Source transmission method based on full-duplex communication and electronic device

Abstract

The application provides a kind of source transmission method and electronic equipment based on full duplex communication, by the first source information obtained is encoded and modulated, interference information sequence is obtained, interference time domain transmission sequence is obtained based on interference information sequence by inverse fourier transform function, and is sent to full duplex transmission channel, when the sending time domain transmission sequence sent by second terminal in full duplex channel is interfered by interference time domain transmission sequence, interference aliasing is obtained, receives time domain transmission sequence, and source reconstruction is carried out based on receiving time domain transmission sequence and interference time domain transmission sequence, eliminate the interference influence existing in receiving time domain transmission sequence in the process of source reconstruction, obtain the source reconstruction information after eliminating interference, realize source reconstruction, and further guarantee that full duplex two-way communication can be carried out normally.

Description

Technical Field

[0001] This application relates to the field of communication technology, and in particular to a source transmission method and electronic device based on full-duplex communication. Background Technology

[0002] In end-to-end bidirectional communication scenarios, using the same frequency and time resources to transmit information in both uplink and downlink simultaneously can lead to strong interference between the information transmission of the uplink and downlink. The downlink received signal of a device at one end of the link will be strongly interfered with by the leakage of its uplink transmitted signal. Moreover, this leakage interference cannot be completely eliminated by methods including analog domain and digital domain interference cancellation, making it difficult to reconstruct the downlink received signal during the transmission of the information source and failing to guarantee normal bidirectional communication. Summary of the Invention

[0003] In view of this, the purpose of this application is to propose a source transmission method and electronic device based on full-duplex communication to solve the above-mentioned technical problems.

[0004] To achieve the above objectives, a first aspect of this application provides a source transmission method based on full-duplex communication, applied to a first terminal, the method comprising:

[0005] Obtain information from the first source;

[0006] The first source information is encoded and modulated to obtain an interference information sequence, and an interference time-domain transmission sequence is obtained based on the interference information sequence through an inverse Fourier transform function.

[0007] The interference time-domain transmission sequence is sent to a full-duplex transmission channel. The full-duplex transmission channel receives the interference time-domain transmission sequence and simultaneously receives a transmission time-domain transmission sequence sent by a second terminal. When the interference time-domain transmission sequence and the transmission time-domain transmission sequence are transmitted in the full-duplex transmission channel, the interference time-domain transmission sequence and the transmission time-domain transmission sequence will interfere and overlap with each other. The transmission time-domain transmission sequence is interfered with and overlapped by the interference time-domain transmission sequence to obtain the received time-domain transmission sequence.

[0008] Receive the received time-domain transmission sequence transmitted through the full-duplex transmission channel;

[0009] Source reconstruction is performed based on the received time-domain transmission sequence and the interference time-domain transmission sequence to obtain source reconstruction information.

[0010] Based on the same inventive concept, a second aspect of this application provides a source transmission method based on full-duplex communication, applied to a second terminal, the method comprising:

[0011] Obtain information from a second source;

[0012] The second source information is encoded and modulated to obtain a transmission sequence, and a transmission time-domain transmission sequence is obtained based on the transmission sequence through an inverse Fourier transform function.

[0013] The transmit time-domain transmission sequence is sent to a full-duplex transmission channel. The full-duplex transmission channel receives the transmit time-domain transmission sequence and simultaneously receives an interference time-domain transmission sequence sent by the first terminal. When the interference time-domain transmission sequence and the transmit time-domain transmission sequence are transmitted in the full-duplex transmission channel, the interference time-domain transmission sequence and the transmit time-domain transmission sequence will interfere and overlap with each other. The transmit time-domain transmission sequence is interfered with and overlapped by the interference time-domain transmission sequence to obtain the receive time-domain transmission sequence.

[0014] Based on the same inventive concept, a third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the program to implement the method described in the first aspect or the method described in the second aspect.

[0015] As can be seen from the above, the source transmission method and electronic device based on full-duplex communication provided in this application obtains an interference information sequence by encoding and modulating the acquired first source information. Then, based on the interference information sequence, an interference time-domain transmission sequence is obtained through the inverse Fourier transform function and sent to the full-duplex transmission channel. In the full-duplex channel, the transmission time-domain transmission sequence sent by the second terminal is interfered with and aliased by the interference time-domain transmission sequence to obtain a received time-domain transmission sequence. Source reconstruction is performed based on the received time-domain transmission sequence and the interference time-domain transmission sequence. During the source reconstruction process, the interference effect present in the received time-domain transmission sequence is eliminated to obtain the source reconstruction information after interference elimination, thus realizing source reconstruction and ensuring normal full-duplex bidirectional communication. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a source transmission method based on full-duplex communication according to an embodiment of this application;

[0018] Figure 2A This is a schematic diagram of the source transmission structure based on full-duplex communication according to an embodiment of this application;

[0019] Figure 2B This is a schematic diagram of a source transmission structure based on full-duplex communication according to another embodiment of this application;

[0020] Figure 2C This is a schematic diagram of the structure of a digital-semantic full-duplex signal detector according to an embodiment of this application;

[0021] Figure 2D This is a schematic diagram of the semantic domain information detection structure in an embodiment of this application;

[0022] Figure 2E This is a schematic diagram illustrating semantic interference cancellation in an embodiment of this application;

[0023] Figure 2F This is a schematic diagram illustrating semantic interference cancellation according to another embodiment of this application;

[0024] Figure 2G This is a schematic diagram illustrating semantic interference cancellation in another embodiment of this application;

[0025] Figure 2H This is a schematic diagram of a source transmission structure based on full-duplex communication according to another embodiment of this application;

[0026] Figure 2I This is a schematic diagram of the structure of a digital-semantic full-duplex signal detector according to another embodiment of this application;

[0027] Figure 2J This is a schematic diagram of a semantic interference cancellation auxiliary digital signal detection structure according to an embodiment of this application;

[0028] Figure 2K This is a schematic diagram comparing the reconstruction quality of source information in embodiments of this application;

[0029] Figure 2L This is a schematic diagram comparing the image reconstruction effects of embodiments of this application;

[0030] Figure 3 A flowchart illustrating a source transmission method based on full-duplex communication according to another embodiment of this application;

[0031] Figure 4 This is a schematic diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0033] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," and similar terms used in the embodiments of this application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0034] In related technologies, the two half-duplex methods, Frequency Division Duplex (FDD) and Time Division Duplexing (TDD), cannot achieve the optimal transmission rate of a full-duplex system. This results in limited spectral efficiency of the wireless communication system, failing to meet the needs of wireless communication scenarios where the system not only needs to expand wireless transmission resources but also needs to maximize spectral efficiency within limited transmission resources.

[0035] Furthermore, the optimal transmission rate for a full-duplex system is achieved through in-band full-duplex mode, which uses the same frequency and time resources to transmit uplink and downlink signals simultaneously. However, this mode leads to strong interference between uplink and downlink information transmissions. Specifically, the downlink received signal at one end of the link is strongly interfered with by the leakage of its uplink transmitted signal. This leakage cannot be completely eliminated by methods including analog and digital domain interference cancellation, making it difficult to reconstruct the downlink received signal during source transmission, and hindering the normal operation of the full-duplex system.

[0036] This application proposes a source transmission method based on full-duplex communication, applied to a first terminal, such as... Figure 1 As shown, the method includes:

[0037] Step 101: Obtain the first source information.

[0038] In this step, the first source information is the source information transmitted by the first terminal. Before transmitting the first source information, the type of source information to be transmitted is determined, including but not limited to high-dimensional source types such as text sources, voice sources, image sources, and video sources, or basic binary or multi-dimensional independent or related sources, etc. Subsequently, the first terminal can select, acquire, or generate the corresponding first source information according to the type of source information to be transmitted.

[0039] The first terminal can be either a sending terminal or a receiving terminal, but a receiving terminal is preferred here.

[0040] Step 102: Encode and modulate the first source information to obtain an interference information sequence. Based on the interference information sequence, obtain the interference time-domain transmission sequence through the inverse Fourier transform function.

[0041] In this step, the first source information is encoded and modulated to convert the first source information to be transmitted into an interference information sequence.

[0042] Based on the interference information sequence, the frequency domain interference information sequence is mapped to the time domain through the inverse Fourier transform function to form the interference time domain transmission sequence.

[0043] Step 103: The interference time-domain transmission sequence is sent to a full-duplex transmission channel. The full-duplex transmission channel receives the interference time-domain transmission sequence and simultaneously receives the transmission time-domain transmission sequence sent by the second terminal. When the interference time-domain transmission sequence and the transmission time-domain transmission sequence are transmitted in the full-duplex transmission channel, the interference time-domain transmission sequence and the transmission time-domain transmission sequence will interfere and overlap with each other. The transmission time-domain transmission sequence is interfered with and overlapped by the interference time-domain transmission sequence to obtain the received time-domain transmission sequence.

[0044] In this step, the interference time-domain transmission sequence is sent to the full-duplex transmission channel. When the interference time-domain transmission sequence and the transmission time-domain transmission sequence sent by the second terminal are transmitted in the full-duplex transmission channel, the transmission time-domain transmission sequence transmitted by the second terminal will be aliased by direct leakage interference and scattering interference caused by the interference time-domain transmission sequence transmitted by the first terminal. The transmission time-domain transmission sequence after being aliased by interference forms the reception time-domain transmission sequence.

[0045] Step 104: Receive the received time-domain transmission sequence transmitted through the full-duplex transmission channel.

[0046] In this step, the transmitted time-domain transmission sequence after interference and aliasing from the full-duplex transmission channel is received to form the received time-domain transmission sequence, providing a data source for source reconstruction.

[0047] Step 105: Based on the received time-domain transmission sequence and the interference time-domain transmission sequence, perform source reconstruction to obtain source reconstruction information.

[0048] In this step, the received time-domain transmission sequence transmitted through the full-duplex transmission channel is reconstructed based on the interference time-domain transmission sequence generated inside the first terminal. During the source reconstruction process, the interference in the received time-domain transmission sequence is eliminated, and the source reconstruction information after interference elimination is obtained. This realizes the source reconstruction and completes the source transmission process, thereby ensuring that full-duplex bidirectional communication can be carried out normally.

[0049] The above scheme involves encoding and modulating the acquired first source information to obtain an interference information sequence. Then, based on this interference information sequence, an interference time-domain transmission sequence is obtained through an inverse Fourier transform function and sent to a full-duplex transmission channel. Within the full-duplex channel, the transmission time-domain transmission sequence sent by the second terminal is interfered with and aliased by the interference time-domain transmission sequence, resulting in a received time-domain transmission sequence. Source reconstruction is performed based on the received time-domain transmission sequence and the interference time-domain transmission sequence. During source reconstruction, the interference in the received time-domain transmission sequence is eliminated, resulting in a reconstructed source information after interference elimination. This achieves source reconstruction, completes the source transmission process, and ensures normal full-duplex bidirectional communication.

[0050] In some embodiments, the interference information sequence includes an interference digital transmission sequence.

[0051] Step 102 includes:

[0052] Step A1: Compress the first source information to obtain the compressed bit sequence of the interference source.

[0053] Step A2: Channel coding and modulation are performed on the compressed bit sequence of the interference source to obtain the interference digital transmission sequence.

[0054] In the above scheme, such as Figure 2A As shown, the interference information sequence includes the interference digital transmission sequence s. I The first terminal uses an interference link digital encoder, while the second terminal uses a semantic encoder, forming a source transmission structure compatible with both digital and semantic transmission links for full-duplex communication. The interference link digital encoder includes a source encoder E. s and channel encoder E c Using source encoder E s For the first source information x I Perform lossless or loss-limited compression processing to output the compressed bit sequence b of the interference source. I Among them, according to the first source information x I Select the appropriate digital source encoder E based on the type of information source.s Perform lossless or loss-limited source coding compression to output the compressed bit sequence b of the interfering source. I .

[0055] First source information x for different types of information sources I Employ the appropriate digital source encoder E s For example, the first source information x I When the information source is a text source, Huffman coding is used for lossless source coding compression, while the first source information x I When the information source is an image source, image compression (JPEG2000) or Better Portable Graphics (BPG) image encoders are used for distortion-limited source coding compression. When the first source information x I When the source is a video source, an upgraded digital video compression format (H.264 / AVC) or digital video compression format (H.265 / HEVC) video encoder is used for distortion-limited source encoding compression.

[0056] Using channel encoder E c Compress the bit sequence b of the interference source I Perform channel coding and modulation, and output the interference digital transmission sequence s I In this process, the channel coding rate is determined based on the channel state information fed back by the receiving signal-to-noise ratio or signal-to-interference-plus-noise ratio from the receiving equipment of the interference link (i.e., the interference link digital encoder). Channel coding methods such as low-density parity-check (LDPC) codes and polar codes are then used to compress the bit sequence b of the interference source. I Error protection is performed; subsequently, through rate adaptation and modulation, the channel-coded interference source compressed bit sequence b is... I Converted into an interference digital transmission sequence s I .

[0057] Using a signal mapping module, the interfering digital transmission sequence s I Converted into an interference time-domain transmission sequence Among them, the interference digital transmission sequence s I Mapping the waveform onto the time-frequency resources occupied by the transmitting end, thereby achieving complete multiplexing within the same time-frequency resource region, the signal mapping module, based on the waveform mapping process of Orthogonal Frequency Division Multiplexing (OFDM), will map the interfering digital transmission sequence s carried in the frequency domain. I Interference time-domain semantic transmission sequence converted to time domain Then, the interference time-domain transmission sequence is transmitted into the full-duplex channel. In this process, the receiving end (i.e., the first terminal) transmits an interference time-domain semantic transmission sequence to another wireless communication device (the transmitting end or other wireless communication device) in the wireless channel through a full-duplex channel.

[0058] For the full-duplex channel W at the receiving end (i.e., full-duplex channel W): the transmission time-domain transmission sequence s transmitted by the sending end (i.e., the second terminal) t The received time-domain transmission sequence is obtained by mixing the direct leakage interference and scattering interference caused by the interference link at the receiving end (i.e., the first terminal).

[0059] Among them, the time-domain transmission sequence s is transmitted within the full-duplex channel W. t The transmitted time-domain transmission sequence s is affected by the interference of the time-domain transmission sequence and the interference of the digital transmission sequence. t The received time-domain transmission sequence is obtained by mixing direct leakage interference and scattering interference caused by interference links at the receiving end. The interference of the interfering link to the receiving link can be expressed as: Where h(·) is the full-duplex channel response function in the digital domain at the receiver, therefore the received time-domain transmission sequence obtained by the receiver... It can be represented as: Where n is a noise sequence.

[0060] For the receiving link at the receiving end (i.e., the first terminal), source reconstruction is performed using a digital-semantic full-duplex signal detector. The structure of the digital-semantic full-duplex signal detector is as follows: Figure 2C The structure shown. Its processing steps are as follows:

[0061] The digital domain interference cancellation module is used to eliminate the received time-domain transmission sequence in the digital domain. The self-interference effect in the process is eliminated, and the received semantic information sequence is output. Receive channel input dimension sequence And optional receive side information sequence

[0062] The digital domain interference cancellation module can be implemented using existing linear or nonlinear filter structures, or constructed using a deep neural network structure. When using existing linear or nonlinear filter structures, it is necessary to consider utilizing the symbol demapping process of Orthogonal Frequency Division Multiplexing (OFDM) to convert the time-domain signal into a frequency-domain signal. When constructing using a deep neural network, it can be implemented using, but is not limited to, convolutional neural networks, deep learning model (Transformer) structures, fully connected neural network structures, or self-designed neural network structures. Nonlinear activation functions are used to implement the nonlinear calculations of the neural network to receive the time-domain transmission sequence. and interference time-domain transmission sequence As input, estimate the received semantic transmission sequence after interference cancellation carried in the frequency domain. or Subsequently, the received semantic transmission sequence after interference cancellation or Separate into sequences of received semantic information Receive channel input dimension side information sequence And optional receive side information sequence Among them, the received channel input dimension side information sequence The received channel input dimension sequence needs to be processed through demodulation, channel decoding, and optional decompression operations. Make an estimate.

[0063] Based on the received side information sequence Using the hyper-prior entropy estimation synthesis module h s Determine the corresponding mean information and standard deviation information Among them, the hyperprior entropy estimation analytical module h s The network is constructed using a deep neural network structure, which can typically be implemented using network structures including but not limited to convolutional neural networks, deep learning models (Transformer) structures, and fully connected neural network structures. Nonlinear activation functions are used to achieve nonlinear computation within the neural network, and additional neural network structures with dimensionality-upgrading capabilities are added to utilize the received side information sequence. Estimate the probability information p of the semantic feature sequence.

[0064] Using the semantic domain information detection module, the source information is reconstructed using a direct detection method. Among them, the semantic domain information detection module adopts, for example, Figure 2DThe semantic domain information detection module shown performs semantic information detection and source information reconstruction. Since the interference link does not employ a semantic communication transmission link design, it cannot provide other useful information for the semantic interference cancellation module to perform semantic interference cancellation. Therefore, only a direct detection method can be used to detect the semantic feature sequence. And reconstruct the source information The processing procedure is as follows:

[0065] When the semantic interference cancellation module is not used, the direct detection method is used to process the received semantic information sequence. Perform semantic feature detection and source information reconstruction (at this time) Figure 2D In The processing steps are as follows: First, the received channel input symbol dimension sequence is... and optional mean information and standard deviation information As a condition, to receive a sequence of semantic information As input, the joint source-channel decoding module f d For the received semantic information sequence Joint source-channel decoding is performed to obtain the received semantic feature sequence after semantic information detection. Then, the semantic feature synthesis module g is used. s , will receive semantic feature sequence To reconstruct the source information and obtain the reconstructed source information. .in:

[0066] Joint source channel decoding module f d Deep neural networks are used for construction, and can typically be implemented using neural network structures including, but not limited to, convolutional neural networks, deep learning models (Transformer) structures, and fully connected neural network structures. After loading the model parameters, the process first involves retrieving the received semantic information sequence according to the input symbol dimension sequence of the receiving channel. The dimensional sequence of the received semantic information sequence Each semantic information subsequence The data is divided and then processed using the joint source-channel decoding module f. d Dimension recovery is performed on each of the partitioned parts to achieve the desired result. Each of them Converted into corresponding reconstructed semantic features Thus, the received semantic feature sequence is obtained.

[0067] Semantic feature synthesis module g sThe system employs deep neural network structures, typically including but not limited to convolutional neural networks, deep learning models (Transformer), and fully connected neural network structures. Nonlinear activation functions are used to achieve nonlinear computation within the neural network, and additional neural network structures with dimensionality-upgrading capabilities are added to enable the processing of source information. The reconstruction. After loading the network parameters, the semantic feature synthesis module g can be used. s For the received semantic feature sequence Perform source information reconstruction and obtain reconstructed source information.

[0068] This enables a source transmission structure that is compatible with both digital and semantic transmission links, where the transmission and receiving links are semantic communication links, and the interference link is a digital communication link.

[0069] In some embodiments, the interference information sequence includes an interference side information sequence and / or an interference channel input symbol dimension sequence and / or an interference semantic feature sequence.

[0070] Step 102 includes:

[0071] Step B1: Extract semantic features from the first source information to obtain an interference semantic feature sequence.

[0072] Step B2: Extract side information from the interference semantic feature sequence to obtain the interference side information sequence.

[0073] Step B3: Based on the interference edge information sequence, perform super-prior entropy estimation analysis to determine the mean and standard deviation information corresponding to each interference semantic feature in the interference semantic feature sequence, and obtain the interference probability information sequence through the conditional Gaussian probability calculation function.

[0074] Step B4: Based on the interference probability information sequence, obtain the interference channel input symbol dimension sequence through a quantization function.

[0075] Step B5: Dimensionally reduce the interference semantic feature sequence according to the interference channel input symbol dimension sequence to obtain the interference semantic feature sequence.

[0076] In the above scheme, such as Figure 2B as well as Figure 2H As shown, the receiving end (i.e., the first terminal) uses an interference link semantic encoder to process the first source information x. I The signal modulation process yields an interference information sequence including the interference side information sequence z. I and / or interference channel input symbol dimension sequence k Iand / or interfering semantic feature sequences s I .

[0077] Using semantic feature extraction module Regarding the first source information x I Extracting the semantic feature sequence y from the interference I Among them, the semantic feature extraction module The system employs a deep neural network structure, typically including but not limited to convolutional neural networks, deep learning models (Transformer), and fully connected neural network structures. Non-linear activation functions are used to achieve non-linear computation within the neural network, and dimensionality reduction capabilities are added to optimize the processing of interfering semantic feature sequences y. I Extraction. After loading the network parameters, the semantic feature extraction module can be used. For the high-dimensional first source information x that needs to be sent I Semantic feature extraction is performed to obtain the corresponding interfering semantic feature sequence y. I .

[0078] Using the analytical module for estimating prior entropy Extracting interference edge information sequence z I Among them, the super-prior entropy estimation analytical module The network is constructed using a deep neural network structure, typically including but not limited to convolutional neural networks, deep learning models (Transformer) structures, and fully connected neural network structures. Non-linear activation functions are used to implement non-linear computations within the neural network, and dimensionality reduction features are added to further reduce interference in the semantic feature sequence y. I Redundant information in the data. After loading the network parameters, the super-prior entropy estimation parsing module can be used. For the interfering semantic feature sequence y I Side information is extracted, and the corresponding lower-dimensional interference side information sequence z is obtained. I .

[0079] Synthesis Module Based on Prior Entropy Estimation Calculate the interference semantic feature sequence y I The corresponding mean sequence μ I , standard deviation sequence σ I and interference probability information sequence p I Among them, the super-prior entropy estimation analytical module The network is constructed using a deep neural network structure, typically including but not limited to convolutional neural networks, deep learning models (Transformer) structures, and fully connected neural network structures. Nonlinear activation functions are used to achieve nonlinear computation within the neural network, and additional neural network structures with dimensionality-upgrading capabilities are added to utilize the interference edge information sequence z. I Estimating the interference probability information sequence p I After loading the network parameters, the hyper-prior entropy estimation synthesis module can be used. Using the input interference side information sequence z I Probability estimation is then performed to obtain the corresponding mean sequence μ. I , standard deviation sequence σ I and interference probability information sequence p I Among them, the interference probability information sequence p I It is necessary to target the interfering semantic feature sequence y I Each value y in I,j , with mean sequence μ I With standard deviation sequence σ I Each value in (u) I,j ,σ I,j Given the condition y, the following conditional Gaussian probability estimation function p(y) is used. I.j |u I,j ,σ I,j Calculate each value of y in y. I,j probability value

[0080]

[0081] Finally, each probability value is combined into a sequence to obtain the interference probability information sequence p. I .

[0082] Using the rate adaptation module For the semantic feature sequence y I Determine the input symbol dimension sequence k of the interference channel I Among them, the rate adaptation module Using the interference probability information sequence p I Convert to the input symbol dimension sequence k of the interference channel I Specifically, the rate adaptation module The interfering semantic feature sequence y can be used I Divided into L groups (L≥1), and based on this partitioning method, the interference probability information sequence p is... I Divided into corresponding number of parts For each part, calculate the corresponding number of output symbols according to the following formula:

[0083]

[0084] Here, the Q(·) function is the quantization function, which quantizes the input -η·log2p. l The result is quantized to a similar integer value; η is the mapping ratio, which will represent the interference probability information sequence p. l Sequence information content - log2p l Converted into the corresponding number of channel input symbols. As a sequence of input symbols for channel reduction, it is output outside the module.

[0085] Using joint source-channel coding module The interference semantic feature sequence y I Converted into a sequence of interference semantic information s that can be input to the channel I Among them, the joint source-channel coding module It is constructed using deep neural networks, which can typically be implemented using neural network structures including, but not limited to, convolutional neural networks, deep learning model (Transformer) structures, and fully connected neural network structures. It uses the perturbation semantic feature sequence y. I As model input, the rate adaptation module is used for each y I The dimension sequence k of the input symbols in the interference channel is determined by the partition. I , will y I Each part [y] 1 ,…,y L According to [k] 1 ,…,k L Dimensionality reduction is performed, and the sequence of interfering semantic information s is obtained. I s I That is, s = [s 1 ,…,s L ], where the output interference semantic information sequence s for the l-th part l Its dimension is k l .

[0086] Using the signal mapping module, the interference semantic information sequence s I Interference channel input dimension sequence k I And optional interference side information sequence z I Mapped to interference temporal semantic transmission sequence Wherein, the interference channel input dimension sequence k I Channel coding is needed for error protection and modulation to obtain the input dimension sequence k of the interference channel. I The sequence of interfering semantic information s I Interference channel input dimension side information sequence k′ Iand optional interference side information sequence z I Integrate into an interference semantic transmission sequence (s I ,k′ I ) or (s I ,k′ I ,z I The signal mapping module maps the waveform mapping process based on Orthogonal Frequency Division Multiplexing (OFDM) technology to the time-frequency resources occupied by the transmitting end, thereby achieving complete multiplexing within the same time-frequency resource region. This mapping process maps the interference semantic transmission sequence (s) carried in the frequency domain to the time-frequency resources. I ,k′ I ) or (s I ,k′ I ,z I Interference time-domain transmission sequence converted to the time domain

[0087] In some embodiments, step 105 includes:

[0088] Step C1: Obtain the input symbol dimension sequence and interference probability information sequence of the interference channel.

[0089] Step C2 involves performing interference cancellation on the received time-domain transmission sequence and the interference time-domain transmission sequence to obtain a received semantic information sequence and / or a received channel input dimension sequence and / or a received side information sequence.

[0090] Step C3: Based on the interference channel input symbol dimension sequence and the interference probability information sequence as semantic interference cancellation conditions, perform semantic interference cancellation on the received semantic information sequence to obtain the interference-cancelled received semantic information sequence.

[0091] Step C4: Based on the received side information sequence and the received channel input dimension sequence as joint source channel decoding conditions, perform joint source channel decoding on the interference-cancelled received semantic information sequence to obtain received semantic feature information.

[0092] Step C5: Perform source reconstruction on the received semantic feature information to obtain the source reconstruction information.

[0093] In the above scheme, Figure 2B and Figure 2H The structure of the digital-semantic full-duplex signal detector is as follows: Figure 2C As shown, Figure 2C The structure of the semantic information detection module in the middle is as follows: Figure 2D As shown.

[0094] The transmission time-domain transmission sequence s transmitted by the sending end (i.e., the second terminal)t Interference time-domain transmission sequence of the receiving end (i.e., the first terminal) link The resulting direct leakage interference and scattering interference are mixed to obtain the received time-domain semantic transmission sequence. The interference of the interfering link to the receiving link can be expressed as: Where h(·) is the full-duplex channel response function in the digital domain at the receiving end (i.e., the first terminal), therefore, the received time-domain semantic transmission sequence obtained by the receiving end... It can be represented as: Where n is a noise sequence.

[0095] For the receiving link at the receiving end (i.e., the first terminal), the structure of the digital-semantic full-duplex signal detector is as follows: Figure 2C As shown, the processing procedure includes:

[0096] The digital domain interference cancellation module is used to eliminate the received time-domain semantic transmission sequence in the digital domain. The self-interference effect in the process is eliminated, and the received semantic information sequence is output. Receive channel input dimension sequence And optional receive side information sequence

[0097] The digital domain interference cancellation module can be implemented using existing linear or nonlinear filter structures, or constructed using a deep neural network structure. When using existing linear or nonlinear filter structures, it is necessary to consider utilizing the symbol demapping process of Orthogonal Frequency Division Multiplexing (OFDM) to convert the time-domain signal into a frequency-domain signal. When constructing using a deep neural network, it can be implemented using, but is not limited to, convolutional neural networks, deep learning model (Transformer) structures, fully connected neural network structures, or self-designed neural network structures. Nonlinear activation functions are used to implement the nonlinear calculations of the neural network to receive the time-domain semantic transmission sequence. and interference time-domain transmission sequence As input, estimate the received semantic transmission sequence after interference cancellation carried in the frequency domain. or Subsequently, the received semantic transmission sequence after interference cancellation or Separate into sequences of received semantic information Receive channel input dimension side information sequence And optional receive side information sequence Among them, the received channel input dimension side information sequence The received channel input dimension sequence needs to be processed through demodulation, channel decoding, and optional decompression operations. Make an estimate.

[0098] Using the hyper-prior entropy estimation synthesis module h s Obtain the corresponding mean sequence with standard deviation series After loading the network parameters, the hyper-prior entropy estimation module h can be used. s Using the input receiver side information sequence Perform probability estimation and obtain the corresponding mean sequence. With received standard deviation sequence (or receive variance sequence) ).

[0099] Using a semantic domain information detection module, methods such as direct detection or semantic interference cancellation are employed to analyze the received semantic feature sequence. The detection is performed to reconstruct the source information. Among them, the semantic domain information detection module receives the channel input symbol dimension sequence. and optional mean series with standard deviation series Optional interference channel input symbol dimension sequence k I Optional interference probability information sequence p I And optional interfering semantic feature sequence y I As a condition, to receive a sequence of semantic information As input, source reconstruction is performed using methods such as direct detection or semantic interference cancellation.

[0100] Figure 2C The structure of the semantic information detection module in the middle is as follows: Figure 2D As shown, this includes an optional semantic interference cancellation module and a joint source-channel decoding module f. d and semantic feature integration module g s constitute.

[0101] For the optional semantic interference cancellation module:

[0102] When the semantic interference cancellation module is not used, the direct detection method is used to process the received semantic information sequence. Perform semantic feature detection and source information reconstruction (at this time) Figure 2D In The processing procedure is as follows: First, the input symbol dimension sequence of the received channel is... and optional mean series with standard deviation series As a condition, to receive a sequence of semantic information As input, the joint source-channel decoding module f d The received semantic feature sequence after semantic information detection is obtained. Then, the semantic feature synthesis module g is used. s The semantic feature sequence will be received. To reconstruct the source information and obtain the reconstructed source information.

[0103] Among them, the joint source channel decoding module f d Deep neural networks are used for construction, and can typically be implemented using neural network structures including, but not limited to, convolutional neural networks, deep learning models (Transformer) structures, and fully connected neural network structures. After loading the model parameters, the process first involves retrieving the received semantic information sequence according to the input dimension sequence of the receiving channel. The dimensional sequence of the received semantic information sequence Each semantic information subsequence The data is divided and then processed using the joint source-channel decoding module f. d Dimension recovery is performed on each part l, thereby... Each Convert to reconstruct semantic features Thus, the received semantic feature sequence is obtained.

[0104] Semantic feature synthesis module g s The system employs deep neural network structures, typically including but not limited to convolutional neural networks, deep learning models (Transformer), and fully connected neural network structures. Nonlinear activation functions are used to achieve nonlinear computation within the neural network, and additional neural network structures with dimensionality-upgrading capabilities are added to enable the processing of source information. The reconstruction. After loading the network parameters, the semantic feature synthesis module g can be used. s For receiving semantic feature information To reconstruct the source information, eliminate interference, and obtain the reconstructed source information.

[0105] When using a semantic interference cancellation module, the semantic interference cancellation method is used to cancel the received semantic information sequence. Semantic interference cancellation, semantic feature detection, and source information reconstruction are performed. The process involves first processing the input symbol dimension sequence k from the interference channel. I And optional interference probability information sequence p I With optional interfering semantic feature sequence y I As a condition, to receive a sequence of semantic information As input, semantic interference cancellation is performed using a semantic interference cancellation module to obtain the received semantic information sequence after interference cancellation. Then, using the direct detection method described above, the received semantic information sequence after interference cancellation is used. Using this as input, semantic information detection and source reconstruction are performed.

[0106] In some embodiments, step 1053 includes:

[0107] Step D1: Based on the interference channel input symbol dimension sequence and the interference probability information sequence as joint source channel decoding conditions, perform joint source channel decoding on the received semantic information sequence to obtain the interference detection semantic feature sequence.

[0108] Step D2: Reconstruct the source information of the interference detection semantic feature sequence to obtain interference detection source information.

[0109] Step D3: Dimensionally reduce the interference detection source information according to the interference channel input symbol dimension sequence to obtain the interference reconstruction semantic information sequence.

[0110] Step D4 involves performing interference cancellation processing on the reconstructed semantic information sequence and the received semantic information sequence to obtain the interference-cancelled received semantic information sequence. Alternatively,

[0111] Step E1: Based on the interference channel input symbol dimension sequence and the interference probability information sequence as joint source channel decoding conditions, perform joint source channel decoding on the received semantic information sequence to obtain the interference detection semantic feature sequence.

[0112] Step E2: Dimensionally reduce the interference detection semantic feature sequence according to the interference channel input symbol dimension sequence to obtain the interference reconstruction semantic information sequence.

[0113] Step E3 involves performing interference cancellation processing on the interference-reconstructed semantic information sequence and the received semantic information sequence to obtain the interference-cancelled received semantic information sequence. Alternatively,

[0114] Step F1: Obtain the sequence of interfering semantic features.

[0115] Step F2: Dimensionally reduce the interference semantic feature sequence according to the interference channel input symbol dimension sequence to obtain the interference reconstruction semantic information sequence.

[0116] Step F3: Based on the reconstructed semantic information sequence of the interference and the received semantic information sequence, the received semantic information sequence after interference cancellation is obtained through a nonlinear activation function.

[0117] In the above scheme, Figure 2DThe structure of the semantic interference cancellation module in the middle can be as follows: Figure 2E As shown, or as Figure 2F As shown, or as Figure 2G As shown.

[0118] like Figure 2E As shown, the processing procedure is as follows:

[0119] First, input the symbol dimension sequence k of the interference channel. I And optional interference probability information sequence p I As a condition, interference detection is used in conjunction with a source-channel decoder. Perform interference semantic information detection to obtain interference detection semantic feature sequences. Then utilize the semantic feature synthesis module g s To reconstruct the interference source information and obtain the interference detection source information.

[0120] Then, the semantic feature parsing module g is used. a Perform interference source feature reconstruction to obtain the interference reconstruction semantic feature sequence. Then, using the interference channel input symbol dimension sequence k I As a condition, the joint source channel coding module is reconstructed using interference. Perform interference signal reconstruction to obtain the interference reconstruction semantic information sequence.

[0121] Finally, the received semantic information sequence after interference cancellation is obtained using the following formula.

[0122] The interference cancellation method expressed in this formula can also be replaced by a deep residual network structure, where the deep neural network f inside the residual network structure... c This can be implemented using network structures including, but not limited to, convolutional neural networks, deep learning models (Transformer) structures, and fully connected neural network structures, and nonlinear activation functions can be used to achieve nonlinear computation in the neural network, such as... Figure 2G The structure within the dashed box is shown. Its processing method can be represented as:

[0123]

[0124] When using a residual network structure, its internal parameters need to be jointly fine-tuned with the parameters of other network structures within the overall network structure, or fine-tuned individually under the condition that the parameters of other network structures are fixed, in order to optimize the network parameters by minimizing the distortion of the reconstructed source information as the loss function.

[0125] Or such as Figure 2F As shown, the processing procedure is as follows:

[0126] First, input the symbol dimension sequence k of the interference channel. I And optional interference probability information sequence p I As a condition, interference detection is used in conjunction with a source-channel decoder. Perform interference semantic information detection to obtain interference detection semantic feature sequences.

[0127] Then, the interference channel input symbol dimension sequence k I As a condition, the joint source channel coding module is reconstructed using interference. Perform interference signal reconstruction to obtain the interference reconstruction semantic information sequence.

[0128] Finally, the received semantic information sequence after interference cancellation is obtained using the following formula.

[0129] The interference cancellation method expressed in this formula can also be replaced by a deep residual network structure, where the deep neural network f inside the residual network structure... c This can be implemented using network structures including, but not limited to, convolutional neural networks, deep learning models (Transformer) structures, and fully connected neural network structures, and nonlinear activation functions can be used to achieve nonlinear computation in the neural network, such as... Figure 2G The structure within the dashed box is shown. Its processing method can be represented as:

[0130]

[0131] When using a residual network structure, its internal parameters need to be jointly fine-tuned with the parameters of other network structures within the overall network structure, or fine-tuned individually under the condition that the parameters of other network structures are fixed, in order to optimize the network parameters by minimizing the distortion of the reconstructed source information as the loss function.

[0132] In addition, adopting such Figure 2F The structure shown reduces the interference between the reconstruction of source information and the reconstruction of semantic information sequences, and theoretically has advantages over other structures. Figure 2E The structure shown has better interference cancellation capability, thereby further achieving better quality of transmitted source reconstruction.

[0133] Or such as Figure 2G As shown, the processing procedure is as follows:

[0134] First, take the interfering semantic feature sequence y I As input, the sequence of symbols k in the interference channel is used as input. I As a condition, the joint source channel coding module is reconstructed using interference. Obtain the semantic information sequence of interference reconstruction

[0135] Then, to receive the semantic information sequence Reconstructing semantic information sequences from interference As input, the received semantic information sequence after interference cancellation is finally obtained using the following formula.

[0136] The interference cancellation method expressed in this formula can also be replaced by a deep residual network structure, where the deep neural network f inside the residual network structure... c This can be implemented using network structures including, but not limited to, convolutional neural networks, deep learning models (Transformer) structures, and fully connected neural network structures, and nonlinear activation functions can be used to achieve nonlinear computation in the neural network, such as... Figure 2G The structure within the dashed box is shown. Its processing method can be represented as:

[0137]

[0138] When using a residual network structure, its internal parameters need to be jointly fine-tuned with the parameters of other network structures within the overall network structure, or fine-tuned individually under the condition that the parameters of other network structures are fixed, in order to optimize the network parameters by minimizing the distortion of the reconstructed source information as the loss function.

[0139] It should be noted that, adopting such Figure 2G The method shown reduces the semantic feature reconstruction process by directly using the existing interfering semantic feature sequence y. I As input in the process of reconstructing interfering semantic information sequences, it theoretically has advantages over other methods. Figure 2F The structure shown has better interference cancellation capability, thereby further achieving better quality of transmitted source reconstruction.

[0140] In some embodiments, step 105 includes:

[0141] Step G1: Obtain the interference semantic feature sequence, the interference channel input symbol dimension sequence, and the interference probability information sequence.

[0142] Step G2 involves performing interference cancellation on the received time-domain transmission sequence and the interfering time-domain transmission sequence to obtain the received digital information sequence.

[0143] Step G3: Based on the interference semantic feature sequence, the interference channel input symbol dimension sequence, and the interference probability information sequence as semantic interference cancellation conditions, semantic interference cancellation is performed on the received digital information sequence to obtain the interference-cancelled received digital information sequence.

[0144] Step G4 involves demodulating and channel decoding the received digital information sequence after interference cancellation to obtain the received source bit sequence to be decoded.

[0145] Step G5: Perform source reconstruction on the received source bit sequence to be decoded to obtain the source reconstruction information.

[0146] In the above scheme, Figure 2H The structure of the digital-semantic full-duplex signal detector shown is as follows: Figure 2I As shown, Figure 2I The structure of the semantic interference cancellation auxiliary digital signal detection module is as follows: Figure 2J As shown, where, Figure 2H The first terminal uses an interference link semantic encoder, while the second terminal uses a digital encoder, forming a source transmission structure for full-duplex communication that is compatible with both digital and semantic transmission links.

[0147] like Figure 2I As shown, the structure of the receiving link at the receiving end (i.e., the first terminal) using a digital-semantic full-duplex signal detector is as follows: Figure 2I As shown, the processing procedure is as follows:

[0148] The digital domain interference cancellation module is used to eliminate the received time-domain digital transmission sequence in the digital domain. The self-interference effect in the output is eliminated, and the received digital information sequence is output. The semantic interference cancellation-assisted digital signal detection module uses methods such as direct detection or semantic interference cancellation to detect the received digital information sequence. The semantic interference information in the source is eliminated, and then the source information is reconstructed by decoding the digital source channel.

[0149] Among them, the semantic interference cancellation-assisted digital signal detection module uses the interference channel input symbol dimension sequence k. I and optional interference probability information sequence p I And optional interfering semantic feature sequence y I As a condition, to receive a sequence of semantic information As input, the received digital information sequence is processed using methods such as direct detection or semantic interference cancellation. The semantic interference information in the source is eliminated, and then the source information is reconstructed by decoding the digital source channel. The internal structure of the semantic interference cancellation module can be selected from... Figure 2E , Figure 2F or Figure 2G Any one of the structures in the dataset can be used to obtain the received digital information sequence after interference cancellation.

[0150] like Figure 2J As shown, the semantic interference cancellation-assisted digital signal detection module includes a semantic interference cancellation module and a channel decoding module D. c and source decoding module D s The processing procedure is as follows:

[0151] First, let k be the input symbol dimension sequence of the interference channel. I And optional interference probability information sequence p I With optional interfering semantic feature sequence y I As a condition, to receive a sequence of semantic information As input, semantic interference cancellation is performed using a semantic interference cancellation module to obtain the received digital information sequence after interference cancellation. The internal structure of the semantic interference cancellation module can be selected from... Figure 2E , Figure 2F or Figure 2G Any one of the structures in the dataset can be used to obtain the received digital information sequence after interference cancellation.

[0152] Subsequently, using the channel decoder D corresponding to the channel encoder in the transmission link... c The received digital information sequence after interference cancellation Demodulation and channel decoding are performed to obtain the received source bit sequence to be decoded.

[0153] Finally, using the source decoder D corresponding to the source encoder in the transmission link... s For the received source bit sequence to be decoded Decode the source information to obtain reconstructed source information.

[0154] Based on the same inventive concept, another embodiment of this application also provides a source transmission method based on full-duplex communication, applied to a second terminal, such as... Figure 3 As shown, the method includes:

[0155] Step 301: Obtain the second source information.

[0156] In this step, the second source information refers to the source information transmitted by the second terminal. Before transmitting the second source information, the type of source information to be transmitted is determined, including but not limited to high-dimensional source types such as text sources, voice sources, image sources, and video sources, or basic binary or multi-dimensional independent or related sources. Subsequently, the second terminal can select, acquire, or generate the corresponding second source information based on the type of source information to be transmitted.

[0157] The second terminal can be either a sending terminal or a receiving terminal, but a sending terminal is preferred here.

[0158] Among them, the source type of the second source information x of the second terminal is the same as that of the first source information x of the first terminal. I The types of information sources can be the same or different.

[0159] Step 302: Encode and modulate the second source information to obtain a transmission sequence, and obtain a transmission time-domain transmission sequence based on the transmission sequence through an inverse Fourier transform function.

[0160] In this step, the second source information is encoded and modulated to convert the second source information to be transmitted into a transmission sequence.

[0161] Based on the transmission sequence, the inverse Fourier transform function is used to map the frequency domain transmission sequence to the time domain, forming the transmission time domain transmission sequence.

[0162] Step 303: The transmission time-domain sequence is sent to a full-duplex transmission channel. The full-duplex transmission channel receives the transmission time-domain sequence and simultaneously receives an interference time-domain transmission sequence sent by the first terminal. When the interference time-domain transmission sequence and the transmission time-domain transmission sequence are transmitted in the full-duplex transmission channel, the interference time-domain transmission sequence and the transmission time-domain transmission sequence will interfere and overlap with each other. The transmission time-domain transmission sequence is interfered with and overlapped by the interference time-domain transmission sequence to obtain the reception time-domain transmission sequence.

[0163] In this step, the transmit time-domain transmission sequence is sent to the full-duplex transmission channel. When the transmit time-domain transmission sequence and the interference time-domain transmission sequence sent by the first terminal are transmitted in the full-duplex transmission channel, the transmit time-domain transmission sequence transmitted by the second terminal will be affected by the direct leakage interference and scattering interference caused by the interference time-domain transmission sequence transmitted by the first terminal. The transmit time-domain transmission sequence after interference and aliasing forms the receive time-domain transmission sequence.

[0164] The above scheme involves encoding and modulating the acquired second source information to obtain a transmission sequence. Then, based on the transmission sequence, an inverse Fourier transform function is used to obtain a transmission time-domain transmission sequence, which is then sent to a full-duplex transmission channel. Within the full-duplex channel, the transmission time-domain transmission sequence sent by the second terminal is interfered with and aliased by the interference time-domain transmission sequence to obtain a reception time-domain transmission sequence. This sequence provides basic data for the first terminal to reconstruct the source and eliminate interference.

[0165] In some embodiments, the transmission sequence includes a digital transmission sequence.

[0166] Step 302 includes:

[0167] Step H1: Compress the second source information to obtain a compressed bit sequence of the source.

[0168] Step H2 involves channel coding and modulation of the source compressed bit sequence to obtain a digital transmission sequence.

[0169] In the above scheme, such as Figure 2G As shown, the transmission sequence includes a digital transmission sequence s. The transmitting end (i.e., the second terminal) uses a digital encoder, while the first terminal (i.e., the receiving end) uses an interference link semantic encoder, forming a source transmission structure for full-duplex communication compatible with both digital and semantic transmission links. The digital encoder includes a source encoder E. s and channel encoder E c Using source encoder E s The second source information x is compressed without distortion or with limited distortion, and the compressed bit sequence b is output.

[0170] For second source information x of different source types, a corresponding and applicable digital source encoder E is used. s For example, when the source type of the second source information x is a text source, Huffman coding is used for lossless source coding compression. When the source type of the second source information x is an image source, a video compression (JPEG2000) or Better Portable Graphics (BPG) image encoder is used for lossy source coding compression. When the source type of the second source information x is a video source, an upgraded digital video compression format (H.264 / AVC) or digital video compression format (H.265 / HEVC) video encoder is used for lossy source coding compression, etc.

[0171] Using channel encoder E c Demodulate and decode x to output a digital transmission sequence s.

[0172] Then, using the signal mapping module, the digital transmission sequence s is converted into a time-domain digital transmission sequence s. t Time-domain digital transmission sequence s t It will be sent to the full-duplex channel W.

[0173] In some embodiments, the transmission sequence includes a side information sequence and / or a channel input symbol dimension sequence and / or a semantic information sequence.

[0174] Step 302 includes:

[0175] Step I1: Extract semantic features from the second source information to obtain a semantic feature sequence.

[0176] Step I2: Extract side information from the semantic feature sequence to obtain a side information sequence.

[0177] Step I3: Perform super-prior entropy estimation analysis based on the edge information sequence to determine the mean and standard deviation information corresponding to each semantic feature sequence in the semantic feature sequence, and obtain the probability information sequence through the conditional Gaussian probability calculation function.

[0178] Step I4: Based on the probability information sequence, obtain the channel input symbol dimension sequence through a quantization function.

[0179] Step I5: Reduce the dimensionality of the semantic feature sequence according to the channel input symbol dimension sequence to obtain the semantic information sequence.

[0180] In the above scheme, such as Figure 2A and Figure 2B As shown, the transmitting end (i.e. the second terminal) uses a semantic encoder. After the semantic encoder stacks the second source information x, it performs signal modulation processing. The resulting transmission sequence includes the side information sequence z and / or the channel input symbol dimension sequence k and / or the semantic information sequence s.

[0181] Using the semantic feature extraction module g a For the second source information x, a semantic feature sequence y is extracted, where the semantic feature extraction module g... a The system employs a deep neural network structure, typically including but not limited to convolutional neural networks, deep learning models (Transformer), and fully connected neural network structures. Non-linear activation functions are used to achieve non-linear computation within the neural network, and dimensionality reduction features are added to extract the semantic feature sequence y. After loading the network parameters, the semantic feature extraction module g can be used... a Semantic features are extracted from the high-dimensional second source information x that needs to be sent to obtain the corresponding semantic feature sequence y.

[0182] Using the prior entropy estimation analytical module h a Extract the edge information sequence z, where the super-prior entropy estimation parsing module h a The network is constructed using a deep neural network structure, typically including but not limited to convolutional neural networks, deep learning models (Transformer) structures, and fully connected neural network structures. Nonlinear activation functions are used to achieve nonlinear computation within the neural network, and dimensionality reduction features are added to further reduce redundant information in the semantic feature sequence y. After loading the network parameters, the hyper-prior entropy estimation parsing module h can be used. aWe extract edge information from the semantic feature sequence y and obtain the corresponding lower-dimensional edge information sequence z.

[0183] Using the hyper-prior entropy estimation synthesis module h s Calculate the mean sequence μ, standard deviation sequence σ, and probability information sequence p corresponding to the semantic feature sequence y, where the prior entropy estimation parsing module h... s The network is constructed using a deep neural network structure, typically including but not limited to convolutional neural networks, deep learning models (Transformer), and fully connected neural network structures. Nonlinear activation functions are used to achieve nonlinear computation within the neural network, and a neural network structure with dimensionality-upgrading capabilities is added. This allows the probability information sequence p to be estimated using the edge information sequence z. After loading the network parameters, the hyper-prior entropy estimation module h can be used. s By utilizing the input side information sequence z and performing probability estimation, the corresponding mean sequence μ, standard deviation sequence σ, and interference probability information sequence p are obtained. The probability information sequence p requires each value y in the semantic feature sequence y to be considered. j For each value in the mean sequence μ and the standard deviation sequence σ (u j ,σ j Given the condition y, the following conditional Gaussian probability estimation function p(y) is used. j |u j ,σ j Calculate each value of y in y. j The probability value:

[0184]

[0185] Finally, each probability value is combined into a sequence to obtain the probability information sequence p.

[0186] Using the rate adaptation module R a The semantic feature sequence y determines the channel input symbol dimension sequence k, where the rate adaptation module R a The probability information sequence p is converted into a channel input symbol dimension sequence k. Specifically, the rate adaptation module R... a The semantic feature sequence y can be divided into L groups y = [y 1 ,…,y L ](L≥1), and based on this partitioning method, the probability information sequence p is divided into a corresponding number of parts p = [p 1 ,…,p L For each part, calculate the corresponding number of output symbols according to the following formula: k l =Q(-η·log2p) l ),

[0187] Here, the Q(·) function is the quantization function, which quantizes the input -η·log2p. l The result is quantized to a similar integer value; η is the mapping ratio, which will represent the probability information sequence p. l Sequence information content - log2p l Convert this to the corresponding channel input symbol number. Then k = [k 1 ,…,k L As a sequence of input symbols for channel reduction, it is output to the outside of the module.

[0188] Using the joint source-channel coding module f e The semantic feature sequence y is converted into a semantic information sequence s that can be input to the channel, wherein the joint source-channel coding module f e It is constructed using deep neural networks, typically employing neural network structures including, but not limited to, convolutional neural networks, deep learning models (Transformer) structures, and fully connected neural network structures. It takes a semantic feature sequence y as the model input, and the rate adaptation module adapts it to each y... I The channel input symbol dimension sequence k, determined by the partition, will y I Each part [y] 1 ,…,y L According to [k] 1 ,…,k L ] Perform dimensionality reduction and obtain the semantic information sequence s, s is s = [s 1 ,…,s L ], where the output semantic information sequence s for the l-th part l Its dimension is k l .

[0189] Using the signal mapping module, the semantic information sequence s, the channel input dimension sequence k, and the optional side information sequence z are mapped to the temporal semantic transmission sequence s. t In this process, the channel input dimension sequence k needs to be error-protected and modulated using channel coding to obtain the channel input symbol dimension side information sequence k′. The semantic information sequence s, the channel input symbol dimension side information sequence k′, and the optional side information sequence z are integrated into a semantic transmission sequence (s,k′) or (s,k′,z), and mapped onto one-end time-frequency resources. The signal mapping module, based on the waveform mapping process of Orthogonal Frequency Division Multiplexing (OFDM), converts the frequency-domain semantic transmission sequence (s,k) or (s,k,z) into a time-domain transmission time-domain semantic transmission sequence s. t .

[0190] The channel input symbol dimension sequence k can first be compressed and encoded without distortion to obtain the channel input dimension compressed sequence k. Then, channel coding is used for error protection and modulation to obtain the channel input symbol dimension side information sequence k′, thereby reducing the transmission length of the channel input symbol dimension side information sequence k′.

[0191] The optional side information sequence can be processed using the same digital encoding as the channel input symbol dimension sequence k, that is, the side information sequence z′ to be transmitted is obtained through the processing method of quantization-lossless compression coding-channel coding-modulation.

[0192] In some embodiments, validation data results for a semantic full-duplex method based on an image source transmission task are provided. Experimental conditions include:

[0193] The selected semantic full-duplex scenario is a scenario where the interference link is a digital communication link scenario. The selected full-duplex transmission source transmission structure is a full-duplex communication source transmission structure that is compatible with both digital transmission links and semantic transmission links. Its sending link and receiving link are semantic communication links, and its interference link is a digital communication link.

[0194] The selected comparison method is the digital full-duplex method, in which the transmitting link, receiving link, and interference link are all digital communication links.

[0195] The image data used for training the semantic communication model and online testing are all from the OpenImages dataset. The face images in the dataset are changed to 3*128*128 images through operations including but not limited to cropping and scaling. Here, 3 represents the R, G and B channels of the image, and 128*128 represents the image resolution (i.e., the number of pixels in height * the number of pixels in width).

[0196] The rate adaptation module in the semantic communication model used needs to divide the semantic feature sequence into regions and adapt the channel transmission rate to each region. The compression ratio of the symbol dimension of the semantic information sequence s in the transmission link to the pixel dimension of the original image source is about 0.25, and the compression ratio of the corresponding side information sequence z dimension to the pixel dimension of the image source is about 0.03, that is, the overall compression ratio is 0.28.

[0197] The overall compression ratio of the transmitting digital encoder used for comparison is equal to the compression ratio of the semantic communication model used, which is 0.28.

[0198] The power of the received noise is set to 0.1, the power of the received time-domain semantic information sequence is set to 1, and the power range of the interference time-domain digital information sequence is [1e3, 1e5], that is, the range of the received signal-to-interference-plus-noise ratio in the digital domain is between -50dB and -30dB.

[0199] The adaptive interference cancellation normalized minimum mean square (NLMS) algorithm from the existing digital domain interference cancellation methods is selected.

[0200] The multi-scale structural similarity loss function (1-MS-SSIM) is selected as the end-to-end distortion metric during model training.

[0201] The end-to-end reconstructed image quality metric used in the model testing process is the Multiscale Structural Similarity Measure (MS-SSIM) (dB).

[0202] like Figure 2K As shown, this application's full-duplex transmission source transmission structure uses semantic communication links for both the transmitting and receiving links (e.g., Figure 2K The semantic encoder in the text, and the traditional digital full-duplex transmission source transmission structure (such as the traditional digital communication link often used in the transmission and reception links) Figure 2K The comparison of BPG+LDPC(1 / 3)+QPSK and BPG+LDPC(7 / 12)+QPSK in the data shows the performance of the two methods under the quality index of source information reconstruction.

[0203] In the full-duplex transmission source transmission structure, the semantic domain information detection module in the receiving link uses direct detection to perform semantic information detection and source information reconstruction.

[0204] In the traditional digital full-duplex transmission source transmission structure, two coding and modulation configurations are used: BPG+LDPC(1 / 3)+QPSK and BPG+LDPC(7 / 12)+QPSK. BPG is the selected image source distortion encoder, LDPC is the selected channel code, the code rate in parentheses is the channel code configuration, and QPSK indicates that Quadrature Phase Shift Keying (QPSK) is used for modulation.

[0205] This application is able to recover the source image from each signal-to-interference-plus-noise ratio (SIR) value within the test range, obtain the reconstruction quality of the source information, and the quality of the reconstructed image steadily improves with the increase of the SIR.

[0206] The traditional method used for comparison exhibits a significant inflection point in both coding configurations. When the signal-to-interference-plus-noise ratio (SINR) is above this inflection point, the reconstructed image quality using the traditional method fails to improve with increasing SINR. Conversely, when the SINR falls below this inflection point, the reconstructed image quality deteriorates dramatically until it becomes impossible to reconstruct the image, resulting in a quality curve that no longer shows improvement. Specifically, at the same compression ratio, a higher channel coding bitrate results in a higher SINR at the inflection point, and also yields a higher highest quality reconstructed image.

[0207] Compared to the traditional method with encoding configuration of BPG+LDPC(7 / 12)+QPSK, the semantic full-duplex method used can achieve a performance gain of 8dB at the signal-to-interference-plus-noise ratio (SINR) scale under the same MS-SSIM metric; and compared to the traditional method with encoding configuration of BPG+LDPC(1 / 3)+QPSK, this application can also achieve a performance gain of 6dB at the same MS-SSIM metric at the SINR scale.

[0208] like Figure 2L As shown, it indicates that... Figure 2K Under the same test conditions, this application compares the image reconstruction results of a portion of the image transmission using a conventional method for comparison. Black images represent images that could not be reconstructed using the conventional method used for comparison.

[0209] like Figure 2L As shown, this application can obtain reconstructed images within a signal-to-interference-plus-noise ratio (SNR) range of -50dB to -30dB, while traditional methods can only obtain reconstructed images within a higher SNR range. Furthermore, even under strong interference conditions ranging from -50dB to -45dB, this application can still observe some features of the source image in the reconstructed image, demonstrating the effectiveness of the design.

[0210] In addition, such as Figure 2L As shown, at signal-to-interference-plus-noise ratios of -35dB and -30dB, although both the present application and the conventional method can obtain reconstructed images, the reconstructed images obtained by the present application are of higher quality and have clearer features than those obtained by the conventional method, while the reconstructed images obtained by the conventional method are more blurry.

[0211] The above verification data results show that, by using this application, the reliability of the reconstructed source information quality can be improved, and compared with traditional methods, the overall transmission performance of the system can be improved, and the effective spectrum efficiency can be increased, thus verifying that this application is feasible and has high reliability.

[0212] It should be noted that the method in this embodiment can be executed by a single device, such as a computer or server. The method can also be applied in a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method in this embodiment, and the multiple devices will interact with each other to complete the method described.

[0213] It should be noted that the above description describes some embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0214] Based on the same inventive concept, corresponding to any of the above embodiments, this application also provides a source transmission device based on full-duplex communication.

[0215] The source transmission device based on full-duplex communication is applied to the first terminal, and the source transmission device based on full-duplex communication includes:

[0216] The first acquisition module is configured to acquire first information source information;

[0217] The first modulation module is configured to perform encoding and modulation processing on the first source information to obtain an interference information sequence, and to obtain an interference time-domain transmission sequence based on the interference information sequence through an inverse Fourier transform function.

[0218] The first aliasing module is configured to send the interference time-domain transmission sequence to a full-duplex transmission channel. The full-duplex transmission channel receives the interference time-domain transmission sequence and simultaneously receives a transmission time-domain transmission sequence sent by a second terminal. When the interference time-domain transmission sequence and the transmission time-domain transmission sequence are transmitted in the full-duplex transmission channel, the interference time-domain transmission sequence and the transmission time-domain transmission sequence will interfere with and alias each other. The transmission time-domain transmission sequence is interfered with and aliased by the interference time-domain transmission sequence to obtain the received time-domain transmission sequence.

[0219] The receiving module is configured to receive the received time-domain transmission sequence transmitted through the full-duplex transmission channel;

[0220] The source reconstruction module is configured to perform source reconstruction based on the received time-domain transmission sequence and the interference time-domain transmission sequence to obtain source reconstruction information.

[0221] Based on the same inventive concept and the same inventive concept as any of the above-described embodiments of the source transmission method based on full-duplex communication, this application also provides a source transmission device based on full-duplex communication.

[0222] The source transmission device based on full-duplex communication is applied to the second terminal, and the source transmission device based on full-duplex communication includes:

[0223] The second acquisition module is configured to acquire second information source information;

[0224] The second modulation module is configured to perform encoding and modulation processing on the second source information to obtain a transmission sequence, and to obtain a transmission time-domain transmission sequence based on the transmission sequence through an inverse Fourier transform function.

[0225] The second aliasing module is configured to send the transmission time-domain transmission sequence to a full-duplex transmission channel. The full-duplex transmission channel receives the transmission time-domain transmission sequence and simultaneously receives an interference time-domain transmission sequence sent by the first terminal. When the interference time-domain transmission sequence and the transmission time-domain transmission sequence are transmitted in the full-duplex transmission channel, the interference time-domain transmission sequence and the transmission time-domain transmission sequence will interfere with and alias each other. The transmission time-domain transmission sequence is interfered with and aliased by the interference time-domain transmission sequence to obtain the reception time-domain transmission sequence.

[0226] For ease of description, the above devices are described in terms of function, divided into various modules. Of course, in implementing this application, the functions of each module can be implemented in one or more software and / or hardware.

[0227] The apparatus of the above embodiments is used to implement the corresponding source transmission method based on full-duplex communication in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0228] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the source transmission method based on full-duplex communication described in any of the above embodiments.

[0229] Figure 4 This illustration shows a more specific hardware structure diagram of an electronic device provided in this embodiment. The device may include: a processor 401, a memory 402, an input / output interface 403, a communication interface 404, and a bus 405. The processor 401, memory 402, input / output interface 403, and communication interface 404 are interconnected internally via the bus 405.

[0230] The processor 401 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0231] The memory 402 can be implemented in the form of ROM (Read-Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 402 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 402 and is called and executed by the processor 401.

[0232] Input / output interface 403 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components in the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touch screens, microphones, various sensors, etc., and output devices may include displays, speakers, vibrators, indicator lights, etc.

[0233] Communication interface 404 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0234] Bus 405 includes a pathway for transmitting information between various components of the device (e.g., processor 401, memory 402, input / output interface 403, and communication interface 404).

[0235] It should be noted that although the above-described device only shows the processor 401, memory 402, input / output interface 403, communication interface 404, and bus 405, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.

[0236] The electronic devices described above are used to implement the corresponding source transmission method based on full-duplex communication in any of the foregoing embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0237] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides a non-transitory computer-readable storage medium that stores computer instructions for causing the computer to execute the source transmission method based on full-duplex communication as described in any of the above embodiments.

[0238] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0239] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the source transmission method based on full-duplex communication as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0240] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this application (including the claims) is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this application as described above, which are not provided in the details for the sake of brevity.

[0241] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this application, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this application, and this also takes into account the fact that the details of the implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this application will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuits) have been set forth to describe exemplary embodiments of this application, it will be apparent to those skilled in the art that the embodiments of this application can be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0242] Although this application has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0243] The embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of this application.

Claims

1. A source transmission method based on full-duplex communication, characterized in that, Applied to a first terminal, the method includes: Obtain information from the first source; The first source information is encoded and modulated to obtain an interference information sequence, and an interference time-domain transmission sequence is obtained based on the interference information sequence through an inverse Fourier transform function. The interference time-domain transmission sequence is sent to a full-duplex transmission channel. The full-duplex transmission channel receives the interference time-domain transmission sequence and simultaneously receives a transmission time-domain transmission sequence sent by a second terminal. When the interference time-domain transmission sequence and the transmission time-domain transmission sequence are transmitted in the full-duplex transmission channel, the interference time-domain transmission sequence and the transmission time-domain transmission sequence will interfere and overlap with each other. The transmission time-domain transmission sequence is interfered with and overlapped by the interference time-domain transmission sequence to obtain the received time-domain transmission sequence. Receive the received time-domain transmission sequence transmitted through the full-duplex transmission channel; Based on the received time-domain transmission sequence and the interference time-domain transmission sequence, source reconstruction is performed to obtain source reconstruction information; The interference information sequence includes an interference side information sequence and / or an interference channel input symbol dimension sequence and / or an interference semantic information sequence; The step of encoding and modulating the first source information to obtain an interference information sequence includes: Semantic features are extracted from the first source information to obtain an interference semantic feature sequence; Side information is extracted from the interference semantic feature sequence to obtain the interference side information sequence; Based on the interference edge information sequence, perform super-prior entropy estimation analysis to determine the mean and standard deviation information corresponding to each interference semantic feature in the interference semantic feature sequence, and obtain the interference probability information sequence through the conditional Gaussian probability calculation function; Based on the interference probability information sequence, the interference channel input symbol dimension sequence is obtained through a quantization function; The interference semantic feature sequence is reduced in dimensionality according to the interference channel input symbol dimension sequence to obtain the interference semantic information sequence. The source reconstruction based on the received time-domain transmission sequence and the interference time-domain transmission sequence, to obtain source reconstruction information, includes: Obtain the input symbol dimension sequence and interference probability information sequence of the interference channel; Interference cancellation is performed on the received time-domain transmission sequence and the interference time-domain transmission sequence to obtain a received semantic information sequence and / or a received channel input dimension sequence and / or a received side information sequence. Based on the input symbol dimension sequence of the interference channel and the interference probability information sequence as semantic interference cancellation conditions, semantic interference cancellation is performed on the received semantic information sequence to obtain the received semantic information sequence after interference cancellation. Based on the received side information sequence and the received channel input dimension sequence as joint source channel decoding conditions, the received semantic information sequence after interference cancellation is jointly decoded by the source channel to obtain received semantic feature information. The received semantic feature information is reconstructed from the source to obtain the reconstructed source information; The step of using the input symbol dimension sequence of the interference channel and the interference probability information sequence as semantic interference cancellation conditions to perform semantic interference cancellation on the received semantic information sequence, resulting in an interference-cancelled received semantic information sequence, includes: Based on the interference channel input symbol dimension sequence and the interference probability information sequence as joint source channel decoding conditions, the received semantic information sequence is subjected to joint source channel decoding to obtain the interference detection semantic feature sequence. Source reconstruction is performed on the interference detection semantic feature sequence to obtain interference detection source information; The interference detection source information is reduced in dimensionality according to the interference channel input symbol dimension sequence to obtain the interference reconstruction semantic information sequence. The interference-reconstructed semantic information sequence and the received semantic information sequence are subjected to interference cancellation processing to obtain the interference-cancelled received semantic information sequence; or, Based on the interference channel input symbol dimension sequence and the interference probability information sequence as joint source channel decoding conditions, the received semantic information sequence is subjected to joint source channel decoding to obtain the interference detection semantic feature sequence. The interference detection semantic feature sequence is reduced in dimensionality according to the interference channel input symbol dimension sequence to obtain the interference reconstruction semantic information sequence. The interference-reconstructed semantic information sequence and the received semantic information sequence are subjected to interference cancellation processing to obtain the interference-cancelled received semantic information sequence; or, Obtain the sequence of interfering semantic features; Using the input symbol dimension sequence of the interference channel as a condition, the interference semantic feature sequence is reconstructed using the interference reconstruction joint source channel coding module to obtain the interference reconstruction semantic information sequence; Based on the reconstructed semantic information sequence of the interference and the received semantic information sequence, the received semantic information sequence after interference cancellation is obtained through a nonlinear activation function; The source reconstruction based on the received time-domain transmission sequence and the interference time-domain transmission sequence, to obtain source reconstruction information, includes: Obtain the interference semantic feature sequence, the interference channel input symbol dimension sequence, and the interference probability information sequence; Interference cancellation is performed on the received time-domain transmission sequence and the interfering time-domain transmission sequence to obtain the received digital information sequence; Based on the interference semantic feature sequence, the interference channel input symbol dimension sequence, and the interference probability information sequence as semantic interference cancellation conditions, semantic interference cancellation is performed on the received digital information sequence to obtain the interference-cancelled received digital information sequence. The received digital information sequence after interference cancellation is demodulated and channel decoded to obtain the received source bit sequence to be decoded; Source reconstruction is performed on the received source bit sequence to be decoded to obtain the source reconstruction information.

2. The method according to claim 1, characterized in that, The interference information sequence includes an interference digital transmission sequence; The step of encoding and modulating the first source information to obtain an interference information sequence includes: The first source information is compressed to obtain a compressed bit sequence of the interference source; The compressed bit sequence of the interference source is channel-coded and modulated to obtain the interference digital transmission sequence.

3. A source transmission method based on full-duplex communication, characterized in that, Applied to a second terminal, the method includes: Obtain information from a second source; The second source information is encoded and modulated to obtain a transmission sequence, and a transmission time-domain transmission sequence is obtained based on the transmission sequence through an inverse Fourier transform function. The transmitted time-domain transmission sequence is sent to a full-duplex transmission channel, which then transmits the transmitted time-domain transmission sequence to a first terminal. The first terminal executes the source transmission method based on full-duplex communication as described in any one of claims 1 to 2. Simultaneously, the first terminal receives an interfering time-domain transmission sequence. When the interfering time-domain transmission sequence and the transmitted time-domain transmission sequence are transmitted within the full-duplex transmission channel, they interfere with and overlap with each other. The transmitted time-domain transmission sequence is affected by the interference and overlap of the interfering time-domain transmission sequence to obtain the received time-domain transmission sequence.

4. The method according to claim 3, characterized in that, The transmission sequence includes a digital transmission sequence; The step of encoding and modulating the second source information to obtain a transmission sequence includes: The second source information is compressed to obtain a compressed bit sequence of the source; The source compressed bit sequence is channel-coded and modulated to obtain a digital transmission sequence.

5. The method according to claim 3, characterized in that, The transmission sequence includes a side information sequence and / or a channel input symbol dimension sequence and / or a semantic information sequence; The step of encoding and modulating the second source information to obtain a transmission sequence includes: The second source information is subjected to semantic feature extraction processing to obtain a semantic feature sequence; The semantic feature sequence is subjected to edge information extraction to obtain an edge information sequence; Based on the edge information sequence, perform super-prior entropy estimation analysis to determine the mean and standard deviation information corresponding to each semantic feature sequence in the semantic feature sequence, and obtain the probability information sequence through the conditional Gaussian probability calculation function; Based on the probability information sequence, the channel input symbol dimension sequence is obtained through a quantization function; The semantic feature sequence is reduced in dimensionality according to the channel input symbol dimension sequence to obtain the semantic information sequence.

6. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 5.

Images