The technical scheme of the invention is further described below with reference to the accompanying drawings.
Examples
A semantic communication transmission method based on non-orthogonal multiple access, as shown in figure 1, comprises the following steps,
S1, constructing a semantic background knowledge base based on a text data set at a base station, sharing the semantic background knowledge base to a plurality of users in a network through broadcasting, and carrying out transcoding processing on the transmitted information by the users by utilizing the semantic knowledge base;
In order to simulate and verify the practical application effect of the method, the embodiment processes the text training set and counts word frequency, generates dictionary and provides the dictionary to different user sending terminals and base station receiving terminals as semantic background knowledge base. According to the knowledge base, the transmitting end transcodes the user information, converts the sentence form into a complex symbol stream, and transmits the complex symbol stream to the receiving end of the base station through a Rayleigh fading channel after the encoding is completed. And the base station receives the user signal, decodes the user signal and restores the user signal to a symbol stream, and restores the symbol to user information by combining a background knowledge base. The text data set is used as a training set for preprocessing, sentence length is divided into 4 to 30 words, character formats are standardized, and corresponding dictionaries, namely a semantic background knowledge base, are generated according to the number of words and are used for realizing the interconversion of sentence information and signal streams. In order to make the transceiver understand the semantic information, the base station broadcasts a semantic knowledge base to each user, and the user transmitting end transcodes the user information according to the knowledge base and converts the transmitted sentences into complex signal streams. Wherein the signals of each user do not affect each other and follow a consistent transcoding rule. The data sent by user i is noted as a set of sentences S i＝[s1,s2,…,sN. Where a single sentence s n＝[w1,w2,…,wL],n＝[1,2,…,N],wl represents the first word in the sentence. The user signals are transmitted to the base station through respective channel superposition. After receiving the user superposition signals, the base station executes serial interference elimination to detect signals of different users, restores the signals to the original data of each user according to a knowledge base after decoding and restoring, and restores the user data to complete sentences according to dictionary indexes.
S2, respectively designing a transmitter and a receiver based on a transducer structure at a base station and a user side to realize semantic information extraction and compression at the user side and semantic information recovery at the base station side;
In order to realize the semantic communication framework, semantic information in user data needs to be extracted and compressed, so that the semantic communication requirement is met. The semantic encoder and decoder in the embodiment are made up of 4 transform encoder layers and 4 transform decoder layers, respectively. At the user transmitting end, the user data is firstly encoded by a semantic encoder, semantic information is extracted and compressed, and then channel encoding is carried out, so that the semantic information is ensured to be successfully transmitted on a physical channel. The signal sent by the user may be denoted as X i＝Cα(Eβ (S)), where C α denotes a channel encoder with parameter α and E β denotes a semantic encoder network with parameter β. After the base station receives the signals and completes signal detection and serial interference elimination, the signals of each user are decoded and recovered independently. The receiver comprises a signal reconstruction network and a decoding network, wherein the signal reconstruction network performs user signal reconstruction and removal from the superimposed signal when performing serial interference cancellation. In the signal reconstruction network, we divide the reconstruction mode into three types of channel reconstruction, semantic reconstruction and direct reconstruction. When performing semantic reconstruction, the base station needs to introduce a semantic encoder based on a transducer after finishing decoding the user signal with strong receiving power. Since the signal dimension changes after decoding, we eliminate embedding layers in the encoder and meet the re-encoding requirement by supplementing the constants of specific dimensions. After the serial interference elimination is completed, the base station detects signals from different users, recovers transmission signals through channel decoding, and then combines a background knowledge base to recover the signals into data sent by the users through a semantic decoder. The recovered data is represented as: Wherein the method comprises the steps of A channel decoder with a parameter delta, and a semantic decoder with a parameter χ.
S3, constructing and designing an uplink non-orthogonal multiple access semantic communication system, reserving semantic information to realize signal reconstruction, and assisting in executing serial interference elimination;
As shown in fig. 2, to implement user communication in conjunction with a non-orthogonal multiple access system, an embodiment designs a communication network in which two users are served by a single base station. The base station and the user terminal are both provided with a single antenna, and in order to ensure the difference of the receiving power of different users at the base station, a user channel gain state |h 1|2>|h2|2 is set, and the two users are respectively marked as a near user and a far user. The users transmit signals to the base station at respective transmit powers. The transmitting power P i of the two users is the same and P 1＝P2 =1, the signals received by the base station are superposition of the signals of the two users, after signal detection, semantic information is reserved, signal reconstruction is realized by means of a channel coder and decoder, channel interference is eliminated, serial interference elimination is assisted, and therefore accurate recovery of user data is ensured.
The processing procedure of the base station receiving end serial interference elimination is that, at the receiving end, after the base station receives the superimposed signals from two users, the base station detects the signals from different users through the serial interference elimination and decodes. The signal transmitted by the user i is recorded as X i, and the signal received by the base station is recorded asWherein H i is the Rayleigh fading channel between the user and the base station, and Z is the varianceIs a free-standing zero-mean additive white gaussian noise. The user signal detection is realized by using a minimum mean square error algorithm, and the formula is shown in the specificationThe total noise variance of the user i includes interference caused by the far-end user and channel noise interference for the near-end user, and the near-end user signal is removed from the total received signal when the signal is detected, so that only the channel noise interference exists
Firstly, signal detection is carried out on a near user, a far user signal is regarded as interference, semantic information assistance is introduced, the influence of a channel is eliminated, and the detected near user signal is subjected toThe reconstruction is specifically that after the near user signal is detected by using the minimum mean square error, the signal needs to be processed, so as to eliminate transmission errors and channel interference and assist the signal detection of the far user. Defining the processing network as R (,), the processing signal is expressed asIn conventional communications, when reconstructing a received signal, it is necessary to demodulate the signal, then perform channel decoding, and then re-encode and modulate the signal. Based on this idea, a semantic reconstruction scheme is proposed. Modulation in the traditional scheme is converted into semantic coding under the semantic communication framework. After the near-user signal channel is decoded, semantic decoding is carried out to replace demodulation in traditional communication, and then encoding is carried out again. Since the dimensionality of the signal after semantic decoding changes, a constant with a specific dimensionality needs to be added to meet the coding requirement, and coding errors are reduced through training. And when recoding is carried out, embedding layers in the semantic encoder are removed, and the consistency of the recovery signal and the transmission signal is ensured. However, under the framework of semantic communication, if the semantic information is decoded and re-extracted, the extracted and compressed semantic information in the current signal is lost. On the basis, a channel reconstruction scheme is further proposed, namely semantic information in the signal is reserved, and the influence caused by channel interference and signal reconstruction is reduced through channel decoding. In order to show the accuracy of channel reconstruction, we also use the direct reconstruction scheme as a contrast scheme, i.e. without any signal processing, directly using the detected signal for reconstruction, i.e.,The method for setting perfect channel state information of the receiving end and reconstructing the channel state information comprises the following steps:
After the near-user signal reconstruction is completed, the user signal is removed from the total received signal, i.e., Y 2＝Y-Y1. After the near user interference is removed, the residual signal only contains the far user signal and Gaussian white noise interference in the channel. And then, detecting the minimum mean square error of the remote user from the residual signal, thereby separating the signals of different users. Therefore, the near user suffers more interference than the far user, while the accuracy of the near user decoding reconstruction has a larger impact on the far user.
In the signal detection process, the accuracy of user signal reconstruction plays a key role in signal transmission of all users in the system. The embodiment provides a channel reconstruction scheme, wherein a minimum mean square error algorithm is used as a signal detection algorithm to perform channel decoding on detected near-user signals so as to reduce interference generated by a physical channel to the maximum extent, and the processed signals contain coded semantic information sent by users. And on the premise of retaining the coding information, repeated channel coding is carried out, the signal state before the transmitting end enters the physical channel is recovered, and signal reconstruction is realized. The scheme can maximally reserve the semantic information sent and coded by the user while removing the physical channel noise, and has lower influence on subsequent semantic decoding. If the semantic information in the signal is reconstructed after the channel decoding is performed, namely, the semantic reconstruction scheme is adopted, the semantic information is subjected to certain loss, and the accuracy of signal transmission is greatly affected. If the transmitted user signal is not reconstructed, that is, the signal passing through the physical channel is directly used as the reconstructed signal, the user information cannot be correctly recovered, and the signal transmission is greatly affected. The network structure of the three types of reconstruction schemes is shown in fig. 3. The performance pairs of the three schemes, i.e. the channel reconstruction, the semantic reconstruction and the direct reconstruction, are shown in fig. 4 and fig. 5 respectively by taking the semantic similarity and the BLEU score as judgment references.
S4, decoding network parameters by the joint training user, collecting communication sending signals and receiving signals, and training a network model according to cross entropy loss;
In the local training stage, the transmitting signals and the receiving signals of each user are collected, the cross entropy loss is used as an optimization function, the transmission loss values of the two users are obtained respectively, and the depth network parameters of the transmitting and receiving ends of the non-orthogonal multiple access semantic communication system are updated in real time according to the average value of the signal loss. The base station is configured with a receiver, decodes and recovers the two detected user signals respectively, and distributes the same signal decoding network structure to the received different user signals. According to the semantic background knowledge base shared by the transmitting end and the receiving end, namely the text data set, the user transmitting signal is encoded, semantic information is extracted, symbols are converted into complex signal streams, and the encoded signals are transmitted to the base station receiving end through a physical channel. At the base station end, when the embodiment decodes the user signal, the embodiment decodes and recovers the near user signal, and then decodes the far user signal from the residual signal. The embodiment adopts channel reconstruction to realize near user signal recovery and combines with design to update decoder parameters of different users, thereby avoiding interference caused by near users to far users during signal decoding in the training process. A comparison of the average scores of the two users' transmission accuracy when updating the parameters jointly and independently is shown in fig. 6. It can be obtained that during training, the decoding parameters of two users can be jointly designed to effectively ensure the receiving accuracy. When the two user signal decoding parameters are different, the average accuracy of the user is lower. When the network is trained, the embodiment decodes the signals of different users by using the same decoding network parameters, and uniformly optimizes the signals according to the training loss, thereby avoiding the problem that the network parameters are difficult to update correctly when the remote users decode. Under the condition of independently updating the model parameters, the updating of the decoding model parameters of the far-end user is interfered by the near-end user, so that the model cannot be updated correctly in the training process, and the decoding signal of the far-end user has larger loss and cannot transmit data correctly.
S5, recovering user data based on the trained model parameters and combining a greedy algorithm and a semantic decoder to realize semantic information transmission.
In the training stage, the signals recovered by the base station are used for a cross entropy loss function, a training result is obtained by calculating the difference between the predicted probability distribution and the real probability distribution, and the signals cannot be recovered into sentences directly corresponding to the background semantic knowledge base. In the embodiment, after training is completed, the base station receiving end decodes the user signal by combining the trained semantic decoder parameters through a greedy algorithm, and then recovers the semantic information sent by the user through a background semantic knowledge base, so that signal transmission is realized. After eliminating channel interference, the base station end adopts greedy decoding to extract semantic information in the signals and recombine the semantic information into sentences. And taking the result of the maximum probability at each word position in the decoding recovery signal, and discarding the rest results to be recombined into a symbol stream. In combination with the semantic knowledge base, the base station can recycle the symbol stream back to the text information that it wants to send. Comparing the text recovered by the receiving end with the text originally sent by the user, the embodiment takes the semantic similarity and bilingual evaluation replacement (BLEU) score as the evaluation of the accuracy of the recovered semantic information. The BLEU score is a common evaluation standard in natural language processing, and can compare and judge the word accuracy of a target sentence, but semantic information cannot be understood and semantic score cannot be combined. The semantic similarity is therefore used as another evaluation criterion to score the accuracy of the transmitted sentences at the semantic level. For a transmitted sentence s with a length of l, the receiving end recovers the lengthIs a sentence of (a)The calculation mode of the BLEU score is as follows:
where u n is the weight of the n-grams and p n is the fraction of the n-grams, expressed as: c k (. Cndot.) is the frequency count function of the kth element in the nth gram.
The semantic similarity is used as a scoring standard of a semantic layer, and the calculation mode is as follows:
Wherein B φ (·) is a pre-trained bert model containing billions of parameter information for extracting semantics. The signals of each user adopt the same flow to restore sentence information. In addition, the trained decoder model is adopted, so that the signal can be directly restored into a symbol stream when the near-user signal is reconstructed. At this time, the semantic encoder consistent with the transmitting end is used, so that the requirement of signal reconstruction can be met, and the change of signal dimension does not need to be further considered.
Although the present invention has been described with reference to the foregoing embodiments, it will be apparent to those skilled in the art that modifications may be made to the embodiments described above, or equivalents may be substituted for elements thereof, and any modifications, equivalents, improvements and changes may be made without departing from the spirit and principles of the present invention.
Claims (4)
1. The semantic communication transmission method based on non-orthogonal multiple access is characterized by comprising the following steps:
s1, a semantic background knowledge base is built at a base station based on a text data set, the semantic background knowledge base is shared to each user in a network through broadcasting, and the user uses the semantic background knowledge base to transcode the sent information;
S2, respectively designing a transmitter and a receiver based on a transducer structure at a base station side and a user side to realize semantic information extraction and compression at the user side and semantic information recovery at the base station side;
S3, constructing an uplink non-orthogonal multiple access semantic communication system, reserving semantic information to realize signal reconstruction, and assisting in executing serial interference elimination;
The step S3 comprises the steps of considering a semantic communication scheme based on uplink non-orthogonal multiple access, serving two users by a base station, enabling the users and the base station to be provided with single antennas, setting a user channel gain state of (1|2>|h2|2), marking the two users as near users and far users respectively, sending signals to the base station by the users according to respective transmitting power, wherein the transmitting power of the two users is P i the same and P 1＝P2 =1, superposing signals received by the base station is two user signals, retaining semantic information after signal detection, realizing signal reconstruction by means of a channel codec, eliminating channel interference and assisting in executing serial interference elimination so as to recover user data, wherein the processing procedure of realizing serial interference elimination by the base station is that after receiving superposition signals from the two users, signals of different users are detected and decoded by the serial interference elimination, marking the signals transmitted by the user i as X i, and the signals received by the base station are Wherein H i is the Rayleigh fading channel between the user and the base station, and Z is the varianceThe method comprises the steps of realizing user signal detection by using a minimum mean square error algorithm, wherein the formula is as follows:
Firstly, signal detection is carried out on a near user, a far user signal is regarded as interference, semantic information assistance is introduced, the influence of a channel is eliminated, and the detected near user signal is subjected to Reconstructing, detecting near user signal with minimum mean square error, processing signal, eliminating transmission error and channel interference, assisting far user signal detection, defining processing network as R (-), and processing signal asAfter finishing the decoding of the near user signal channel, carrying out semantic decoding and then re-encoding, adding a dimensional constant to meet the encoding requirement and reducing the encoding error through training, removing embedding layers in the semantic encoder during re-encoding to ensure that the recovered signal is consistent with the transmission signal, using the detected signal for reconstruction, namely,The method for setting the receiving end to have channel state information comprises the following steps:
After the near user signal is reconstructed, the user signal is removed from the total received signal, namely Y 2＝Y-Y1, after the near user interference is removed, the residual signal Y 2 only contains the far user signal and Gaussian white noise interference Z in the channel, and the far user signal is detected from the signal Y 2 Thereby separating out signals of different users;
S4, designing a user signal unified decoding network at a base station side to acquire received data, collecting the transmitted data of a user, calculating a cross entropy loss function, and training a depth network model of a receiving end and a transmitting end of the non-orthogonal multiple access semantic communication system;
Step S4, in the local training stage, collecting the transmitted signals and the received signals of each user, using cross entropy loss as an optimization function, respectively obtaining transmission loss values of the two users, and updating the depth network parameters of a receiving and transmitting end of the non-orthogonal multiple access semantic communication system in real time according to the signal loss average value;
s5, recovering user data received by the base station based on model parameters trained by the depth network model and combining a greedy algorithm and a semantic decoder to realize semantic information transmission.
2. The method of claim 1, wherein the step S1 is characterized in that a text data set is preprocessed as a training set, the sentence length is divided into 4 to 30 words, character formats are standardized, corresponding dictionaries, namely semantic background knowledge bases are generated according to the number of words and used for realizing the mutual conversion of sentence information and signal streams, in order to enable a transceiver mechanism to solve the semantic information, a base station broadcasts a semantic knowledge base to each user in a network, a user transmitting end transcodes the user information according to the knowledge base, the transmitted sentences are converted into complex signal streams, signals of each user are independent of each other, consistent transcoding rules are followed, data transmitted by a user i are a group of sentences S i＝[s1,s2,…,sN, wherein a single sentence S n＝[w1,w2,…,wL],n＝[1,2,…,N],wl represents a first word in the sentences, the user signals are transmitted to a base station through respective channel superposition, after receiving the user superposition signals, serial interference elimination is performed to detect signals of different users, the original data of each user are restored according to the knowledge base after decoding restoration, and the user data are restored into complete sentences according to dictionary indexes.
3. The method of claim 1, wherein the step S2 is specifically that the semantic encoder and the decoder are respectively composed of 4 transducer encoder layers and 4 transducer decoder layers, at the user transmitting end, user data is firstly encoded by the semantic encoder, semantic information is extracted and compressed, then channel encoding is performed to ensure that the semantic information is successfully transmitted on a physical channel, a signal transmitted by the user is represented as X i＝Cα(Eβ (S)), wherein C α represents a channel encoder network with a parameter of alpha, E β represents a semantic encoder network with a parameter of beta, after the base station receives the signal and completes signal detection and serial interference elimination, the signal of each user is independently decoded and recovered, the receiver comprises a signal reconstruction network and a decoding and recovering network, wherein the signal reconstruction network realizes user signal reconstruction and removes from an overlapped signal when serial interference elimination is performed, the reconstruction mode is divided into three modes of channel reconstruction, semantic reconstruction and direct reconstruction in the signal reconstruction network, the signal reconstruction mode is represented as X i＝Cα(Eβ (S)), wherein the signal transmitted by the user represents a channel encoder with a parameter of alpha, E β represents a signal encoder network with a parameter of beta, the signal is decoded by the base station after the base station receives the signal and completes signal detection and serial interference elimination, the signal recovery network is completed by the base station, the signal recovery network is completed after the signal recovery is completed, the signal recovery network is completed by decoding and the signal recovery network is completed, the signal recovery network is completed by the base station has the signal recovery system, and the signal recovery network is completed after the signal recovery has been completed, and the signal recovery network is decoded by the decoding and the signal has been completed, and the signal has been recovered by the user signal decoder: Wherein the method comprises the steps of A channel decoder with a parameter delta is indicated,A semantic decoder representing a parameter χ.
4. The method of claim 1, wherein step S5 comprises the steps of obtaining training results by calculating differences between predictive probability distribution and true probability distribution in training phase, obtaining training results by combining trained semantic decoder network with greedy algorithm, obtaining maximum probability results at each word position in decoded recovery signals, discarding the rest results and recombining the results into symbol stream, combining semantic knowledge base, recovering text information intended to be sent by users by the symbol stream by the base station, comparing text recovered by the receiving end with original text sent by users to obtain BLU score and semantic similarity for evaluating accuracy of recovery sentences, wherein for a sent sentence S with length of l, recovering length by the receiving endIs a sentence of (a)The calculation mode of the BLEU score is as follows:
where u n is the weight of the n-grams and p n is the fraction of the n-grams, expressed as: c k (·) is a frequency count function of the kth element in the nth gram, and semantic similarity is added as a scoring standard of a semantic layer, and the calculation mode is as follows:
Wherein B φ (.) is a pre-trained bert model, which contains billions of parameter information for extracting semantics, the signals of each user adopt the same flow to restore sentence information, and the trained decoder model is adopted to directly restore the signals into symbol streams when the near-user signals are reconstructed, and the semantic encoder consistent with the transmitting end is used to satisfy the signal reconstruction.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310419000.4A CN116390134B (en) | 2023-04-18 | 2023-04-18 | Semantic communication transmission method based on non-orthogonal multiple access |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310419000.4A CN116390134B (en) | 2023-04-18 | 2023-04-18 | Semantic communication transmission method based on non-orthogonal multiple access |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116390134A CN116390134A (en) | 2023-07-04 |
| CN116390134B true CN116390134B (en) | 2025-09-09 |
Family
ID=86976797
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310419000.4A Active CN116390134B (en) | 2023-04-18 | 2023-04-18 | Semantic communication transmission method based on non-orthogonal multiple access |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116390134B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN119324759A (en) * | 2023-07-17 | 2025-01-17 | 北京邮电大学 | Uplink protocol, downlink protocol and device based on multiple access technology |
| WO2025055249A1 (en) * | 2023-09-13 | 2025-03-20 | Huawei Technologies Co., Ltd. | Transmission method and apparatuses for controlling information |
| CN117614584B (en) * | 2023-10-26 | 2024-07-26 | 北京邮电大学 | Channel-migratable semantic communication method and related equipment |
| CN120475438A (en) * | 2024-02-05 | 2025-08-12 | 索尼集团公司 | Electronic device and method for wireless communication, and computer-readable storage medium |
| CN118590185A (en) * | 2024-02-27 | 2024-09-03 | 北京邮电大学 | Multi-user information transmission method based on semantic basis and related device |
| CN119740099A (en) * | 2025-03-03 | 2025-04-01 | 深圳大学 | Edge multi-task semantic communication method, device, computer equipment and storage medium |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7248841B2 (en) * | 2000-06-13 | 2007-07-24 | Agee Brian G | Method and apparatus for optimization of wireless multipoint electromagnetic communication networks |
| TR202005532A1 (en) * | 2020-04-07 | 2021-10-21 | Aselsan Elektronik Sanayi Ve Ticaret As | A METHOD TO IMPROVE THE PERFORMANCE OF UPWAY CONNECTION NON-ORANGLE MULTI-ACCESS METHOD |
| CN112911714B (en) * | 2021-02-01 | 2022-06-21 | 福州大学 | NOMA two-user downlink decoding method based on power distribution |
| CN115309869B (en) * | 2022-08-19 | 2025-08-19 | 南京邮电大学 | One-to-many multi-user semantic communication model and communication method |
| CN115866622B (en) * | 2022-11-24 | 2025-01-03 | 南京邮电大学 | Transmission method for uplink multi-cell MIMO semantic communication system with distributed architecture |
-
2023
- 2023-04-18 CN CN202310419000.4A patent/CN116390134B/en active Active
Non-Patent Citations (1)
| Title |
|---|
| 《Uplink NOMA Semantic Communications: Semantic Reconstruction for SIC》;Bingcen Chen等;《 IEEE/CIC International Conference on Communications in China (ICCC)》;20230905;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116390134A (en) | 2023-07-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116390134B (en) | Semantic communication transmission method based on non-orthogonal multiple access | |
| Luo et al. | Semantic communications: Overview, open issues, and future research directions | |
| CN116436567B (en) | A semantic communication method based on deep neural network | |
| Lu et al. | Rethinking modern communication from semantic coding to semantic communication | |
| Weng et al. | Semantic communication systems for speech transmission | |
| Zhou et al. | Cognitive semantic communication systems driven by knowledge graph | |
| Hu et al. | One-to-many semantic communication systems: Design, implementation, performance evaluation | |
| Wu et al. | Semantic segmentation-based semantic communication system for image transmission | |
| Yi et al. | Deep learning-empowered semantic communication systems with a shared knowledge base | |
| Tian et al. | The achievable distortion region of sending a bivariate Gaussian source on the Gaussian broadcast channel | |
| CN114461816B (en) | Knowledge-graph-based implementation method of information supplement semantic communication system | |
| CN115114928A (en) | An Interpretable Semantic Communication System Based on Feature Selection | |
| Tang et al. | Cooperative semantic communication with on-demand semantic forwarding | |
| CN115333900B (en) | A Cooperative Channel Estimation and CSI Feedback Method for Unmanned Aerial Vehicles Based on Deep Learning | |
| CN115687921A (en) | Robust semantic communication implementation method and system capable of resisting semantic noise | |
| CN115866622B (en) | Transmission method for uplink multi-cell MIMO semantic communication system with distributed architecture | |
| Chen et al. | Uplink noma semantic communications: Semantic reconstruction for sic | |
| Tang et al. | Text semantic communication systems with sentence-level semantic fidelity | |
| Tong et al. | Alternate Learning-Based SNR-Adaptive Sparse Semantic Visual Transmission | |
| Li et al. | Content adaptive distributed joint source-channel coding for image transmission with hyperprior | |
| Zhu et al. | Deep learning for waveform level receiver design with natural redundancy | |
| Dong et al. | Innovative semantic communication system | |
| CN103050123B (en) | A kind of method and system of transmitting voice information | |
| Yi et al. | Semantic communication systems with a shared knowledge base | |
| Ghassemi et al. | Generative AI-enabled Blockage Prediction for Robust Dual-Band mmWave Communication |
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 |