An entropy model-based digital semantic perception OFDM subcarrier allocation method

Abstract

The application provides a digital semantic perception OFDM subcarrier allocation method based on an entropy model, and relates to the field of wireless communication. The method comprises the following steps: a semantic encoder in a joint coding and modulation module is used to perform semantic coding on an original image to obtain semantic features; the joint coding and modulation module is used to modulate the semantic features to obtain a modulation symbol sequence; a semantic importance evaluation module is used to process the semantic features to obtain a semantic importance vector; an orthogonal frequency division multiplexing transmitter is used to process the modulation symbol sequence into M to-be-transmitted orthogonal frequency division multiplexing data packets; the joint coding and modulation module is used to allocate the M to-be-transmitted orthogonal frequency division multiplexing data packets to M subcarriers according to the semantic importance vector and feedback channel state information; and the orthogonal frequency division multiplexing transmitter is used to send the M to-be-transmitted orthogonal frequency division multiplexing data packets to an orthogonal frequency division multiplexing receiver in a receiving end, so as to realize efficient and reliable transmission of digital semantic information in a wireless channel.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of wireless communication, in particular to a digital semantic-aware OFDM subcarrier allocation method based on an entropy model. BACKGROUND

[0002] In recent years, with the rapid development of deep learning technology, semantic communication research has made significant progress. Current research mainly uses neural network architecture to achieve semantic information extraction and encoding, and has been successfully applied to efficient transmission of various data types, including text, speech, images, video and multi-modal data.

[0003] In the digital implementation of semantic communication systems, the main challenge currently faced is the performance loss of analog-to-digital signal conversion. To improve this limitation, a joint coding-modulation (JCM) system is proposed, which uses a probability encoder to generate a constellation point transition probability matrix and randomly generates transmit symbols based on the probability distribution. Performance tests show that in the low signal-to-noise ratio region, the difference between the JCM model and the analog system is small; when the signal-to-noise ratio rises, although the gap between the JCM model and the analog system increases, it is still better than other digital schemes.

[0004] However, there are still difficulties in the compatibility of semantic encoders with current digital communication infrastructure, and the performance is also significantly reduced in wireless channel (such as frequency selective channel) environment. Therefore, how to realize the efficient and reliable transmission of digital semantic information in the wireless channel is a technical problem to be solved by the present application. SUMMARY

[0005] Based on the above technical problems, the present application provides a digital semantic-aware OFDM subcarrier allocation method based on an entropy model, which aims to overcome the above problems or at least partially solve the above problems.

[0006] The first aspect of the present application provides a digital semantic-aware OFDM subcarrier allocation method based on an entropy model, which comprises:

[0007] The sender encodes the original image through the semantic encoder in the joint coding-modulation module to obtain semantic features;

[0008] The sender modulates the semantic features through the joint coding-modulation module to obtain a modulation symbol sequence, and the sender processes the semantic features through a semantic importance evaluation module to obtain a semantic importance vector, the mth element of the semantic importance vector corresponding to the semantic importance of the orthogonal frequency division multiplexing data packet carried by the mth subcarrier, wherein the value of m is 1 to M, and M is the number of subcarriers;

[0009] The sending end processes the modulation symbol sequence into M orthogonal frequency division multiplexing data packets to be transmitted through the joint coding and modulation module.

[0010] The sending end allocates the M orthogonal frequency division multiplexing data packets to be transmitted to M subcarriers according to the semantic importance vector and the channel state information fed back by the receiving end of the orthogonal frequency division multiplexing receiver for the last image transmission, and the principle that the orthogonal frequency division multiplexing data packets with high semantic importance are preferentially allocated to subcarriers with good channel state, and transmits the M orthogonal frequency division multiplexing data packets to be transmitted to the orthogonal frequency division multiplexing receiver of the receiving end through the orthogonal frequency division multiplexing transmitter in the sending end using the M subcarriers.

[0011] In the digital semantic perception OFDM subcarrier allocation method based on an entropy model provided in the application, the sending end not only modulates the semantic features of the original image through the joint coding and modulation module to obtain a modulation symbol sequence, but also processes the semantic features of the original image through the semantic importance evaluation module to obtain a semantic importance vector; after processing the modulation symbol sequence into M orthogonal frequency division multiplexing data packets to be transmitted through the joint coding and modulation module, the M orthogonal frequency division multiplexing data packets to be transmitted are allocated to M subcarriers according to the semantic importance vector and the channel state information fed back by the receiving end of the orthogonal frequency division multiplexing receiver for the last image transmission, and the principle that the orthogonal frequency division multiplexing data packets with high semantic importance are preferentially allocated to subcarriers with good channel state, so that the M orthogonal frequency division multiplexing data packets to be transmitted are transmitted to the orthogonal frequency division multiplexing receiver of the receiving end through the orthogonal frequency division multiplexing transmitter using the M subcarriers, and the transmission of digital semantic information in the wireless channel is completed. In this way, the application realizes the conversion between semantic information and orthogonal frequency division multiplexing signals by deeply integrating deep learning digital semantic coding with orthogonal frequency division multiplexing physical layer transmission, and realizes the intelligent matching of high-quality subcarriers and important semantic features between the orthogonal frequency division multiplexing transmitter and the orthogonal frequency division multiplexing receiver by comprehensively considering the semantic importance vector corresponding to the digital semantic information and the channel state information of the transmission channel, thereby significantly reducing the information transmission loss, improving the robustness of the digital semantic information in the complex channel environment and the overall information transmission efficiency, and realizing the efficient and reliable transmission of digital semantic information in the wireless channel. BRIEF DESCRIPTION OF DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the description of the embodiments of the application. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0013] Figure 1 is a step flow chart of a digital semantic-aware OFDM subcarrier allocation method based on an entropy model according to an embodiment of the present application;

[0014] Figure 2 is a structural block diagram of a channel equalization module according to an embodiment of the present application;

[0015] Figure 3 is a processing schematic diagram of an OFDM transmitter and an OFDM receiver according to an embodiment of the present application;

[0016] Figure 4 is a structural schematic diagram of a semantic importance evaluation module according to an embodiment of the present application;

[0017] Figure 5 is a structural schematic diagram of a digital semantic-aware OFDM subcarrier allocation system based on an entropy model according to an embodiment of the present application;

[0018] Figure 6 is a structural block diagram of a digital semantic-aware OFDM subcarrier allocation system based on an entropy model according to an embodiment of the present application;

[0019] Figure 7 is a schematic diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0021] Please refer to Figure 1 , Figure 1 is a step flow chart of a digital semantic-aware OFDM subcarrier allocation method based on an entropy model according to an embodiment of the present application. As shown in Figure 1 , the digital semantic-aware OFDM subcarrier allocation method based on an entropy model provided by the embodiment includes at least the following steps:

[0022] Step S11: The sender performs semantic encoding on the original image through a semantic encoder in a joint coding modulation module to obtain semantic features.

[0023] In this embodiment, the sending end at least includes a joint encoding modulation module and an orthogonal frequency division multiplexing transmitter (OFDM transmitter). For an original image to be transmitted over a wireless channel, the original image is input to the joint encoding modulation module, and the sending end can first perform semantic encoding on the original image by a semantic encoder in the joint encoding modulation module to obtain semantic features of the original image.

[0024] Step S12: The sending end modulates the semantic features by the joint encoding modulation module to obtain a modulation symbol sequence, and processes the semantic features by a semantic importance evaluation module to obtain a semantic importance vector.

[0025] In this embodiment, after obtaining the semantic features of the original image, the sending end can modulate the semantic features by the joint encoding modulation module to obtain a modulation symbol sequence as input to the OFDM transmitter. The modulation symbol sequence is a modulation signal with semantic features, i.e., a modulated semantic feature symbol suitable for wireless channel transmission.

[0026] In addition, after obtaining the semantic features of the original image, the sending end can also process the semantic features by the semantic importance evaluation module to obtain a semantic importance vector. The semantic importance evaluation module in this embodiment acts on the semantic features to evaluate and quantify the importance of data units carried by different sub-channels. Specifically, the semantic features are processed by the semantic importance evaluation module to obtain a semantic importance vector The mth element of the semantic importance vector corresponds to the semantic importance of an OFDM data packet carried by the mth sub-carrier; that is, the mth element of the semantic importance vector I corresponds to the relative importance degree of a data packet carried by the mth sub-carrier used in a single OFDM symbol transmission period in the OFDM transmitter and the OFDM receiver; where m is an integer from 1 to M, M is the number of sub-carriers, i.e., M is the number of sub-carriers used in a single OFDM symbol transmission in OFDM, and M is an integer greater than 1.

[0027] Step S13: The sending end processes the modulation symbol sequence into M OFDM data packets to be transmitted by the OFDM transmitter.

[0028] In this embodiment, the sending end inputs the modulation symbol sequence output by the joint encoding modulation module to the OFDM transmitter, and the OFDM transmitter is responsible for mapping the modulated semantic features (i.e., the modulation symbol sequence) into a standard OFDM symbol sequence: the OFDM transmitter processes the modulation symbol sequence into M OFDM data packets to be transmitted.

[0029] ​Step S14: The sender allocates the M to-be-transmitted OFDM data packets to M subcarriers according to the principle that OFDM data packets with high semantic importance are preferentially allocated to subcarriers with good channel states, and sends the M to-be-transmitted OFDM data packets to the OFDM receiver of the receiver through the OFDM transmitter using the M subcarriers.

[0030] In this embodiment, the OFDM receiver of the receiver feeds back channel state information in the image transmission process to the joint coding and modulation module for each image transmission. For the current image transmission, the joint coding and modulation module can allocate M to-be-transmitted OFDM data packets to M subcarriers according to the principle that OFDM data packets with high semantic importance are preferentially allocated to subcarriers with good channel states, according to the semantic importance vector and the channel state information fed back by the OFDM receiver for the last image transmission. For example, the OFDM data packet with the highest semantic importance can be allocated to the subcarrier with the best channel state, the OFDM data packet with the second highest semantic importance can be allocated to the subcarrier with the second best channel state, and so on, until the M to-be-transmitted OFDM data packets are allocated to the M subcarriers respectively.

[0031] It should be noted that this embodiment does not limit the number of transmitted images for each image transmission, i.e., one or more original images can be transmitted each time.

[0032] After the M to-be-transmitted OFDM data packets are allocated to the M subcarriers, the sender can send the allocated M to-be-transmitted OFDM data packets to the OFDM receiver of the receiver through the OFDM transmitter using the M subcarriers through the wireless channel.

[0033] In an optional embodiment, the wireless channel can be a wireless channel in a complex environment, including but not limited to a frequency selective channel, a multipath fading channel, etc.

[0034] In this embodiment, by deeply integrating deep learning digital semantic coding with OFDM physical layer transmission, the conversion between semantic information and OFDM signals is realized, and by comprehensively considering the semantic importance vector corresponding to the digital semantic information and the channel state information of the transmission channel, the intelligent matching between the high-quality subcarriers and the important semantic features between the OFDM transmitter and the OFDM receiver is realized, thereby significantly improving the robustness of the digital semantic information in the complex channel environment and the overall information transmission efficiency, and realizing the efficient and reliable transmission of the digital semantic information in the wireless channel.

[0035] In combination with the above embodiments, in an alternative embodiment, the joint coding modulation module can include a probability converter and a constellation symbol generator in addition to the semantic encoder. After obtaining the semantic features through the semantic encoder, the semantic features can be input into the probability converter, the semantic information contained in the semantic features is mapped into the transition probability from the latent feature space to the discrete modulation symbol space through the probability converter, and then the transition probability is input into the constellation symbol generator. The constellation symbol generator performs a random sampling operation according to the transition probability distribution, thereby randomly generating a specific modulation symbol sequence as the input of the OFDM transmitter. The modulation symbol sequence of the present embodiment is a constellation symbol modulated from the semantic features, which is a discrete variable with a classification distribution, and is obtained by sampling the transition probability generated by the probability encoder.

[0036] In combination with the above embodiments, in an embodiment, the present application further provides an entropy model-based digital semantic perception OFDM subcarrier allocation method. In the method, in addition to the above steps, steps S21 to S25 can be further included.

[0037] Step S21: The receiving end processes the received OFDM data packets through the OFDM receiver to obtain intermediate signals and provides the intermediate signals to the channel estimation module and the channel equalization module.

[0038] In the present embodiment, after the OFDM transmitter of the sending end transmits the allocated M OFDM data packets to be transmitted through the wireless channel to the OFDM receiver of the receiving end using M subcarriers, the receiving end can receive these OFDM data packets through the OFDM receiver, process the received OFDM data packets through the OFDM receiver to obtain intermediate signals, and then provide the intermediate signals to the channel estimation module and the channel equalization module in the OFDM receiver, respectively.

[0039] In the present embodiment, the receiving end at least includes an OFDM receiver and two parallel probability decoders (parallel semantic decoders), wherein the OFDM receiver demodulates the received signals and recovers the transmitted semantic features, and the dual probability decoder architecture (i.e., the two parallel probability decoders) recovers the source data and the original semantic information from the received signals.

[0040] Step S22: The channel estimation module obtains the channel state information of the present image transmission according to the intermediate signals, and sends the channel state information of the present image transmission to the channel equalization module in the OFDM receiver.

[0041] In this embodiment, the channel estimation module processes the intermediate signal according to the intermediate signal to obtain the channel state information of the current image transmission output by the channel estimation module. The channel estimation module can send the channel state information of the current image transmission to the channel equalization module in the OFDM receiver.

[0042] Step S23: The channel equalization module obtains the channel equalization result according to the channel state information of the current image transmission and the intermediate signal.

[0043] In this embodiment, the channel equalization module can process the channel state information of the current image transmission and the intermediate signal according to the received channel state information of the current image transmission and the intermediate signal to obtain the channel equalization result.

[0044] Step S24: The receiving end performs semantic reconstruction according to the channel equalization result through the semantic information reconstruction module to obtain the reconstructed semantic features.

[0045] In this embodiment, the receiving end includes two parallel semantic decoders: the semantic information reconstruction module and the source information reconstruction module, which are respectively used for semantic recovery and image recovery tasks, and both take the channel equalization result as input. Among them, the receiving end can decode and restore the original semantic information from the channel equalization result through the semantic information reconstruction module, perform semantic reconstruction, reconstruct the semantic content corresponding to the original image, and obtain the reconstructed semantic features.

[0046] Step S25: The receiving end performs image reconstruction according to the channel equalization result through the source information reconstruction module to obtain the reconstructed image.

[0047] In this embodiment, the receiving end can perform image reconstruction according to the channel equalization result through the source information reconstruction module to reconstruct the visual content of the original image from the channel equalization result and obtain the reconstructed image.

[0048] In combination with any of the above embodiments, the application also provides an OFDM subcarrier allocation method based on an entropy model. In addition to the above steps, the method can further include the following steps S31 to S32:

[0049] Step S31: The channel estimation module feeds back the channel state information of the current image transmission to the joint coding and modulation module.

[0050] In this embodiment, after obtaining the channel state information of the current image transmission, the channel estimation module will also feed back the channel state information of the current image transmission to the joint coding and modulation module in the sending end.

[0051] Step S32: the joint coding modulation module performs subcarrier allocation for the next image transmission based on the channel state information of the current image transmission.

[0052] In the embodiment, after receiving the channel state information of the current image transmission, the joint coding adjustment module can perform subcarrier allocation for the next image transmission based on the channel state information of the current image transmission: in the next image transmission, the received channel state information of the current image transmission is taken as the channel state information fed back by the last image transmission, so that the M OFDM data packets to be transmitted in the next image transmission are allocated to M subcarriers according to the principle that the OFDM data packets with high semantic importance are preferentially allocated to subcarriers with good channel state, so as to complete subcarrier allocation.

[0053] In combination with any of the above embodiments, the application further provides an OFDM subcarrier allocation method based on an entropy model, in which, in addition to the above steps, the method can further include the following step S101:

[0054] Step S101: the semantic encoder performs semantic encoding on the original image of the next image transmission based on the channel state information of the current image transmission.

[0055] In the embodiment, after obtaining the feedback channel state information of the current image transmission, the joint coding modulation module can perform semantic encoding on the original image of the next image transmission based on the received channel state information of the current image transmission by the semantic encoder in the joint coding modulation module, so as to obtain the semantic features in the next image transmission and realize channel adaptive semantic encoding.

[0056] In an optional embodiment, in order to improve transmission performance, the sending end of the embodiment needs to have the ability to preferentially allocate more important semantic information to subchannels with better channel conditions, so as to reduce information loss and improve transmission efficiency. Therefore, the selection of evaluation criteria directly affects the allocation effect, and how to effectively evaluate the channel conditions of subchannels and the importance of semantic information data on each subcarrier becomes the key of the subcarrier allocation algorithm. In the embodiment, the subchannel gain G is used as a quantitative index of channel quality, and G can be represented as: ; wherein, H1 to H M is the channel gain of the mth subcarrier, which can be obtained by the channel estimation module of the receiving end. That is, the channel state information in the embodiment can include: channel gain, and the channel gain is used as a quantitative index of channel quality.

[0057] In combination with the above embodiments, in an optional embodiment, as shown in Figure 2 , Figure 2 is a structural block diagram of a channel equalization module according to an embodiment of the present application. In Figure 2 , the channel equalization module can use a convolutional neural network as a subnetwork to learn the residual error in traditional channel equalization (such as MMSE), and compensate for the result of traditional channel equalization. The subnetwork is composed of two convolutional layers (such as 2 Conv2d) and two normalization layers (such as 2 Norm2d), the input of the channel equalization module is channel state information and intermediate signal r, wherein the channel state information is represented by the channel gain H of the subcarrier; the output of the channel equalization module is the channel equalization result .

[0058] In combination with the above embodiments, in an optional embodiment, as shown in Figure 3 , Figure 3 is a processing schematic diagram of an OFDM transmitter and an OFDM receiver according to an embodiment of the present application. In Figure 3 , the OFDM transmitter can first perform inverse fast Fourier transform (IFFT) on the modulated symbol sequence to obtain a time-domain representation, then append a cyclic prefix and insert a pilot, and then the obtained OFDM symbol is subjected to peak clipping by a nonlinear peak clipping function to obtain the input t of the wireless channel, that is, the OFDM data packet to be transmitted, the purpose is to reduce the peak-to-average ratio.

[0059] In Figure 3 , in the OFDM receiver, for the received OFDM data packet, the cyclic prefix is removed and fast Fourier transform (FFT) is performed, and then the intermediate signal is obtained, and then the intermediate signal is input into the channel estimation module and the channel equalization module. The channel estimation module obtains the channel state information based on the intermediate signal, inputs the channel state information into the channel equalization module, and feeds back the channel state information to the joint coding and modulation module at the receiving end. The channel equalization module processes the channel equalization result based on the intermediate signal and the channel state information.

[0060] In combination with any of the above embodiments, in an implementation, the present application also provides an OFDM subcarrier allocation method based on digital semantic perception of an entropy model. In the method, the step S12 of "processing the semantic features by the semantic importance evaluation module to obtain a semantic importance vector" can specifically include the following steps S41 to S44:

[0061] Step S41: the sender inputs the semantic feature into the hyper-encoder in the semantic importance evaluation module to obtain a hyper-prior feature, and inputs the semantic feature into the first quantizer in the semantic importance evaluation module to obtain a semantic feature quantized representation, and inputs the semantic feature quantized representation into the context analysis model in the semantic importance evaluation module.

[0062] In this embodiment, the semantic importance evaluation module at least includes a hyper-encoder, a first quantizer, a context analysis model, a second quantizer, a hyper-decoder, a Gaussian parameter estimation module, and an entropy calculation module. Wherein, the sender can input the semantic feature (which can be understood as the potential representation of the original image with the most information quantity extracted and compressed from the high-dimensional pixel space) into the hyper-encoder in the semantic importance evaluation module to obtain the hyper-prior feature output by the hyper-encoder. The hyper-prior feature is a deeper and more global statistical feature in the semantic feature.

[0063] In addition, the sender can input the semantic feature into the first quantizer in the semantic importance evaluation module, quantize the semantic feature by the first quantizer for transmission, obtain a semantic feature quantized representation, and input the semantic feature quantized representation into the context analysis model to obtain the output of the context analysis model. The context model is an autoregressive model based on potential representation, and its essence is to use the spatial dependence relationship inside the quantized potential feature (i.e. the semantic feature quantized representation) for local prediction.

[0064] Step S42: input the hyper-prior feature into the second quantizer in the semantic importance evaluation module to obtain a hyper-prior feature quantized representation, and input the hyper-prior feature quantized representation into the hyper-decoder in the semantic importance evaluation module.

[0065] In this embodiment, after obtaining the hyper-prior feature, the hyper-prior feature can be input into the second quantizer to be quantized by the second quantizer to obtain a hyper-prior feature quantized representation, and the hyper-prior feature quantized representation is input into the hyper-decoder to obtain the output of the hyper-decoder. In this embodiment, the hyper-encoder and the hyper-decoder are used to further extract deeper and more global statistical features in the semantic feature, and learn how to represent information that helps to correct the context prediction.

[0066] Step S43: processing the output of the context analysis model and the output of the hyper-decoder by the Gaussian parameter estimation module in the semantic importance evaluation module to obtain a Gaussian parameter estimation value.

[0067] In this embodiment, the output of the context analysis model and the output of the super decoder are input to the Gaussian parameter estimation module. Based on the output of the context analysis model and the output of the super decoder, the Gaussian parameter estimation module obtains global statistical information and local prediction information of the decoded neighborhood, and generates Gaussian parameter estimates. These Gaussian parameter estimates include the mean and scale parameters required by the conditional Gaussian entropy model.

[0068] Step S44: The Gaussian parameter estimate is processed by the entropy calculation module in the semantic importance assessment module to obtain the entropy value of the semantic feature, which is used as the semantic importance vector.

[0069] In this embodiment, after obtaining the Gaussian parameter estimate, the Gaussian parameter estimate is input into the entropy calculation module. The entropy calculation module processes the Gaussian parameter estimate to obtain the entropy value of the semantic feature, and uses the entropy value of the semantic feature as the semantic importance vector.

[0070] In this embodiment, the probability distribution of each element in the semantic feature is obtained by using the Gaussian parameters of each element of the semantic feature obtained by the semantic importance assessment module, and then the entropy value of each element in the semantic feature is obtained. This entropy value is used as an indicator to assess the semantic importance, so as to effectively assess the importance of data packets on each subchannel (subcarrier).

[0071] In conjunction with the above embodiments, in an optional embodiment, such as Figure 4 As shown, Figure 4 This is a schematic diagram illustrating the structure of a semantic importance evaluation module according to an embodiment of the present invention. Figure 4 The semantic importance evaluation module includes at least the following components: a super encoder, a first quantizer, a context analysis model, a second quantizer, a super decoder, a Gaussian parameter estimation module, and an entropy calculation module. Semantic feature Y is input into the semantic importance evaluation module. Within this module, semantic feature Y is fed into the super encoder and the first quantizer, yielding the super-prior feature Z and the quantized representation of the semantic feature, respectively. The prior feature Z is quantized using a second quantizer to obtain the quantized representation of the prior feature. Then, the prior features are quantized and represented. Input the super decoder and obtain its output; quantize the semantic features. Input the context analysis model to obtain its output; then input the output of the context analysis model and the output of the super decoder into the Gaussian parameter estimation module to obtain the Gaussian parameter estimate; finally input the Gaussian parameter estimate into the entropy calculation module to obtain the entropy value of the semantic feature, which is used as the semantic importance vector I.

[0072] In combination with any of the above embodiments, in an implementation, the application further provides a digital semantic-aware OFDM subcarrier allocation method based on an entropy model. In the method, the digital semantic-aware OFDM subcarrier allocation method based on the entropy model is implemented through a subcarrier allocation model based on the entropy model, and a training process of the subcarrier allocation model based on the entropy model at least includes a first stage. The training target of the first stage is to minimize the length of the bit stream to be transmitted and minimize the distortion of the sample reconstructed image relative to the sample original image. Through the first stage, the semantic encoder to be trained, the semantic importance evaluation module to be trained, and the source information reconstruction module to be trained are trained to obtain the trained semantic importance evaluation module, the semantic encoder trained through the first stage, and the source information reconstruction module trained through the first stage.

[0073] In the embodiment, the bit stream to be transmitted at least includes the sample semantic feature quantized representation corresponding to the sample original image and the sample hyper-prior feature quantized representation corresponding to the sample original image. The length of the bit stream to be transmitted at least includes the length of the bit stream of the sample semantic feature quantized representation corresponding to the sample original image and the length of the bit stream of the sample hyper-prior feature quantized representation corresponding to the sample original image. The sample original image is an original image used in the training process of the subcarrier allocation model based on the entropy model. The sample semantic feature quantized representation is a semantic feature quantized representation obtained by the semantic importance evaluation module to be trained in the training process of the subcarrier allocation model based on the entropy model. The sample hyper-prior feature quantized representation is a hyper-prior feature quantized representation obtained by the semantic importance evaluation module to be trained in the training process of the subcarrier allocation model based on the entropy model.

[0074] The subcarrier allocation model based on the entropy model in the embodiment is trained in an end-to-end manner. In the training process of the subcarrier allocation model based on the entropy model, the sample original image is input into the subcarrier allocation model based on the entropy model to be trained to obtain the output sample reconstructed image and the sample reconstructed semantic feature. The subcarrier allocation model based on the entropy model to be trained at least includes a joint coding and modulation module to be trained (at least including a semantic encoder to be trained, a semantic importance evaluation module to be trained), a source information reconstruction module to be trained, a semantic information reconstruction module to be trained, an orthogonal frequency division multiplexing receiver, and an orthogonal frequency division multiplexing transmitter. In the training process, the processing process in the subcarrier allocation model based on the entropy model to be trained is similar to the digital semantic-aware OFDM subcarrier allocation method based on the entropy model in the above embodiments, and the specific process can be referred to any of the above embodiments.

[0075] In combination with any of the above embodiments, in an embodiment, the present application further provides an entropy model based digital semantic perception OFDM subcarrier allocation method. In the method, the loss function value of the first stage is determined according to the difference between the sample original image and the sample reconstructed image, and the bit stream length of the sample semantic feature quantization representation corresponding to the sample original image and the bit stream length of the sample hyper-prior feature quantization representation corresponding to the sample original image.

[0076] In the embodiment, the first stage focuses on jointly training the semantic encoder, the semantic importance evaluation module and the source information reconstruction module, aiming to improve the ability of the semantic importance evaluation module to evaluate the semantic importance. Therefore, in the first stage, the network parameters of the semantic encoder to be trained, the semantic importance evaluation module to be trained and the source information reconstruction module to be trained can be updated by the loss function of the first stage until the loss function of the first stage converges, the network parameters are fixed, and the trained semantic importance evaluation module, the semantic encoder trained in the first stage and the source information reconstruction module trained in the first stage are obtained.

[0077] In the embodiment, the first stage focuses on jointly training the semantic encoder, the semantic importance evaluation module and the source information reconstruction module, aiming to improve the ability of the semantic importance evaluation module to evaluate the semantic importance. Therefore, in the first stage, the network parameters of the semantic encoder to be trained, the semantic importance evaluation module to be trained and the source information reconstruction module to be trained can be updated by the loss function of the first stage until the loss function of the first stage converges, the network parameters are fixed, and the trained semantic importance evaluation module, the semantic encoder trained in the first stage and the source information reconstruction module trained in the first stage are obtained.

[0078] In the embodiment, in the training process of the first stage, the difference between the sample original image and the sample reconstructed image can be determined based on the sample original image and the sample reconstructed image, wherein the sample reconstructed image is a reconstructed image obtained in the training process of the entropy model based subcarrier allocation model. And the bit stream length of the sample semantic feature quantization representation corresponding to the sample original image can be obtained according to the sample semantic feature quantization representation corresponding to the sample original image, and the bit stream length of the sample hyper-prior feature quantization representation corresponding to the sample original image can be obtained according to the sample hyper-prior feature quantization representation corresponding to the sample original image.

[0079] In the embodiment, in combination with the training target of the first stage, the cost of transmitting hyper-potential features and the reconstruction error distortion metric need to be considered, therefore, the loss function value of the first stage can be obtained according to the difference between the sample original image and the sample reconstructed image, the bit stream length of the sample semantic feature quantization representation corresponding to the sample original image, and the bit stream length of the sample hyper-prior feature quantization representation corresponding to the sample original image.

[0080] In a specific optional embodiment, the loss function of the first stage is :

[0081] + ;

[0082] wherein, is a sample semantic feature quantized representation, is a sample hyper-prior feature quantized representation, is a bitstream length of the sample semantic feature quantized representation corresponding to the sample original image, is a bitstream length of the sample hyper-prior feature quantized representation corresponding to the sample original image, X is the sample original image, is a sample reconstructed image, corresponds to a distortion of the sample reconstructed image relative to the sample original image; is a trade-off hyper-parameter used to balance the image classification task and the image quality task.

[0083] In combination with any of the above embodiments, in an implementation, the present application further provides a digital semantic-aware OFDM subcarrier allocation method based on an entropy model. In the method, the training process of the subcarrier allocation model based on the entropy model further includes a second stage; through the second stage, the semantic encoder trained in the first stage, the source information reconstruction module trained in the first stage, and the semantic information reconstruction module to be trained are trained under the condition that the parameters of the trained semantic importance evaluation module are frozen, and the subcarrier allocation model based on the entropy model is obtained.

[0084] In the embodiment, since the functions of the semantic importance evaluation module and the semantic encoder and decoder are relatively independent, the subcarrier allocation model based on the entropy model adopts a two-stage training mode. The second stage is to further train the semantic encoder and decoder to strengthen the ability of extracting semantic information and subcarrier allocation of the to-be-sent information.

[0085] Specifically, in the second stage of the training process of the subcarrier allocation model based on the entropy model, the semantic encoder trained in the first stage, the source information reconstruction module trained in the first stage, and the semantic information reconstruction module to be trained are trained under the condition that the network parameters of the trained semantic importance evaluation module are frozen, and the trained subcarrier allocation model based on the entropy model is obtained, thereby realizing the digital semantic-aware OFDM subcarrier allocation method based on the entropy model of the foregoing embodiments.

[0086] In combination with any of the above embodiments, in an implementation, the present application further provides a digital semantic-aware OFDM subcarrier allocation method based on an entropy model. In the method, the loss function value of the second stage is obtained at least according to the following information:

[0087] a difference between a class label of the sample semantic feature corresponding to the sample original image and a class prediction result corresponding to the sample reconstructed semantic feature;

[0088] a difference between the sample original image and the sample reconstructed image.

[0089] In this embodiment, the image classification task is used for training, and the semantic information is limited to the class label. After obtaining the sample reconstructed semantic feature, the class prediction result corresponding to the sample reconstructed semantic feature can be determined. The class prediction result corresponding to the sample reconstructed semantic feature is equivalent to a reconstructed class label, so that the difference between the class label of the sample semantic feature corresponding to the sample original image and the class prediction result corresponding to the sample reconstructed semantic feature can be determined. Then, based on the difference between the class label of the sample semantic feature corresponding to the sample original image and the class prediction result corresponding to the sample reconstructed semantic feature, and the difference between the sample original image and the sample reconstructed image, the loss function value of the second stage is determined.

[0090] In combination with any of the above embodiments, in an implementation, the present application further provides an OFDM subcarrier allocation method based on an entropy model. In this method, the loss function value of the second stage is further obtained according to the following information:

[0091] a rank correlation coefficient representing a similarity between the sample semantic importance vector of the sample original image and the sample channel state information fed back in the image transmission process of the sample original image. In order to train the allocation performance of the subchannel, the matching degree of the channel state information and the importance of the semantic data packet (i.e., the orthogonal frequency division multiplexing data packet) is reflected in the loss function of the second stage in this embodiment. For example, the rank correlation coefficient of the sample semantic importance vector of the sample original image and the sample channel state information fed back in the image transmission process of the sample original image can be selected as a measurement index. The rank correlation coefficient represents a similarity between the sample semantic importance vector of the sample original image and the sample channel state information fed back in the image transmission process of the sample original image.

[0092] In an optional example, the rank correlation coefficient can be a Kendall correlation coefficient. The Kendall correlation coefficient is based on the rank of the data object to evaluate the correlation between two random variables, and is mainly used to evaluate the degree of cooperation of the ordering of two variables.

[0093] In this embodiment, the Kendall correlation coefficient is intended to reflect the similarity between the semantic importance sequence and the sub-channel gain sequence. However, it can be observed that the Kendall correlation coefficient is non-differentiable because it requires both in-order and out-of-order pairs to participate in the calculation, hindering its direct integration into the loss function for gradient optimization. Therefore, this embodiment improves the Kendall correlation coefficient by replacing the non-differentiable in-order and out-of-order pair counting with a differentiable sigmoid function. This retains some counting functionality while ensuring the differentiability of the loss function.

[0094] The improved Kendall correlation coefficient in this embodiment can be expressed as:

[0095] ;

[0096] in, The rank correlation coefficient, Let i be the Sigmoid function, i = 1, 2, ... M j=1,2,…, M , M The length of the semantic importance sequence. The i-th element of the semantic importance sequence Let j be the j-th element of the semantic importance sequence. Let i be the i-th element of the subchannel gain sequence. It is the j-th element of the subchannel gain sequence.

[0097] In a specific alternative embodiment, the loss function of the second stage for:

[0098] ;

[0099] Where N is the number of samples in each batch during training. The rank correlation coefficient, For the original image of the nth sample, Reconstruct the image for the nth sample. The difference between the reconstructed image of the corresponding sample and the original image of the sample. These are two trade-off hyperparameters used to balance image classification, image quality, and subcarrier matching tasks. And, For cross-entropy loss, For mean square error loss, The loss is related to semantic importance.

[0100] In conjunction with the above embodiments, in one embodiment, such as Figure 5 As shown, Figure 5 This is a schematic diagram illustrating the structure of a digital semantic-aware orthogonal frequency division multiplexing subcarrier allocation system based on an entropy model, according to an embodiment of the present invention.Figure 5 In the embodiment, the digital semantic perception orthogonal frequency division multiplexing subcarrier allocation system based on an entropy model comprises a sending end and a receiving end; the sending end at least comprises a joint coding modulation module, a dimension reorganization module and an orthogonal frequency division multiplexing transmitter, wherein the core task of the joint coding modulation module is to perform deep analysis and conversion on an input image, and finally output a modulated semantic feature symbol suitable for wireless channel transmission; the joint coding modulation module comprises a semantic encoder, a probability converter, a constellation symbol generator and a semantic importance evaluation module. The receiving end at least comprises an orthogonal frequency division multiplexing receiver, a dimension reorganization module, a semantic information reconstruction module and a source information reconstruction module.

[0101] The sending end inputs an original image into the joint coding modulation module, and the original image is denoted as X. Wherein c represents the color number of the original image, h represents the height of the original image, and w represents the width of the original image. The joint coding modulation module outputs a modulated symbol sequence C after processing the original image. Wherein n represents the number of channels used. Inside the joint coding modulation module, the original image X is first sent to the semantic encoder. The semantic encoder is essentially a deep neural network, which extracts semantic features Y from the original pixel data through multiple layers of nonlinear transformation. The semantic features Y are input into the probability converter to obtain transition probability P, and then the transition probability P is input into the constellation symbol generator to obtain the modulated symbol sequence C as the output of the joint coding modulation module; and the semantic features Y are input into the semantic importance evaluation module to obtain a semantic importance vector I, and then a semantic importance related loss can be obtained based on the semantic importance vector I.

[0102] Then, the dimension of the modulated symbol sequence C can be changed from to by the dimension reorganization module, wherein S represents the number of OFDM symbols in an orthogonal frequency division multiplexing data packet, and the number of subcarriers, and then the modulated symbol sequence after dimension reorganization is input into the orthogonal frequency division multiplexing transmitter to obtain a to-be-transmitted orthogonal frequency division multiplexing data packet t as the input of the channel.

[0103] The receiving end receives the orthogonal frequency division multiplexing data packet r through the orthogonal frequency division multiplexing receiver, and then processes the received orthogonal frequency division multiplexing data packet r to obtain a channel equalization result , and feeds back the obtained channel state information to the joint coding modulation module; then, the channel equalization result is input into the semantic information reconstruction module and the source information reconstruction module after dimension change by the dimension reorganization module. Finally, the semantic information reconstruction module reconstructs the original image according to the channel equalization result after dimension change semantic reconstruction is performed to obtain reconstructed semantic features , and further cross-entropy loss can be obtained based on the reconstructed semantic features ; the source information reconstruction module performs image reconstruction based on the channel equalization result after dimension transformation , and further mean square error loss can be obtained based on the reconstructed image .

[0104] It should be noted that the semantic importance evaluation module can belong to the joint coding modulation module, or can be an independent module in the sending end (independent of the joint coding modulation module), and the semantic importance evaluation module has the same effect in the two cases.

[0105] The present application proposes and constructs an innovative digital semantic communication system model based on OFDM technology (i.e., a subcarrier allocation model based on an entropy model). The core of the model is to deeply integrate the advanced deep learning digital semantic coding scheme with the mature OFDM physical layer transmission technology, and the embodiment designs the conversion mechanism between semantic information and OFDM signals and formulates the corresponding end-to-end model training strategy, the main goal of which is to significantly improve the robustness of the system in a complex channel environment and the overall information transmission efficiency. The model exhibits stable and excellent performance under a wide range of signal-to-noise ratios, and shows good adaptability and robustness to different modulation orders. In different modulation modes, the fidelity of semantic transmission can be effectively guaranteed. In a typical multipath fading channel model, the model designed by the present application successfully maintains stable communication performance, verifying the effectiveness of the neural network assisted equalization and semantic perception resource allocation strategy in overcoming frequency selective fading.

[0106] Specifically, in an embodiment, the simulation experiment is performed on a system equipped with an AMD EPYC 9754 128-core CPU and a 24 GB NVIDIA GeForce RTX 4090D graphics card, and another system equipped with an AMD Ryzen 5 5600H CPU and a 4 GB NVIDIA GeForce RTX 3050 Laptop GPU, using Pytorch and CUDA 12.6 driver. The experiments are all performed on the CIFAR10 dataset, in which the classification label is set as semantic information. The CIFAR10 dataset contains 60000 32x32 color images of 10 categories, of which 50000 are used for training and 10000 are used for testing. In the two stages of training, the training batch size of the model is 32 samples. The neural network parameters in the semantic encoder and decoder are trained using the Adam optimizer, and the learning rate is decayed using the cosine annealing strategy, with an initial learning rate of​ , to 0.1 in 300 training rounds .

[0107] The embodiment analyzes the influence law of key system parameters such as modulation order and channel signal-to-noise ratio on classification accuracy and PSNR through a control variable experiment, the experimental results show the robustness of the designed scheme under various signal-to-noise ratios and modulation schemes, and prove that the system can still maintain stable communication performance in a multipath channel environment. At the same time, using ablation experiments, the contribution of each module to the overall performance is verified, proving that the designed subcarrier adaptive allocation strategy can still maintain stable performance in a multipath fading channel and effectively improve the information transmission capacity of the communication system.

[0108] It should be noted that for the method embodiments, in order to simply describe, they are all described as a series of action combinations, but those skilled in the art should know that the embodiments of the present application are not limited to the order of the described actions, because according to the embodiments of the present application, certain steps can be performed in other order or at the same time. Secondly, those skilled in the art should know that the embodiments described in the specification all belong to preferred embodiments, and the actions involved are not necessarily the necessary of the embodiments of the present application.

[0109] Based on the same inventive concept, an embodiment of the present application provides a digital semantic perception OFDM subcarrier allocation system based on an entropy model. Referring to Figure 6 , Figure 6 is a structural block diagram of a digital semantic perception OFDM subcarrier allocation system based on an entropy model provided by an embodiment of the present application. As Figure 6 shown, the system includes a sending end and a receiving end; the sending end is used for:

[0110] The semantic encoder in the joint coding modulation module is used for performing semantic coding on the original image to obtain semantic features;

[0111] The joint coding modulation module is used for modulating the semantic features to obtain a modulation symbol sequence, and the semantic importance evaluation module is used for processing the semantic features to obtain a semantic importance vector, an mth element of the semantic importance vector corresponds to semantic importance of an orthogonal frequency division multiplexing data packet carried by an mth subcarrier, where m is 1 to M, and M is the number of subcarriers;

[0112] The joint coding modulation module is used for processing the modulation symbol sequence into M to-be-transmitted orthogonal frequency division multiplexing data packets;

[0113] allocating, by the joint coding and modulation module, the M to-be-transmitted OFDM data packets to M subcarriers according to a principle that OFDM data packets with high semantic importance are preferentially allocated to subcarriers with good channel states, based on the semantic importance vector and channel state information fed back by the receiving end for a previous image transmission, and transmitting, by an OFDM transmitter in the sending end, the M to-be-transmitted OFDM data packets to the OFDM receiver using the M subcarriers.

[0114] Optionally, the receiving end is configured to:

[0115] processing, by the OFDM receiver, the received OFDM data packets to obtain intermediate signals and providing the intermediate signals to a channel estimation module and a channel equalization module;

[0116] obtaining, by the channel estimation module, channel state information for the current image transmission based on the intermediate signals, and sending the channel state information for the current image transmission to the channel equalization module in the OFDM receiver;

[0117] obtaining, by the channel equalization module, channel equalization results based on the channel state information for the current image transmission and the intermediate signals;

[0118] performing semantic reconstruction based on the channel equalization results by a semantic information reconstruction module to obtain reconstructed semantic features;

[0119] performing image reconstruction based on the channel equalization results by a source information reconstruction module to obtain a reconstructed image.

[0120] Optionally, the receiving end is further configured to:

[0121] feeding back, by the channel estimation module, the channel state information for the current image transmission to the joint coding and modulation module;

[0122] the sending end is further configured to:

[0123] performing subcarrier allocation for a next image transmission based on the channel state information for the current image transmission by the joint coding and modulation module.

[0124] Optionally, the sending end is further configured to perform semantic encoding for an original image for the next image transmission based on the channel state information for the current image transmission by the semantic encoder.

[0125] Optionally, the sending end is specifically configured to:

[0126] inputting the semantic feature into a hyper-encoder in the semantic importance evaluation module to obtain a hyper-prior feature, and inputting the semantic feature into a first quantizer in the semantic importance evaluation module to obtain a semantic feature quantized representation, and inputting the semantic feature quantized representation into a context analysis model in the semantic importance evaluation module;

[0127] inputting the hyper-prior feature into a second quantizer in the semantic importance evaluation module to obtain a hyper-prior feature quantized representation, and inputting the hyper-prior feature quantized representation into a hyper-decoder in the semantic importance evaluation module;

[0128] processing the output of the context analysis model and the output of the hyper-decoder by a Gaussian parameter estimation module in the semantic importance evaluation module to obtain a Gaussian parameter estimation value;

[0129] processing the Gaussian parameter estimation value by an entropy calculation module in the semantic importance evaluation module to obtain an entropy value of the semantic feature as the semantic importance vector.

[0130] Optionally, the digital semantic perception OFDM subcarrier allocation system based on an entropy model is implemented through a subcarrier allocation model based on an entropy model, and a training process of the subcarrier allocation model based on the entropy model at least includes a first stage.

[0131] The training target of the first stage is to minimize the bit stream length to be transmitted and to minimize the distortion of a sample reconstructed image relative to a sample original image.

[0132] Through the first stage, a semantic encoder to be trained, a semantic importance evaluation module to be trained, and a source information reconstruction module to be trained are trained to obtain a trained semantic importance evaluation module, a semantic encoder trained through the first stage, and a source information reconstruction module trained through the first stage.

[0133] Optionally, a loss function value of the first stage is determined according to a difference between the sample original image and the sample reconstructed image, a bit stream length of a sample semantic feature quantized representation corresponding to the sample original image, and a bit stream length of a sample hyper-prior feature quantized representation corresponding to the sample original image.

[0134] Optionally, the training process of the subcarrier allocation model based on the entropy model further includes a second stage.

[0135] Through the second stage, the first stage trained semantic encoder and the first stage trained source information reconstruction module and the to-be-trained semantic information reconstruction module are trained with the parameters of the trained semantic importance evaluation module being frozen, to obtain the subcarrier allocation model based on the entropy model.

[0136] Optionally, the loss function value of the second stage is obtained according to at least the following information:

[0137] a difference between a class label of the sample semantic feature corresponding to the sample original image and a class prediction result corresponding to the sample reconstructed semantic feature;

[0138] a difference between the sample original image and the sample reconstructed image.

[0139] Optionally, the loss function value of the second stage is obtained according to at least the following information:

[0140] a rank correlation coefficient, the rank correlation coefficient representing a similarity between the sample semantic importance vector of the sample original image and sample channel state information fed back in an image transmission process of the sample original image.

[0141] Based on the same inventive concept, another embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps in the method for digital semantic perception OFDM subcarrier allocation based on an entropy model according to any one of the above embodiments of the present application.

[0142] Based on the same inventive concept, another embodiment of the present application provides an electronic device, as shown in Figure 7 Figure 7 is a schematic diagram of an electronic device according to an embodiment of the present application. The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps in the method for digital semantic perception OFDM subcarrier allocation based on an entropy model according to any one of the above embodiments of the present application when executed.

[0143] For the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts are referred to the part of the method embodiment.

[0144] Each embodiment in the specification is described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same and similar parts between the embodiments are referred to each other.

[0145] ​Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0146] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0147] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0148] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 Figure 1 The steps of the function specified in one or more boxes.

[0149] Although preferred embodiments of the present invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present invention.

[0150] Finally, it is to be understood that the phraseology or terminology such as "first" and "second" etc. used herein is merely intended to differentiate one entity or operation from another entity or operation, without necessarily requiring or implying any actual such relationship or order between such entities or operations. Moreover, the terms "comprising", "including", or any other closure, are intended to cover the non-exclusive inclusion such that a process, method, article, or apparatus that comprises a list of elements does not include those elements alone but can include other elements not expressly listed or even include elements inherent in such process, method, article, or apparatus. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0151] The above describes in detail the digital semantic perception OFDM subcarrier allocation method based on the entropy model provided by the present application. The principles and implementation modes of the present application are described by using specific examples. The above description of the embodiments is only used to help understand the method of the present application and its core idea. Meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range will be changed. In summary, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A digital semantic-aware OFDM subcarrier allocation method based on an entropy model, characterized in that, The method includes: The transmitting end performs semantic encoding on the original image through the semantic encoder in the joint coding and modulation module to obtain semantic features; The transmitting end modulates the semantic features through the joint coding and modulation module to obtain a modulation symbol sequence. The transmitting end processes the semantic features through the semantic importance evaluation module to obtain a semantic importance vector. The m-th element of the semantic importance vector corresponds to the semantic importance of the orthogonal frequency division multiplexing data packet carried by the m-th subcarrier, where m takes the value from 1 to M, and M is the number of subcarriers. The transmitting end processes the modulation symbol sequence into M orthogonal frequency division multiplexing data packets to be transmitted using an orthogonal frequency division multiplexing transmitter. The transmitting end, through the joint coding and modulation module, allocates the M orthogonal frequency division multiplexing (OFDM) data packets to the M subcarriers according to the semantic importance vector and the channel state information fed back by the orthogonal frequency division multiplexing receiver at the receiving end for the previous image transmission, following the principle of prioritizing the allocation of semantically important OFDM data packets to subcarriers with better channel state. The transmitting end then uses the M subcarriers to transmit the M OFDM data packets to the orthogonal frequency division multiplexing receiver via the OFDM transmitter. The sending end processes the semantic features through a semantic importance evaluation module to obtain a semantic importance vector, including: The sending end inputs the semantic features into the super encoder in the semantic importance evaluation module to obtain the super prior features, and inputs the semantic features into the first quantizer in the semantic importance evaluation module to obtain the semantic feature quantization representation, and inputs the semantic feature quantization representation into the context analysis model in the semantic importance evaluation module; The super-prior features are input into the second quantizer in the semantic importance evaluation module to obtain the super-prior feature quantization representation, and the super-prior feature quantization representation is input into the super decoder in the semantic importance evaluation module. The Gaussian parameter estimation module in the semantic importance assessment module processes the output of the context analysis model and the output of the super decoder to obtain the Gaussian parameter estimate. The Gaussian parameter estimate is processed by the entropy calculation module in the semantic importance assessment module to obtain the entropy value of the semantic feature, which is used as the semantic importance vector.

2. The digital semantic-aware OFDM subcarrier allocation method based on the entropy model according to claim 1, characterized in that, The method further includes: The receiving end processes the received orthogonal frequency division multiplexing data packets through the orthogonal frequency division multiplexing receiver to obtain an intermediate signal and provides the intermediate signal to the channel estimation module and the channel equalization module; The channel estimation module obtains the channel state information of this image transmission based on the intermediate signal, and sends the channel state information of this image transmission to the channel equalization module in the orthogonal frequency division multiplexing receiver; The channel equalization module obtains the channel equalization result based on the channel state information of the current image transmission and the intermediate signal; The receiving end uses a semantic information reconstruction module to perform semantic reconstruction based on the channel equalization result to obtain reconstructed semantic features. The receiving end uses the source information reconstruction module to reconstruct the image based on the channel equalization result, and obtains the reconstructed image.

3. The digital semantic-aware OFDM subcarrier allocation method based on the entropy model according to claim 2, characterized in that, The method further includes: The channel estimation module feeds back the channel state information of the current image transmission to the joint coding and modulation module; The joint coding and modulation module allocates subcarriers for the next image transmission based on the channel state information of the current image transmission.

4. The digital semantic-aware OFDM subcarrier allocation method based on the entropy model according to claim 3, characterized in that, The method further includes: The semantic encoder performs semantic encoding on the original image for the next image transmission based on the channel state information of the current image transmission.

5. The digital semantic-aware OFDM subcarrier allocation method based on the entropy model according to claim 2, characterized in that, The digital semantic-aware OFDM subcarrier allocation method based on the entropy model is implemented through an entropy model-based subcarrier allocation model, and the training process of the entropy model-based subcarrier allocation model includes at least a first stage. The training objective of the first stage is to minimize the length of the bit stream to be transmitted and to minimize the distortion of the reconstructed image relative to the original image. In the first stage, the semantic encoder to be trained, the semantic importance evaluation module to be trained, and the source information reconstruction module to be trained are trained to obtain the trained semantic importance evaluation module, the semantic encoder trained in the first stage, and the source information reconstruction module trained in the first stage.

6. The digital semantic-aware OFDM subcarrier allocation method based on the entropy model according to claim 5, characterized in that, The loss function value in the first stage is determined based on the difference between the original sample image and the reconstructed sample image, as well as the bitstream length of the sample semantic feature quantization representation corresponding to the original sample image and the bitstream length of the sample hyperprior feature quantization representation corresponding to the original sample image.

7. The digital semantic-aware OFDM subcarrier allocation method based on the entropy model according to claim 5, characterized in that, The training process of the subcarrier allocation model based on the entropy model also includes a second stage; In the second stage, while freezing the parameters of the trained semantic importance evaluation module, the semantic encoder trained in the first stage, the source information reconstruction module trained in the first stage, and the semantic information reconstruction module to be trained are trained to obtain the subcarrier allocation model based on the entropy model.

8. The digital semantic-aware OFDM subcarrier allocation method based on the entropy model according to claim 7, characterized in that, The loss function value in the second stage is obtained based on at least the following information: The difference between the category label of the semantic features corresponding to the original image of the sample and the category prediction result corresponding to the semantic features of the reconstructed sample; The difference between the original image of the sample and the reconstructed image of the sample.

9. The digital semantic-aware OFDM subcarrier allocation method based on the entropy model according to claim 8, characterized in that, The loss function value in the second stage is also obtained based on the following information: The rank correlation coefficient characterizes the similarity between the sample semantic importance vector of the original sample image and the sample channel state information fed back during the image transmission process of the original sample image.

Images