Semantic perception joint source channel coding method and related device

Abstract

The invention relates to the technical field of semantic communication, in particular to a semantic perception joint source channel coding method and a related device. The method comprises the steps of obtaining semantic features of to-be-transmitted data for representing semantic importance of the to-be-transmitted data; and selecting an information source coding mode according to the semantic features, wherein the information source coding mode comprises Polar code information source coding or LDPC code information source coding. According to the semantic perception combined source channel coding method, the semantic features of the to-be-transmitted data are extracted to quantify the semantic importance of the to-be-transmitted data, and an accurate basis is provided for subsequent coding decisions; lDPC codes are adopted for semantic secondary data to improve the information source coding efficiency, and accurate matching of an information source coding mode and semantic importance is achieved. According to the semantic perception joint source channel coding method provided by the invention, through coding collaborative design taking semantic features as a core, the problem that semantic important data transmission reliability and overall transmission efficiency are difficult to consider at the same time in the prior art is effectively solved.

Description

Technical Field

[0001] This invention relates to the field of semantic communication technology, specifically to a semantically aware joint source-channel coding method and related apparatus. Background Technology

[0002] Joint Source-Channel Coding (JSCC), a technique that unifies source coding and channel coding, effectively improves the overall performance of transmission systems by breaking down their independent boundaries. In recent years, the rise of semantic communication has injected new vitality into JSCC, enabling differentiated transmission based on semantic importance by extracting semantic features of data, thus further optimizing resource allocation.

[0003] However, existing semantically aware JSCC methods still have shortcomings: although some schemes introduce semantic features for encoding decisions, they do not deeply integrate semantic information with the encoding mode selection and parameter adjustment of Polar codes and LDPC codes. Polar codes have excellent asymptotic performance and reliability, but their encoding complexity is high; LDPC codes are known for their low complexity and high encoding efficiency, but their reliability at high signal-to-noise ratios is slightly inferior to that of Polar codes.

[0004] Existing methods either fixate on a single encoding mode, failing to dynamically adjust based on semantic features, or while achieving mode switching, the switching logic doesn't adequately consider the matching between semantic importance and encoding mode characteristics. This results in insufficient reliability of semantically important data transmission, while overall transmission efficiency is also affected by inappropriate encoding mode selection. Therefore, how to achieve dynamic coordination between Polar codes and LDPC codes at the source-channel coding level within a semantically aware framework, thereby balancing the reliability of semantically important data transmission and overall transmission efficiency in dynamic scenarios, has become a pressing technical problem to be solved. Summary of the Invention

[0005] This disclosure proposes a semantically aware joint source-channel coding method and related apparatus, aiming to overcome at least one of the defects existing in the prior art.

[0006] To achieve the above objectives, the technical solution disclosed in this invention is as follows: According to one aspect of this disclosure, a semantically aware joint source-channel coding method is provided, comprising the steps of: Obtain the semantic features of the data to be transmitted, which are used to characterize the semantic importance of the data to be transmitted; The source coding mode is selected based on the semantic features, and the source coding mode includes Polar code source coding or LDPC code source coding; Combining the semantic features and the output of the source coding mode, the channel coding method is adaptively switched, including Polar code channel coding or LDPC code channel coding; Based on the feedback information of the semantic features, the joint coding parameters of the source coding mode and the channel coding method are adjusted to optimize coding performance.

[0007] Furthermore, a Transformer-based semantic encoder is used to encode the data to be transmitted, and the semantic entropy used to characterize the semantic uncertainty of the data to be transmitted or the semantic priority used to characterize the semantic importance of the data to be transmitted is extracted as the semantic feature.

[0008] Furthermore, when the semantic importance corresponding to the semantic feature is higher than the first threshold, Polar code source coding is selected to improve the reliability of source coding; when the semantic importance corresponding to the semantic feature is lower than or equal to the first threshold, LDPC code source coding is selected to improve the efficiency of source coding.

[0009] Furthermore, the adaptive switching channel coding method includes: when the semantic priority corresponding to the semantic feature is higher than the second threshold and the channel state is fading, Polar code channel coding is selected to combat channel interference; when the semantic priority corresponding to the semantic feature is lower than or equal to the second threshold and the channel state is Gaussian white noise, LDPC code channel coding is selected to reduce coding complexity.

[0010] Furthermore, the adjustment of the joint coding parameters of the source coding mode and the channel coding method includes: receiving the semantic decoding result of the data to be transmitted, calculating the semantic loss, and iteratively adjusting the coding rate of the source coding mode and the code length of the channel coding method according to the semantic loss, wherein the semantic loss characterizes the degree of loss of semantic information of the data to be transmitted.

[0011] Furthermore, the selection of Polar code source encoding includes: adjusting the frozen bit positions of the Polar code according to the semantic features, and mapping data bits with high semantic importance to reliable bits of the Polar code to prioritize the protection of semantically important data; the selection of LDPC code source encoding includes: selecting the parity check matrix structure of the LDPC code according to the semantic features, with data bits with high semantic importance corresponding to the height nodes of the LDPC code to enhance the error correction capability of semantically important data.

[0012] Furthermore, the semantic features also include semantic relevance, which characterizes the semantic association between the data to be transmitted and the context data. The step of selecting the source coding mode based on the semantic features includes: when the semantic relevance is higher than a third threshold, selecting Polar code source coding to preserve the context semantic association; when the semantic relevance is lower than or equal to the third threshold, selecting LDPC code source coding to simplify the coding process.

[0013] Furthermore, the adjustment method of the joint coding parameters includes: when the semantic loss increases, reducing the source coding rate to improve the source coding accuracy, while increasing the channel coding rate to reduce transmission delay; when the semantic loss decreases, increasing the source coding rate to improve coding efficiency, while reducing the channel coding rate to save channel resources.

[0014] Furthermore, the data to be transmitted comes from multiple users; obtaining the semantic features of the data to be transmitted includes: obtaining the semantic features of the data to be transmitted for each user respectively; selecting the source coding mode based on the semantic features includes: independently selecting the source coding mode for the data to be transmitted for each user; adaptively switching the channel coding method includes: independently switching the channel coding method for the data to be transmitted for each user; adjusting the joint coding parameters includes: adjusting the joint coding parameters of multiple users based on the feedback information of the semantic features of multiple users to achieve multi-user resource allocation optimization.

[0015] According to another aspect of this disclosure, a semantically aware joint source-channel coding system is provided for implementing the semantically aware joint source-channel coding method as described above, comprising: A semantic feature extraction module is used to obtain the semantic features of the data to be transmitted, wherein the semantic features are used to characterize the semantic importance of the data to be transmitted; The source coding mode selection module is used to select a source coding mode based on the semantic features, wherein the source coding mode includes Polar code source coding or LDPC code source coding. The channel coding adaptive switching module is used to adaptively switch the channel coding method by combining the semantic features and the output of the source coding mode. The channel coding method includes Polar code channel coding or LDPC code channel coding. The joint coding parameter optimization module is used to adjust the joint coding parameters of the source coding mode and the channel coding mode based on the feedback information of the semantic features, so as to optimize coding performance.

[0016] According to another aspect of this disclosure, a semantically aware joint source-channel coding apparatus is provided, comprising: a processor, a memory, and a communication interface, wherein the memory stores a computer program, the processor executes the computer program to implement the semantically aware joint source-channel coding method as described above, and the communication interface is used to transmit encoded data or receive semantic feedback information.

[0017] According to another aspect of this disclosure, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program that, when executed by a processor, implements the semantically aware joint source-channel coding method as described above.

[0018] The beneficial effects of this invention are: The semantically aware joint source-channel coding method provided by this invention effectively solves the problem of balancing the reliability of semantically important data transmission and overall transmission efficiency in the prior art through coding collaborative design with semantic features as the core.

[0019] Specifically, the semantic-aware joint source-channel coding method of this invention extracts semantic features of the data to be transmitted to quantify its semantic importance, providing a precise basis for subsequent coding decisions. Then, it selects the source coding mode based on the semantic features: prioritizing Polar codes for semantically important data to ensure the reliability of source coding, and using LDPC codes for semantically less important data to improve source coding efficiency, achieving a precise match between the source coding mode and semantic importance. Subsequently, it adaptively switches the channel coding mode by combining semantic features and source coding output, further optimizing the channel protection strategy for semantically important data. Finally, it adjusts the joint coding parameters through semantic feedback information to ensure that coding performance dynamically improves with the scenario. This collaborative design of Polar codes and LDPC codes, using semantic features as a bridge, ensures accurate transmission of semantically important data through the high reliability of Polar codes and improves overall transmission performance through the high efficiency of LDPC codes, achieving an optimal balance between the reliability of semantically important data transmission and overall transmission efficiency in dynamic scenarios. It is particularly suitable for scenarios such as 6G where semantic accuracy and transmission performance requirements are extremely high, demonstrating significant technological advancement and application value.

[0020] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0021] Figure 1 This is a flowchart of a semantically aware joint source-channel coding method in one embodiment of the present invention; Figure 2This is a schematic diagram of semantic feature hierarchical spatial mapping in one embodiment of the present invention; Figure 3 This is a comparison diagram of the joint encoding of Polar code and LDPC code in one embodiment of the present invention; Figure 4 This is a schematic diagram of the joint coding parameter optimization surface in one embodiment of the present invention; Figure 5 This is a schematic diagram illustrating multi-user resource allocation optimization in one embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0023] In embodiments of the present invention, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" or "for example" in embodiments of the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Rather, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0024] The present invention provides the following preferred embodiments: Example 1: To address the problem that existing joint source-channel coding methods do not fully utilize the semantic features of the data to be transmitted, leading to a mismatch between coding strategies and the semantic importance requirements of the data, this example provides a semantically aware joint source-channel coding method. This method refines the entire processing logic from semantic feature extraction to feedback optimization, achieving dynamic adaptation between coding decisions and the semantic value of the data. For example... Figure 1 As shown, the steps of the channel coding method are as follows: S100: Obtain the semantic features of the data to be transmitted, which are used to characterize the semantic importance of the data to be transmitted.

[0025] S200: Select the source coding mode based on semantic features. The source coding mode includes Polar code source coding or LDPC code source coding.

[0026] S300: Combining the output of semantic features and source coding modes, it adaptively switches the channel coding mode, which includes Polar code channel coding or LDPC code channel coding.

[0027] S400: Based on feedback information from semantic features, adjust the joint coding parameters of the source coding mode and the channel coding mode to optimize coding performance.

[0028] Specifically, the semantic feature extraction and quantization of the data to be transmitted employs the Transformer model to parse the semantic information in the data, generating semantic vectors containing multi-dimensional features such as object recognition, scene understanding, and relationship reasoning, for example... Figure 2 As shown in the diagram, the hierarchical spatial mapping of semantic features illustrates the distribution of these multidimensional features in three-dimensional space. Different categories of features are represented by scatter points of different colors, clearly depicting the hierarchical structure of the semantic features. Furthermore, the semantic vectors are mapped to continuous semantic importance coefficients through a quantization module. These coefficients, through nonlinear transformation, compress the high-dimensional semantic features into single-dimensional indicators, accurately representing the differences in semantic value among different data blocks, and providing a quantitative basis for subsequent encoding mode selection and parameter adjustment.

[0029] Furthermore, the source coding mode is selected based on the semantic importance coefficient. The source coding mode selection module divides the data to be encoded into different semantic value regions. When the semantic importance coefficient exceeds a preset threshold, the Polar code source coding mode is used, leveraging its polarization characteristics to accurately compress high semantic value data, ensuring low-loss encoding of key semantic information. When the semantic importance coefficient is below the threshold, the LDPC code source coding mode is used, utilizing its sparse parity-check matrix structure to improve the compression efficiency of large-scale conventional semantic data, such as... Figure 3 As shown, the comparison diagram of joint encoding of Polar codes and LDPC codes presents the distribution of the two encoding modes in the semantic importance and transmission efficiency space. The decision boundary curve clarifies the switching conditions of the two modes. Polar codes are distributed in the high semantic importance region, while LDPC codes are distributed in the high transmission efficiency region, forming a complementarity between semantic value and encoding efficiency.

[0030] Based on this, the system adaptively switches the channel coding scheme by combining semantic features and source coding output. After receiving the codewords from the source coding output, the system monitors channel state information and semantic importance coefficients in real time. When the semantic importance coefficient corresponding to the source codeword is high, it switches to the Polar code channel coding scheme, using its strong error correction capability to ensure reliable transmission of high semantic value information. When the semantic importance coefficient corresponding to the source codeword is low, it switches to the LDPC code channel coding scheme, using its high throughput characteristics to optimize the transmission efficiency of conventional semantic data. This switching process is based on the compatible reconstruction of the source codeword structure and the channel coding matrix, ensuring the consistency of the codeword format during coding switching and avoiding additional transmission overhead.

[0031] Furthermore, the joint coding parameters are adjusted based on feedback information from semantic features. A semantic feature feedback loop is established, continuously collecting semantic fidelity evaluation data from the receiver and semantic loss indicators from the transmitter. The source coding rate and channel coding code length are dynamically corrected through a gradient feedback adjustment mechanism, such as... Figure 4As shown, the joint coding parameter optimization surface diagram depicts the impact of source coding rate and channel coding code length on semantic transmission performance. The surface presents a continuous trend of semantic transmission performance as the parameters change, and the iterative optimization path demonstrates the process of parameters converging from the initial state to the optimal operating point. For example, when feedback data indicates that the current semantic fidelity is lower than the system requirements, the source coding redundancy of key semantic regions is increased, and the number of coding bits is increased to reduce semantic information loss; when the channel quality is significantly improved, the channel coding code length is reduced according to a non-linear relationship, thereby reducing transmission overhead while ensuring semantic transmission performance.

[0032] The advantage of this embodiment lies in constructing a semantic feature-driven closed-loop joint coding system, forming a complete technology chain from semantic feature extraction and coding mode selection to parameter optimization, achieving adaptive matching between coding strategy and data semantic importance. Figures 1 to 4 The collaborative display clearly demonstrates the fundamental role of semantic features in coding decisions, as well as the dynamic adjustment process of joint coding parameters with semantic feedback, forming a full-process technical support from semantic parsing to coding optimization.

[0033] Example 2: To address the problem that existing semantic feature extraction methods are unable to accurately characterize the semantic uncertainty and importance of the data to be transmitted, this example further refines the semantic feature extraction process based on Transformer. By generating two types of features, semantic entropy and semantic priority, multi-dimensional quantification of semantic information is achieved, providing a more comprehensive decision-making basis for subsequent encoding mode selection.

[0034] First, a Transformer-based semantic encoder architecture is constructed. The encoder's input layer converts the data to be transmitted (such as text, images, or video frames) into fixed-dimensional word embedding vectors and adds positional encoding to preserve the temporal or spatial structural information of the data. The encoder layer consists of multiple stacked self-attention mechanisms and feedforward neural network layers. The self-attention mechanism captures the internal dependencies of semantic content by calculating the association weights between data elements, such as the relationship between subject and predicate in text, or the relationship between foreground objects and background in images. The feedforward neural network performs non-linear transformations on these associated features to enhance the representation ability of semantic information. The output layer converts the output of the encoder layer into a global semantic vector through global average pooling, and then generates semantic entropy and semantic priority through two parallel fully connected layers. Semantic entropy is calculated using a probability distribution, reflecting the uncertainty of the semantic content of the data to be transmitted; for example, text with ambiguous descriptions or low-resolution images have higher semantic entropy. Semantic priority is calculated using the cosine similarity between the semantic vector and a pre-trained semantic importance template, reflecting the degree of contribution of the data to overall semantic understanding; for example, image patches containing key object recognition information have higher semantic priority.

[0035] Furthermore, semantic entropy and semantic priority are weighted and fused to generate comprehensive semantic features. The fusion weights are determined based on the system's need to balance semantic uncertainty and importance. For example, when the system focuses more on the accuracy of semantic content, the weight of semantic entropy is increased; when it focuses more on the criticality of semantic content, the weight of semantic priority is increased. The fused comprehensive semantic features are input into the semantic feature quantization module, where a linear transformation compresses the high-dimensional features into a one-dimensional semantic importance coefficient. This coefficient serves as the core basis for subsequent source coding mode selection.

[0036] It's important to understand that the combined use of semantic entropy and semantic priority can overcome the limitations of a single feature: semantic entropy reflects the ambiguity of semantic content, providing a reference for "how much redundancy is needed" in encoding pattern selection; semantic priority reflects the importance of semantic content, providing a reference for "how much redundancy is worthwhile" in encoding pattern selection. The combination of the two makes encoding decisions more aligned with the semantic essence of the data. For example... Figure 2 As shown in the diagram, the semantic feature hierarchical spatial mapping diagram illustrates the three-dimensional distribution of semantic entropy, semantic priority, and comprehensive semantic importance. The positions of different data points correspond to the differences in their semantic features. Data points with high semantic entropy and high semantic priority are distributed in the upper right corner of the space, corresponding to the region with high comprehensive semantic importance, providing an intuitive decision boundary for the selection of source coding mode.

[0037] The advantage of this embodiment is that it achieves accurate multi-dimensional extraction of semantic features through a Transformer-based semantic encoder. The combined use of semantic entropy and semantic priority makes the representation of semantic information more comprehensive, providing a more reliable basis for subsequent encoding mode selection and ensuring a high degree of adaptation between the encoding strategy and the semantic requirements of the data.

[0038] Example 3: To address the issue that the selection of source coding modes does not fully consider differences in semantic importance, this example further refines the decision-making logic of source coding modes based on the comparison of semantic importance and a first threshold, achieving a dynamic trade-off between reliability and efficiency, and making the source coding strategy more aligned with the semantic value of the data.

[0039] Specifically, the first threshold value is determined based on the statistical distribution of the semantic importance coefficient of the data to be transmitted. By analyzing the semantic importance coefficients of a large amount of historical data, a critical value that maximizes the product of system reliability and efficiency is selected. For example, when the semantic importance coefficient exceeds the first threshold, it means that the data's contribution to the overall semantic understanding exceeds the system's preset importance threshold. In this case, the system needs to prioritize ensuring the encoding reliability of the data. When the semantic importance coefficient is lower than or equal to the first threshold, it means that the data's contribution has not reached the preset threshold. In this case, the system can prioritize improving encoding efficiency to save resources.

[0040] Furthermore, the source coding mode is selected based on the comparison between semantic importance and a first threshold. When the semantic importance coefficient is higher than the first threshold, Polar code source coding is selected. Polar codes separate reliable and unreliable channels through channel polarization, allocating more reliable channels to data with high semantic importance, achieving accurate compression with low loss. For example, for text data blocks containing user instructions, Polar code source coding adjusts the position of frozen bits to encode key semantic information in the instruction, such as "turn on" and "air conditioner" in "turn on the air conditioner," into reliable channels, reducing semantic loss during compression and ensuring that the receiver can accurately parse the instruction content. When the semantic importance coefficient is lower than or equal to the first threshold, LDPC code source coding is selected. The sparse parity-check matrix structure of LDPC codes enables highly parallel coding, allowing for rapid processing of large-scale conventional semantic data and improving coding efficiency. For example, for image data blocks containing background scene information, such as the sky and ground, LDPC code source coding significantly shortens coding time and improves system throughput by rapidly operating the sparse matrix while ensuring the basic preservation of background semantic information.

[0041] It's important to understand that the choice between Polar codes and LDPC codes is not absolutely contradictory, but rather a trade-off based on semantic importance: Polar codes excel in reliable compression of high semantic value data, but their encoding complexity is higher; LDPC codes excel in efficient processing of low semantic value data, but their reliability is slightly lower. The setting of the first threshold achieves complementarity between the two, allowing the system to ensure the reliability of high semantic value data while also considering the processing efficiency of low semantic value data. For example... Figure 3 As shown, the comparison diagram of joint encoding of Polar code and LDPC code illustrates the distribution of the two encoding modes in the semantic importance and transmission efficiency space. The first threshold corresponds to the decision boundary curve in the diagram. The left side of the curve is the efficient region of LDPC code (low semantic importance), and the right side is the reliable region of Polar code (high semantic importance), clearly depicting the logic of mode selection.

[0042] The advantage of this embodiment is that by setting the first threshold, dynamic adaptation between the source coding mode and semantic importance is achieved. The complementary use of Polar codes and LDPC codes enables the system to achieve a balance between reliability and efficiency, ensuring that high semantic value data is reliably compressed and low semantic value data is efficiently processed, thereby improving the overall performance of the joint source-channel coding system.

[0043] Example 4: To address the issue that channel coding methods cannot simultaneously adapt to semantic requirements and channel states, this example further refines the adaptive switching logic that combines semantic priority, a second threshold, and channel states. This achieves a balance between channel interference mitigation and coding complexity, making the channel coding strategy more aligned with data semantic requirements and the transmission environment.

[0044] Specifically, the value of the second threshold is determined based on the distribution of semantic priority and the system's requirements for interference mitigation. For example, when the semantic priority exceeds the second threshold, it indicates that the impact of the data on the user experience exceeds the system's preset threshold, such as data packets containing critical user requests. In this case, the system needs to prioritize combating channel interference to ensure reliable data transmission. When the semantic priority is lower than or equal to the second threshold, it indicates that the data impact has not reached the preset threshold, such as data packets containing advertising pushes. In this case, the system can prioritize reducing coding complexity to save terminal device energy consumption.

[0045] Furthermore, the system monitors the channel state. It estimates the channel state using pilot signals fed back from the receiver, distinguishing between fading channels (such as multipath fading channels) and Gaussian white noise channels: the signal amplitude of fading channels varies with time or frequency, easily leading to data transmission errors; Gaussian white noise channels have a stable noise distribution and a lower error rate. Channel state information and semantic priority together serve as the basis for switching channel coding methods.

[0046] Furthermore, the channel coding method is selected based on a combination of semantic priority, the second threshold, and the channel state. When the semantic priority is higher than the second threshold and the channel state is fading, Polar code channel coding is selected. The polarization characteristics of Polar codes give them stronger error correction capabilities in fading channels, effectively combating interference caused by signal amplitude variations. For example, for data packets containing user medical data, the semantic priority is high. When the transmission channel is experiencing multipath fading, Polar code channel coding increases the number of error correction bits by adjusting the code length and code rate, ensuring that critical information in the medical data (such as blood pressure and heart rate) is not corrupted by channel interference. When the semantic priority is lower than or equal to the second threshold and the channel state is Gaussian white noise, LDPC code channel coding is selected. The iterative decoding algorithm of LDPC codes has low complexity and can quickly complete decoding in Gaussian white noise channels, reducing the power consumption of terminal devices. For example, for data packets containing news pushes, the semantic priority is low. When the transmission channel is Gaussian white noise, LDPC code channel coding reduces the computational burden of terminal devices and extends battery life through fast decoding of the sparse parity-check matrix.

[0047] It's important to understand that switching channel coding methods doesn't involve considering semantic requirements or channel conditions in isolation, but rather a combination of both: semantic priority determines "whether interference needs to be mitigated," while channel conditions determine "what type of interference needs to be mitigated." This combined logic makes channel coding strategies more aligned with actual transmission scenarios, ensuring the reliability of high-semantic-value data in poor channels while reducing the complexity of low-semantic-value data in good channels.

[0048] The advantage of this embodiment is that by combining semantic priority, second threshold and adaptive switching logic of channel state, dynamic adaptation of channel coding method with semantic requirements and transmission environment is achieved. Reasonable switching between Polar code and LDPC code enables the system to achieve a balance between combating channel interference and reducing coding complexity, thereby improving the transmission performance and resource utilization of joint source channel coding system.

[0049] Example 5: To address the problem that existing joint coding parameter adjustments do not fully consider the dynamic changes in semantic loss, resulting in insufficient adaptability of coding performance to semantic requirements, this example further refines the joint coding parameter adjustment logic based on iterative semantic loss, and achieves a dynamic balance between source coding rate and channel coding code length through semantic decoding result feedback.

[0050] First, the receiving end performs semantic decoding on the transmitted data. It calculates the semantic loss using the cosine similarity between the semantic vector and the original data's semantic vector. This loss value characterizes the degree of semantic information loss in the transmitted data; that is, the lower the similarity, the greater the semantic loss. The semantic loss is then input into the parameter optimization module, which is based on a pre-trained joint encoding parameter optimization model, such as... Figure 4 As shown, the joint coding parameter optimization surface diagram illustrates the three-dimensional relationship between source coding rate, channel coding code length, and semantic loss, outputting the initial adjustment direction: When semantic loss increases, it indicates that source coding is over-compressing semantic information or channel coding is insufficiently resistant to interference. In this case, the source coding rate is reduced to increase the redundancy of source coding and improve the accuracy of semantic information preservation; simultaneously, the channel coding rate is increased to shorten the channel coding code length and reduce transmission delay. It is important to understand that the combination of reducing the source coding rate and increasing the channel coding rate not only compensates for the accuracy problem caused by the increase in semantic loss but also avoids the delay degradation caused by the increase in channel coding code length. When semantic loss decreases, it indicates that the current coding parameters meet the semantic preservation requirements. In this case, the source coding rate is increased to reduce the redundancy of source coding and improve coding efficiency; simultaneously, the channel coding rate is reduced to increase the redundancy of channel coding and save channel resources. This adjustment logic achieves a balance between efficiency and resources, maximizing coding efficiency and resource utilization within an acceptable range of semantic loss.

[0051] Furthermore, parameter adjustment employs an iterative approach. After each round of transmission, the receiver reports the semantic loss, and the parameter optimization module adjusts the encoding parameters based on the latest loss value until the semantic loss stabilizes within a preset acceptable range. During the iteration process, the step size of parameter adjustment gradually decreases to avoid performance fluctuations caused by over-adjustment. For example, in the initial rounds when the semantic loss is large, the adjustment step size is large to quickly approach the optimal parameter region; in subsequent rounds when the semantic loss is small, the adjustment step size is small to finely optimize the parameters. This iterative approach ensures that parameter adjustment always follows changes in semantic loss, achieving dynamic adaptation.

[0052] The advantage of this embodiment is that it achieves precise matching between the source coding rate and the channel coding code length through dynamic feedback and iterative adjustment of semantic loss, so that the joint coding parameters are always optimized around semantic requirements, which not only ensures the effective preservation of semantic information, but also takes into account transmission efficiency and resource utilization.

[0053] Example 6: To address the problem that existing Polar and LDPC source coding is not optimized for semantically important data, resulting in key semantic information being susceptible to coding loss, this example further refines the semantic adaptation mechanism of Polar and LDPC codes. By adjusting the frozen bit positions of Polar codes and the parity check matrix structure of LDPC codes, priority protection of semantically important data is achieved.

[0054] For Polar code source encoding, data bits are first sorted according to semantic features such as semantic priority, mapping semantically important data bits to reliable bits in the Polar code. Polar codes divide the channel into reliable and unreliable channels through channel polarization; reliable bits correspond to reliable channels, and their error rate is much lower than that of unreliable bits. Through this mapping, semantically important data bits are allocated to more reliable transmission paths during encoding, reducing semantic loss caused by encoding compression. It's important to understand that the adjustment of frozen bit positions is not random, but a precise mapping based on semantic features, ensuring that each semantically important data bit receives optimal protection resources. For example, mapping text bits containing core user instruction information to the first 10% of reliable bits in the Polar code reduces the error rate of that information to 10%. -6 the following.

[0055] Furthermore, for LDPC source coding, the structure of the parity-check matrix is ​​selected based on semantic features. Specifically, data bits with high semantic importance correspond to high-weight nodes in the LDPC parity-check matrix. High-weight nodes are information bit nodes with a larger number of connected parity bits, and their corresponding parity equations are more numerous, resulting in stronger error correction capabilities. For example, in image data coding, data bits containing key object edge information (high semantic importance) are assigned to nodes with larger row weights in the parity-check matrix, where row weight refers to the number of parity bits connected to that node. Conversely, background information data bits (low semantic importance) are assigned to nodes with smaller row weights. This structural selection allows semantically important data bits to obtain more parity information support during decoding, enhancing their error correction capabilities. It can be understood that the structural adjustment of the LDPC parity-check matrix essentially tilts limited error correction resources toward semantically important data bits, achieving efficient resource utilization.

[0056] The advantage of this embodiment is that by freezing the bit positions of the Polar code and adjusting the semantic adaptation of the LDPC code parity-check matrix structure, the source coding process is made to better fit the semantic essence of the data. Semantically important data bits are more effectively protected, the loss of key semantic information is reduced, and the semantic preservation capability of source coding is improved.

[0057] Example 7: To address the problem that existing source coding does not consider the semantic correlation between the data to be transmitted and the context data, which leads to the easy loss of context-dependent semantic information, this example further introduces semantic relevance features to characterize semantic correlation and optimizes the source coding mode selection logic based on these features.

[0058] Specifically, semantic relevance features are calculated using the semantic vector similarity between the data to be transmitted and the context data. Context data includes historical transmission data, preceding data from the current session, or related data from the same scene. Semantic vectors are generated using a pre-trained Transformer model. For example, in a dialogue system, the cosine similarity between the semantic vector of the current statement and the semantic vector of the previous statement represents the semantic relevance of that statement; in video transmission, the similarity between the semantic vector of the current frame and the semantic vector of the previous frame represents the semantic relevance of that frame. Higher semantic relevance indicates a closer connection between the data to be transmitted and the context data, and a greater impact of the loss of semantic information on overall semantic understanding.

[0059] Furthermore, the third threshold is determined based on the degree of influence of contextual relevance on semantic understanding. For example, when the semantic relevance is higher than the third threshold, it indicates that the data to be transmitted is highly correlated with the contextual data, and the loss of its semantic information will lead to a break in the semantic understanding of the context. For example, the loss of a transitional sentence in a dialogue will make the meaning of the entire dialogue incomprehensible. When the semantic relevance is lower than or equal to the third threshold, it indicates that the data to be transmitted is not highly correlated with the contextual data, and the loss of its semantic information has a smaller impact on the overall semantic understanding. For example, the loss of independent advertising information will not affect the understanding of the main content.

[0060] Furthermore, the source coding mode is selected based on the comparison between semantic relevance and the third threshold: when the semantic relevance is higher than the third threshold, Polar code source coding is selected. The channel polarization characteristics of Polar codes enable them to more reliably preserve the semantic structure of the data and avoid the loss of contextual information. For example, concluding sentences in a dialogue are encoded using Polar codes to ensure that their semantic connection with the preceding text is not destroyed. When the semantic relevance is lower than or equal to the third threshold, LDPC code source coding is selected. The sparse parity-check matrix structure of LDPC codes simplifies the coding process, enables the rapid processing of semantically weakly related data, and improves coding efficiency. For example, independent advertising information is encoded using LDPC codes, which shortens the coding time while ensuring basic semantic preservation.

[0061] The advantage of this embodiment is that by introducing semantic relevance features and setting a third threshold, the source coding mode and the context semantic relevance are adapted. Data to be transmitted with high context relevance receives more reliable coding protection and avoids the break in context semantic understanding; data with low relevance receives more efficient coding processing, thus improving the overall coding efficiency of the system.

[0062] Example 8: To address the problem that existing joint coding parameter adjustments fail to balance semantic precision and transmission efficiency, resulting in either significant semantic loss or high latency, this example further refines the semantic loss-driven joint coding parameter adjustment method. By adjusting the source coding rate and the channel coding rate inversely, a dynamic balance between semantic precision and transmission efficiency is achieved.

[0063] When semantic loss increases, it indicates that the current source coding is over-compressing semantic information, leading to the loss of key semantic information. In this case, reducing the source coding rate increases redundancy per unit of data, allowing the source encoder to more accurately retain semantic information and reduce semantic loss. Simultaneously, increasing the channel coding rate shortens the code length per unit of data, reducing transmission time and latency. It's important to understand that the combination of decreasing the source coding rate and increasing the channel coding rate addresses the accuracy issue caused by increased semantic loss while avoiding the latency degradation caused by increased channel coding code length, achieving a balance between accuracy and latency. For example, when the semantic loss increases from 0.2 to 0.5, decreasing the source coding rate from 0.8 to 0.5 and increasing the channel coding rate from 0.6 to 0.9 reduces the semantic loss back to 0.3, while simultaneously reducing latency from 100ms to 80ms.

[0064] Furthermore, when semantic loss decreases, it indicates that the current source coding is sufficiently accurate and semantic information is well preserved. In this case, increasing the source coding rate reduces redundancy, improves coding efficiency, and allows for the processing of more data. Simultaneously, decreasing the channel coding rate increases code length and redundancy, but since semantic loss is small, high channel protection is unnecessary, and reducing the code rate conserves channel resources. This adjustment logic achieves a balance between efficiency and resources, maximizing coding efficiency and resource utilization within an acceptable range of semantic loss. For example, when semantic loss decreases from 0.3 to 0.1, increasing the source coding rate from 0.5 to 0.8 and decreasing the channel coding rate from 0.9 to 0.6 increases coding efficiency from 50 Mbps to 80 Mbps, while simultaneously reducing channel resource occupancy from 90% to 60%.

[0065] Furthermore, the magnitude of parameter adjustment is determined based on the rate of change of semantic loss. The faster the semantic loss increases, the greater the decrease in source coding rate and the greater the increase in channel coding rate, thus quickly suppressing the growth of semantic loss; conversely, the faster the semantic loss decreases, the greater the increase in source coding rate and the greater the decrease in channel coding rate, thus rapidly improving coding efficiency and resource utilization. This dynamic adjustment method makes parameter adjustment more closely match the changing trend of semantic loss, improving the timeliness and effectiveness of parameter adjustment.

[0066] The advantage of this embodiment is that by adjusting the source coding rate and the channel coding rate inversely, a dynamic balance between semantic accuracy and transmission efficiency is achieved. When semantic loss increases, accuracy and latency are prioritized, and when semantic loss decreases, efficiency and resources are prioritized, thereby improving the adaptive capability of the joint coding system.

[0067] Example 9: To address the problem that existing multi-user scenarios do not optimize coding mode selection and resource allocation for each user's semantic features, resulting in unfair or inefficient resource allocation, this example further refines the semantic-aware joint source-channel coding logic in multi-user scenarios. By allowing each user to independently select a coding mode and optimizing multi-user resource allocation, the semantic needs and resources of multiple users are adapted.

[0068] Specifically, for each user's data to be transmitted, its semantic features are independently acquired. These features include semantic entropy, semantic priority, and semantic relevance, and are generated using a dedicated semantic encoder for each user. For example, user A's text data uses a text semantic encoder to generate semantic priority, while user B's image data uses an image semantic encoder to generate semantic entropy. Then, a source coding mode is independently selected based on each user's semantic features. For instance, user A's data to be transmitted has a high semantic priority, such as containing medical data, so Polar code source coding is chosen to preserve key semantic information; user B's data has a low semantic priority, such as containing news feeds, so LDPC code source coding is chosen to improve coding efficiency. Simultaneously, the channel coding method is independently switched based on each user's semantic features and channel state. For example, user A's channel state is fading and has a high semantic priority, so Polar code channel coding is chosen to combat channel interference; user B's channel state is Gaussian white noise and has a low semantic priority, so LDPC code channel coding is chosen to reduce coding complexity. It is important to understand that each user's coding mode selection is independent, based on their own semantic features and channel state, ensuring that each user's coding strategy aligns with their specific needs.

[0069] Furthermore, based on semantic feature feedback information from multiple users, including each user's semantic loss, transmission delay, and channel occupancy, the joint coding parameters for multiple users are adjusted to optimize resource allocation. The goal of resource allocation optimization is to maximize the overall resource utilization of the system while meeting the semantic loss requirements of each user. Figure 5 As shown in the diagram, the multi-user resource allocation optimization illustrates the relationship between resource allocation and semantic features for each user. Users with higher semantic priority receive more resources. For example, when user A's semantic loss increases, its source coding redundancy is increased (source coding rate is reduced), while its channel resource allocation is increased (channel coding rate is increased) to meet its semantic accuracy requirements. Conversely, when user B's semantic loss decreases, its source coding redundancy is reduced (source coding rate is increased), while its channel resource allocation is reduced (channel coding rate is decreased) to conserve resources for users who need them more (such as user A). The weight of resource allocation is based on the user's semantic priority; users with higher semantic priority receive higher resource allocation weights, ensuring that their semantic needs are met first.

[0070] The advantage of this embodiment is that by allowing each user to independently select an encoding mode and optimizing resource allocation for multiple users, it achieves precise matching of semantic requirements and resources in multi-user scenarios. Each user's encoding strategy is tailored to its own semantic characteristics and channel state, and resource allocation is tilted towards users with high semantic priority, thereby improving the overall semantic preservation capability and resource utilization of the system.

[0071] Example 10: To address the problem that existing joint source-channel coding systems do not fully integrate semantic features and module collaboration, resulting in insufficient adaptability between coding strategies and semantic requirements, this example provides a semantically aware joint source-channel coding system. Through semantic feature-driven inter-module information transfer and closed-loop optimization, a dynamic balance between semantic preservation and coding performance is achieved.

[0072] The semantic feature extraction module, serving as the system's perception layer, performs semantic analysis on the data to be transmitted using a pre-trained multimodal semantic encoder that supports text, images, audio, and other data types, generating semantic features that characterize semantic importance. Semantic priority is calculated by matching the semantic vector with a pre-defined key semantic vector library, while semantic entropy is obtained by measuring the uncertainty of the semantic distribution. These semantic features not only accurately depict the semantic essence of the data to be transmitted but also provide a foundational basis for decisions made by subsequent modules. For example, key indicators in medical diagnostic reports have high semantic priority and require more reliable encoding protection; while repetitive advertising push information has low semantic entropy and can be encoded using more efficient methods. The output of the semantic feature extraction module is not an isolated numerical value but a semantic description deeply bound to data type and application scenario, ensuring that subsequent decisions always revolve around semantic requirements.

[0073] Furthermore, based on the output of the semantic feature extraction module, the source coding mode selection module serves as the system's decision layer, achieving precise matching between semantic features and source coding modes. When the semantic priority is higher than a preset threshold, Polar code source coding is selected. The channel polarization characteristics of Polar codes enable them to recursively map semantically important data bits to reliable bits, effectively preserving key semantic information. When the semantic priority is lower than or equal to the preset threshold, LDPC code source coding is selected. The sparse parity-check matrix structure of LDPC codes results in low coding process complexity, improving coding efficiency while ensuring basic semantic preservation. It is important to understand that the core of the selection logic is the synergy between semantic features and coding mode characteristics: the high reliability of Polar codes is suitable for semantically important data that needs precise preservation, while the high efficiency of LDPC codes is suitable for semantically secondary data that prioritizes transmission speed. This matching avoids a "one-size-fits-all" coding strategy, making source coding more closely aligned with the semantic value of the data.

[0074] Furthermore, the adaptive switching module for channel coding, acting as a collaborative layer of the system, combines the outputs of the semantic feature extraction module, the source coding mode selection module, and real-time channel conditions, such as channel fading and noise type, to dynamically switch channel coding methods. For example, when the source coding mode is Polar code (corresponding to semantically important data) and the channel condition is frequency-selective fading, Polar code channel coding is selected. Its polarization characteristics can allocate reliable channels to semantically important data through channel polarization, effectively combating interference caused by channel fading. When the source coding mode is LDPC code (corresponding to semantically secondary data) and the channel condition is Gaussian white noise, LDPC code channel coding is selected. Its iterative decoding characteristics of the sparse parity-check matrix can achieve sufficient error correction capability with low complexity, reducing transmission delay. It is understandable that "adaptive switching" is not a simple mode switch, but a collaborative decision-making process integrating semantic features, source coding strategies, and channel conditions to ensure that channel coding and source coding form a "dual protection" or "efficiency-first" combination, achieving adaptation of the coding system to the channel environment.

[0075] Furthermore, the joint coding parameter optimization module, as the system's optimization layer, dynamically adjusts coding performance through a closed-loop feedback mechanism. This module receives semantic feature feedback information from the receiver, such as semantic loss (the difference between the transmitted and original semantic information), and adjusts the joint coding parameters of the source coding mode and channel coding mode, such as the source coding rate and channel coding code length, based on the feedback. For example, when the semantic loss increases, it indicates that the current source coding is over-compressing the semantic information. In this case, the source coding rate is reduced to increase the redundancy of the source coding and improve the accuracy of semantic information preservation; simultaneously, the channel coding rate is increased to shorten the channel coding code length and reduce transmission delay. This adjustment compensates for the semantic loss while avoiding delay degradation. When the semantic loss decreases, it indicates that the current coding parameters meet the semantic preservation requirements. In this case, the source coding rate is increased to reduce the redundancy of the source coding and improve coding efficiency; simultaneously, the channel coding rate is reduced to increase the redundancy of the channel coding and save channel resources. This adjustment achieves a balance between efficiency and resources while ensuring semantic requirements are met. The parameter optimization process adopts a cyclic iterative approach, updating the parameters after each round of transmission until the semantic loss stabilizes within a preset acceptable range, ensuring that the coding parameters are always optimized in accordance with changes in semantic requirements and channel conditions.

[0076] The advantage of this embodiment lies in achieving precise adaptation of the joint source-channel coding system to semantic requirements through semantic feature-driven inter-module collaboration and closed-loop optimization. The semantic feature extraction module provides the foundation for semantic awareness, the source coding mode selection module achieves the matching of semantics and source coding, the channel coding adaptive switching module achieves the collaboration between source and channel coding, and the joint coding parameter optimization module achieves dynamic optimization of coding performance, ensuring both effective preservation of semantic information and consideration of transmission efficiency and resource utilization.

[0077] Example 11: To address the problem that existing coding devices do not fully integrate the collaboration between hardware components and semantic-aware joint source-channel coding methods, resulting in a difficulty in balancing semantic preservation and transmission efficiency, this example provides a semantic-aware joint source-channel coding device. It further refines the hardware architecture and component collaboration mechanism of the semantic-aware joint source-channel coding device. Through the logical connection and functional division of the processor, memory, and communication interface, it achieves full-link optimization of the semantic-driven coding process.

[0078] Specifically, the processor executes computer programs stored in memory to implement the entire logic process of semantic feature extraction, source coding mode selection, channel coding mode switching, and joint coding parameter optimization. For example, after receiving data to be transmitted, the processor calls a semantic feature extraction algorithm to perform semantic analysis on the data, generating features such as semantic priority and semantic entropy; then, based on the semantic features, it selects either Polar code or LDPC code source coding mode; next, it switches the channel coding mode by combining real-time channel state information; finally, based on the semantic loss information fed back from the receiver, it adjusts the source coding rate and channel coding rate to achieve a balance between semantic accuracy and transmission efficiency. The processor's multi-core architecture supports parallel processing, allowing semantic feature extraction and coding parameter calculation to be performed simultaneously, thus improving processing speed.

[0079] Furthermore, the memory is used to store computer programs and related data, including data to be transmitted, semantic features, encoding parameters, and feedback information from the receiving end. The memory employs a hierarchical structure: a cache is used for temporary storage of frequently accessed semantic feature data to accelerate processor read speeds; non-volatile memory is used for long-term storage of computer programs and preset semantic vector libraries, ensuring data is not lost after device restart. The memory capacity is designed to adapt to different application scenarios. For example, in medical data transmission scenarios, the memory reserves sufficient space to store historical encoding parameters for high semantic priority data, facilitating rapid retrieval.

[0080] Furthermore, the communication interface, acting as a bridge between the device and the external environment, supports multiple network protocols for transmitting encoded data to the receiving end and receiving semantic feedback information from the receiving end. The low latency of the communication interface ensures that feedback information is delivered to the processor in a timely manner, supporting real-time parameter adjustment. Multi-protocol support allows the device to adapt to different network environments; for example, it uses high-frequency transmission of encoded data under 5G networks and adaptive modulation and coding under Wi-Fi networks, improving compatibility. The communication interface and the processor are connected via a high-speed bus, and the data transmission rate meets the requirements of real-time encoding.

[0081] The advantage of this embodiment is that through the logical division of labor and functional coordination of hardware components, the semantic-aware joint source-channel coding method is executed efficiently. The computing power of the processor, the storage capacity of the memory, and the transmission capacity of the communication interface complement each other, which not only ensures the effective preservation of semantic information, but also improves the efficiency of coding and transmission.

[0082] Example 12: To address the problem that existing computer-readable storage media are not optimized for semantically aware joint source-channel coding methods, resulting in low program execution efficiency or poor compatibility, this example provides a computer-readable storage medium. Through the adaptation design of the storage medium and the coding method, efficient program execution and accurate satisfaction of semantic requirements are achieved.

[0083] The computer program stored on a computer-readable storage medium adopts a modular design, including a semantic feature extraction module, a source coding mode selection module, a channel coding adaptive switching module, and a joint coding parameter optimization module. These modules interact with each other through standardized interfaces. For example, the semantic priority data output by the semantic feature extraction module is passed to the source coding mode selection module via the interface, serving as the basis for selecting Polar or LDPC codes; the channel coding adaptive switching module receives source coding mode information and real-time channel state information, and outputs the corresponding channel coding scheme; the joint coding parameter optimization module receives semantic feedback information and adjusts the source coding rate and channel coding rate. This modular design makes the program easy to maintain, allowing different modules to be updated independently, adapting to the iterative needs of semantic awareness algorithms.

[0084] When the processor executes the computer program, it first reads the code of the semantic feature extraction module from the storage medium and loads it into memory. The processor then performs semantic analysis on the input data to be transmitted, generating semantic features. Next, it calls the source coding mode selection module to determine the source coding mode based on the semantic features. Then, combining this with channel state information, it calls the channel coding adaptive switching module to select the channel coding method. Finally, the processor transmits the encoded data to the receiving end through the communication interface, receives feedback information, and calls the joint coding parameter optimization module to adjust the coding parameters. The program's execution flow corresponds perfectly to the logic of the semantic-aware joint source-channel coding method, ensuring that each step of the method can be implemented through the program.

[0085] The advantages of this embodiment are that through the adaptation design of computer-readable storage medium and semantic-aware joint source-channel coding method, efficient storage and execution of program are achieved. The modular program structure enables the functional modules to work together, high-speed storage technology supports real-time coding requirements, cross-platform compatibility improves the versatility of the device, and ensures that the semantic-driven coding process can run stably in different hardware environments.

[0086] Although the present invention has been specifically described above with reference to preferred embodiments, it should be understood that the present invention is not limited to the embodiments described above. Various modifications and variations can be made by those skilled in the art without departing from the spirit of the present invention, and such modifications and variations should fall within the scope defined by the appended claims and their equivalents.

Claims

1. A semantically aware joint source-channel coding method, characterized in that the steps include... include: Obtain the semantic features of the data to be transmitted, which are used to characterize the semantic importance of the data to be transmitted; The source coding mode is selected based on the semantic features, and the source coding mode includes Polar code source coding or LDPC code source coding; Combining the semantic features and the output of the source coding mode, the channel coding method is adaptively switched, including Polar code channel coding or LDPC code channel coding; Based on the feedback information of the semantic features, the joint coding parameters of the source coding mode and the channel coding method are adjusted to optimize coding performance.

2. The semantically aware joint source-channel coding method as described in claim 1, characterized in that, The data to be transmitted is encoded using a Transformer-based semantic encoder, and the semantic entropy used to characterize the semantic uncertainty of the data to be transmitted or the semantic priority used to characterize the semantic importance of the data to be transmitted is extracted as the semantic feature.

3. The semantically aware joint source-channel coding method as described in claim 1, characterized in that, When the semantic importance corresponding to the semantic feature is higher than the first threshold, Polar code source coding is selected to improve the reliability of source coding; when the semantic importance corresponding to the semantic feature is lower than or equal to the first threshold, LDPC code source coding is selected to improve the efficiency of source coding.

4. The semantically aware joint source-channel coding method as described in claim 1, characterized in that, The adaptive switching channel coding method includes: when the semantic priority corresponding to the semantic feature is higher than the second threshold and the channel state is fading, Polar code channel coding is selected to combat channel interference; when the semantic priority corresponding to the semantic feature is lower than or equal to the second threshold and the channel state is Gaussian white noise, LDPC code channel coding is selected to reduce coding complexity.

5. The semantically aware joint source-channel coding method as described in claim 1, characterized in that, The adjustment of the joint coding parameters of the source coding mode and the channel coding method includes: receiving the semantic decoding result of the data to be transmitted, calculating the semantic loss, and iteratively adjusting the coding rate of the source coding mode and the code length of the channel coding method according to the semantic loss, wherein the semantic loss characterizes the degree of loss of semantic information of the data to be transmitted.

6. The semantically aware joint source-channel coding method as described in claim 3, characterized in that, The Polar code source encoding selection includes: adjusting the frozen bit positions of the Polar code according to the semantic features, and mapping data bits with high semantic importance to reliable bits of the Polar code to prioritize the protection of semantically important data; the LDPC code source encoding selection includes: selecting the parity check matrix structure of the LDPC code according to the semantic features, with data bits with high semantic importance corresponding to the height nodes of the LDPC code to enhance the error correction capability of semantically important data.

7. The semantically aware joint source-channel coding method as described in claim 1, characterized in that, The semantic features also include semantic relevance, which characterizes the semantic association between the data to be transmitted and the context data. The step of selecting the source coding mode based on the semantic features includes: when the semantic relevance is higher than a third threshold, selecting Polar code source coding to preserve the context semantic association; when the semantic relevance is lower than or equal to the third threshold, selecting LDPC code source coding to simplify the coding process.

8. The semantically aware joint source-channel coding method as described in claim 5, characterized in that, The adjustment methods for the joint coding parameters include: when the semantic loss increases, reducing the source coding rate to improve the source coding accuracy, while increasing the channel coding rate to reduce transmission delay; when the semantic loss decreases, increasing the source coding rate to improve coding efficiency, while reducing the channel coding rate to save channel resources.

9. The semantically aware joint source-channel coding method as described in claim 1, characterized in that, The data to be transmitted comes from multiple users; The acquisition of semantic features of the data to be transmitted includes: acquiring the semantic features of the data to be transmitted for each user; the selection of source coding mode based on the semantic features includes: independently selecting a source coding mode for the data to be transmitted for each user; the adaptive switching of channel coding mode includes: independently switching the channel coding mode for the data to be transmitted for each user; the adjustment of joint coding parameters includes: adjusting the joint coding parameters of multiple users based on feedback information of the semantic features of multiple users to achieve multi-user resource allocation optimization.

10. A semantically aware joint source-channel coding system, used to implement the semantically aware joint source-channel coding method as described in any one of claims 1-9, characterized in that, include: A semantic feature extraction module is used to obtain the semantic features of the data to be transmitted, wherein the semantic features are used to characterize the semantic importance of the data to be transmitted; The source coding mode selection module is used to select a source coding mode based on the semantic features, wherein the source coding mode includes Polar code source coding or LDPC code source coding. The channel coding adaptive switching module is used to adaptively switch the channel coding method by combining the semantic features and the output of the source coding mode. The channel coding method includes Polar code channel coding or LDPC code channel coding. The joint coding parameter optimization module is used to adjust the joint coding parameters of the source coding mode and the channel coding mode based on the feedback information of the semantic features, so as to optimize coding performance.

11. A semantically aware joint source-channel coding apparatus, characterized in that, include: The system includes a processor, a memory, and a communication interface. The memory stores a computer program, and the processor executes the computer program to implement the semantic-aware joint source-channel coding method as described in any one of claims 1 to 9. The communication interface is used to transmit encoded data or receive semantic feedback information.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the semantically aware joint source-channel coding method as described in any one of claims 1 to 9.

Images