A task adaptive semantic communication method applied to space-ground integrated network

Abstract

The application relates to a task adaptive semantic communication method applied to a space-ground integrated network, and the method comprises the following steps: constructing a multi-dimensional semantic service quality evaluation system to generate a comprehensive semantic service quality index for quantifying the service quality of semantic communication based on path transmission quality indexes and knowledge base compatibility; based on the comprehensive semantic service quality index, a first preference function of a user to a worker node and a second preference function of the worker node to the user are constructed; based on the first preference function and the second preference function, a stable matching algorithm with capacity constraints is run to realize dynamic service matching between the user and the worker node, and a stable matching set is obtained; based on the stable matching set, a UCB algorithm is used to select an optimal path from an available candidate path set for semantic communication, and a reconstructed semantic result is obtained. The method has the advantages of task adaptability, long-term stable matching and theoretical performance guarantee, can effectively support multiple businesses such as disaster rescue and remote sensing monitoring, and has good scalability.

Description

Technical Field

[0001] This invention relates to the fields of communication networks and artificial intelligence technology, specifically to a task-adaptive semantic communication method applied to an integrated space-ground network. Background Technology

[0002] With the large-scale deployment of low-Earth orbit satellite constellations (such as Starlink and OneWeb), integrated space-air-ground networks are gradually becoming an important infrastructure for global communication coverage. This network, through the collaboration of ground base stations, high-altitude platforms, and satellite systems, provides wide-area access capabilities to regions such as oceans, remote areas, and disaster sites. However, due to the limited bandwidth of satellite links, the time-varying nature of links, and frequent topology changes, traditional bit-based communication modes face bottlenecks in terms of efficiency and flexibility.

[0003] Semantic communication technology breaks away from the traditional approach of "bit-level accurate transmission," focusing instead on the "semantic value of information" and "task completion quality." Through semantic feature extraction and compression, it transmits only the critical information necessary to complete the task, thus offering significant advantages in bandwidth-constrained satellite links. Recent research indicates that in image, remote sensing, and video applications, deep learning-based semantic communication can substantially reduce data volume while maintaining high semantic accuracy.

[0004] However, the integrated air-space-ground network environment is highly dynamic: satellites move at high speeds, link quality fluctuates, user task requirements change over time, and the network simultaneously contains ground users, ground "worker nodes" (such as edge servers and ground stations), and multiple air-space-ground transmission paths. How to achieve task-oriented adaptive communication in a complex and dynamic environment, while taking into account both "semantic task requirements" and "physical link resource constraints," is a key issue of common concern in current engineering practice and academia.

[0005] Existing technologies have made progress in areas such as single semantic link optimization, semantic index design, and resource allocation strategies, but most of them are aimed at ground or low-altitude networks and mainly address static scenarios and local problems, making them difficult to apply directly to highly dynamic and heterogeneous integrated air-space-ground networks.

[0006] Therefore, it is necessary to propose a collaborative design that enables task-oriented adaptive communication and long-term stable resource matching in an integrated air-space-ground network, from semantic perception and task understanding to resource optimization. Summary of the Invention

[0007] To address the aforementioned problems in the existing technology, this invention provides a task-adaptive semantic communication method applied to integrated space-ground networks. This invention provides a task-adaptive semantic communication method applied to an integrated space-ground network, comprising: Step 1: Construct a multi-dimensional semantic service quality evaluation system to generate a comprehensive semantic service quality index for quantifying the quality of semantic communication services based on path transmission quality indicators and knowledge base compatibility. Step 2: Based on the comprehensive semantic service quality index, construct the first preference function of users for worker nodes and the second preference function of worker nodes for users; Step 3: Based on the first preference function and the second preference function, run the stable matching algorithm with capacity constraints to realize dynamic service matching between users and worker nodes, and obtain a stable matching set; Step 4: Based on the stable matching set, the UCB algorithm is used to select the optimal path from the available candidate path set for semantic communication to obtain the reconstructed semantic result.

[0008] In one embodiment of the present invention, step 1 includes: The semantic similarity of users on the path is obtained by fitting a function using channel CNR and semantic symbol number; Based on the minimum number of semantic symbols under the constraint of task semantic accuracy, the semantic rate of the user on the path is obtained. The path transmission quality index is constructed based on the semantic rate and the semantic similarity. Obtain the knowledge base compatibility between users and worker nodes, where the knowledge base compatibility is represented as:

[0009] in, For users With worker nodes The compatibility of knowledge bases between them For knowledge similarity function, For users semantic knowledge base For worker nodes The semantic knowledge base; The path transmission quality index and the knowledge base compatibility are weighted and fused to obtain the comprehensive semantic service quality index. In one embodiment of the present invention, the path transmission quality index is expressed as:

[0010] in, For users via path In the time slot Path transmission quality metrics for transmission tasks For users via path In the time slot Semantic similarity of transmission tasks For users via path In the time slot The semantic rate of the transmission task. For users To the worker nodes The One transmission path, For users In the time slot The minimum semantic similarity, For users in time slots The lowest rate, , This is a function in the Sigmoid form. In one embodiment of the present invention, the comprehensive semantic service quality index includes:

[0011] in, For users via path In the time slot The overall semantic service quality index after transmission These are the weighting coefficients. In one embodiment of the present invention, the first preference function is expressed as:

[0012] in, For users In the time slot To worker nodes Preference for transmission tasks user With worker nodes Historical average comprehensive semantic service quality index of interactions To explore coefficients, For users With worker nodes Number of selections, As weight, For indicator functions, user The previously selected worker node; The second preference function is expressed as:

[0013] in, For worker nodes For users In the time slot preference For users With worker nodes The compatibility of knowledge bases between them Based on historical average service quality, As weight.

[0014] In one embodiment of the present invention, step 3 includes: Step 3.1: Unmatched users submit service requests to worker nodes according to their first preference function and in preference order; Step 3.2: Provided that the worker node does not exceed its own service capacity limit, it accepts users with higher preferences and rejects other users according to the second preference function; Step 3.3: Iteratively execute steps 3.1 and 3.2 until there are no more blocking pairs, and obtain the stable matching set. In one embodiment of the present invention, step 4 includes: Step 4.1: Determine the average empirical reward value and the number of times each path in the set of available candidate paths is selected; Step 4.2: Calculate the upper confidence bound for each path in the set of available candidate paths based on the average reward value and the number of times it has been selected; Step 4.3: Select the path with the largest upper confidence threshold as the optimal path for semantic communication with the current user, and obtain the reconstructed semantic result. In one embodiment of the present invention, the upper confidence threshold is represented as:

[0015] in, For users via path In the time slot The upper confidence threshold for the transmission task. For users In the path Time slot Average experience reward For path By user Number of selections To explore coefficients. In one embodiment of the present invention, step 4.3 includes: The user's information to be transmitted is mapped into a semantic symbol vector through a semantic encoder and a channel encoder; The semantic symbol vector is transmitted through the optimal path to obtain the received signal; The received signal is recovered step by step by a channel decoder and a semantic decoder to obtain the reconstructed semantic result. In one embodiment of the present invention, the method further includes the following step after step 4.3: The actual SC-QoS reward for current semantic communication is calculated based on the formula for calculating the comprehensive semantic service quality index. And update the average reward value of the optimal path and the number of times it was selected. Compared with the prior art, the beneficial effects of the present invention are as follows: The purpose of this invention is to construct a semantic-driven system architecture and propose a multi-timescale dynamic matching and path optimization algorithm to address the characteristics of integrated air-space-ground networks, such as strong link dynamism, heterogeneous nodes, and diverse task requirements. First, at the system architecture level, a three-layer collaborative architecture of "semantic perception layer—task understanding layer—resource optimization layer" is designed. Data flow and control flow are organized around semantic elements and task requirements to achieve cross-modal semantic feature processing and adaptive resource allocation. Then, at the resource matching level, a user-worker node preference model is constructed using knowledge base similarity, historical service quality, and user willingness to pay. Stable matching theory is used to achieve stable dynamic user-worker node matching under capacity constraints, ensuring long-term stability. Finally, at the path selection level, the "user-worker node-path" selection problem is transformed into a multi-armed slot machine problem. A multi-timescale MTUCB algorithm is designed to ensure matching stability while utilizing the UCB mechanism to complete path exploration—using trade-offs to improve semantic communication quality and obtain a theoretically guaranteed upper bound on regret.

[0016] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a task-adaptive semantic communication method for an integrated space-ground network provided by the present invention. Detailed Implementation

[0018] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0019] Example 1 Existing technical solutions: (1) Research on semantic communication links in satellite scenarios Huang et al. systematically studied the satellite-to-ground semantic communication link structure in the paper “HUANG JH, SUN MY, HAN SJ, et al. Prospects of semantic communication technology for 6G satellite communication [J]. ZTEtechnology journal, 2024, 30(5): 3-8.” They proposed a semantic feature extraction method based on deep learning, which increased the effective information transmission volume by about 3–5 times under the same bandwidth conditions, verifying the bandwidth utilization advantage of semantic communication in satellite links.

[0020] (2) Preliminary work on resource matching and optimization in semantic communication In broader semantic communication research, the problem of multidimensional resource matching and optimization has attracted attention. Wang et al., in their paper "Wang L, Wu W, Zhou F, et al. Adaptive resource allocation for semantic communication networks[J]. IEEE Transactions on Communications, 2024, 72(11):6900-6916," proposed an adaptive semantic resource allocation paradigm, using a hybrid quantization method to establish a mapping between semantic indicators and traditional communication indicators, providing a quantitative basis for resource allocation in semantic scenarios. Xia et al., in their paper "Xia, Le, Yao Sun, Dusist Niyato, Xiaoqian Li and Muhammad Ali Imran. 'Joint UserAssociation and Bandwidth Allocation in Semantic Communication Networks.' IEEE Transactions on Vehicular Technology 73 (2022): 2699-2711," introduced an auxiliary knowledge base system, which, by sharing background knowledge in semantic communication, increased the system's message throughput by more than 40%.

[0021] (3) Semantic communication architecture and task allocation for air-space / ground-ground / non-terrestrial networks In the paper "H. Peng, Z. Zhang, Y. Liu, Z. Su, TH Luan and N. Cheng, "Semantic Communication in Non-Terrestrial Networks: A Future-Ready Paradigm," in IEEE Network, vol. 38, no. 4, pp. 119-127, July 2024," Peng et al. proposed a semantic communication-based network architecture for non-terrestrial networks (NTNs). This architecture considers semantic transmission under various link types and provides an architectural approach for semantic communication in satellite scenarios. However, it has not yet established a dynamic mapping relationship between task requirements and semantic features.

[0022] In summary, existing technologies have made progress in areas such as single semantic link optimization, semantic index design, and resource allocation strategies. However, most of these technologies are aimed at ground or low-altitude networks and are mainly designed for static scenarios and localized problems. They are difficult to apply directly to highly dynamic and heterogeneous integrated air-space-ground networks.

[0023] Analysis of the above studies reveals a progressive series of challenges in integrated air-space-ground semantic communication, ranging from flexibility and integrity to stability. Firstly, at the task adaptation level, the system is constrained by fixed encoding strategies and lacks dynamic adaptive mechanisms for specific tasks, making it impossible to flexibly adjust the semantic extraction granularity according to specific scenarios (such as disaster relief and remote sensing monitoring). Secondly, at the resource scheduling level, the system lacks collaborative optimization of multi-dimensional resources, often separating computational and communication resources, making it difficult to uniformly coordinate indicators such as semantic accuracy, physical link quality, and task completion, thus hindering efficient scheduling of integrated air-space-ground systems. Finally, this limitation extends to the long-term operational level. Due to the lack of stability assurance mechanisms, algorithms are often limited to pursuing instantaneous optimal performance, leading to frequent node switching and high signaling overhead in dynamic topologies, severely damaging service continuity and user experience.

[0024] Therefore, it is necessary to propose a new system architecture and communication method that can achieve task-oriented adaptive communication and long-term stable resource matching in an integrated air-space-ground network, from semantic perception and task understanding to resource optimization through full-link collaborative design.

[0025] This invention provides a task-adaptive semantic communication method applied to an integrated space-air-ground network. To implement this method, the integrated space-air-ground semantic communication architecture is first described. The integrated space-air-ground network includes ground base stations, low-Earth orbit satellites, ground worker nodes, ground users, and a central controller. Ground worker nodes possess semantic encoders and semantic knowledge bases for semantic extraction and compression of task data. User-side nodes possess semantic decoders and semantic knowledge bases for decoding received semantic representations and task recovery. The semantic knowledge base supports semantic compression and reconstruction and may include vectors of semantic features and keywords for text tasks; different nodes may vary depending on their needs. The central controller periodically broadcasts available worker nodes, path sets, and network topology information to users via a low-bandwidth control channel. The system operates in discrete time slots; within each time slot, the topology and channel conditions can be considered static, but may change across time slots.

[0026] Please see Figure 1 , Figure 1 This is a flowchart illustrating a task-adaptive semantic communication method for integrated space-ground networks provided by the present invention. The task-adaptive semantic communication method for integrated space-ground networks provided by the present invention includes: Step 1: Construct a multidimensional semantic service quality (SC-QoS) evaluation system to generate a comprehensive semantic service quality index for quantifying semantic communication service quality based on path transmission quality indicators and knowledge base compatibility.

[0027] Step 1.1: Obtain the semantic similarity of users on the path by fitting the channel CNR (Carrier-to-Noise Ratio) and the number of semantic symbols.

[0028] Specifically, semantic similarity is defined based on task type (such as different modalities like text, image, and speech) and channel conditions. , User In the path The semantic similarity is empirically fitted to the channel CNR and the number of semantic symbols. (in path) The number of semantic symbols transmitted is a function of the channel CNR (Cross-Noise Ratio) and the number of semantic symbols, which can be expressed as a logical function. Semantic similarity is obtained through fitting to achieve a mapping from network metrics to semantic quality.

[0029] Step 1.2: Based on the minimum number of semantic symbols under the task semantic accuracy constraint, obtain the semantic rate of the user on the path.

[0030] Specifically, based on the method proposed in the literature "Yan L, Qin Z, Zhang R, et al. Resource allocation for text semantic communications[J]. IEEE Wireless Communications Letters, 2022,11(7): 1394-1398.", which is based on the idea of ​​semantic entropy, the minimum number of semantic symbols that satisfy the semantic accuracy constraint of the task is defined, and the semantic rate, that is, the amount of effective semantic information per unit time, is obtained accordingly.

[0031] Step 1.3: Construct path transmission quality indicators based on semantic rate and semantic similarity.

[0032] Specifically, for a given user and worker nodes The Path Based on the gap between semantic rate and semantic similarity and task requirements, a path transmission quality index is constructed, which is expressed as:

[0033] in, For users via path In the time slot Path transmission quality metrics for transmission tasks For users via path In the time slot Semantic similarity of transmission tasks For users via path In the time slot The semantic rate of the transmission task. For users To the worker nodes The One transmission path, For users In the time slot The minimum semantic similarity, For users in time slots The lowest rate, and Determined based on the actual situation. , This is a function in the Sigmoid form.

[0034] Step 1.4: Obtain the knowledge base compatibility between users and worker nodes.

[0035] Specifically, to characterize the consistency of the semantic knowledge bases of users and worker nodes, the knowledge base compatibility is defined as follows:

[0036] in, For users With worker nodes The compatibility of knowledge bases between them For knowledge similarity function, For users semantic knowledge base For worker nodes A semantic knowledge base.

[0037] Step 1.5: Weighted fusion of path transmission quality indicators and knowledge base compatibility to obtain comprehensive semantic service quality indicators.

[0038] Here, the comprehensive semantic service quality index is expressed as:

[0039] in, For users via path In the time slot The overall semantic service quality index after transmission These are the weighting coefficients.

[0040] Step 2: Based on the comprehensive semantic service quality index, construct the first preference function of users to worker nodes and the second preference function of worker nodes to users.

[0041] In this embodiment, in each time slot The central controller collects the set of available worker nodes. and the capacity limit for each worker node. (The maximum number of users that each worker node can serve); Collect user sets. and each user's task requirements (such as semantic rate and semantic similarity requirements) and willingness to pay parameters. Construct a knowledge base compatibility matrix .

[0042] Here, the first preference function is expressed as:

[0043] in, For users In the time slot To worker nodes Preference for transmission tasks user With worker nodes The historical average comprehensive semantic service quality index of the interaction, the second item is the UCB (Upper Confidence Bound) exploration item, To explore coefficients, For users With worker nodes Number of selections, As weight, This is an indicator function for whether a worker handover has occurred. The function takes a value of 1 if the current candidate worker is different from the worker matched with the user in the previous time slot, and 0 otherwise. For users The worker node selected previously.

[0044] Here, the second preference function is expressed as:

[0045] in, For worker nodes For users In the time slot preference For users With worker nodes The compatibility of knowledge bases between them Based on historical average service quality, As weight.

[0046] Step 3: Based on the first preference function and the second preference function, run the stable matching algorithm with capacity constraints to realize dynamic service matching between users and worker nodes, and obtain a stable matching set.

[0047] Step 3.1: Unmatched users submit service requests to worker nodes in order of preference based on their first preference function.

[0048] Specifically, firstly, the user's preference for each worker node is calculated using the first preference function. Then, these preferences are sorted from highest to lowest, and a corresponding user preference list is built according to the sorted order. This user preference list The system records the order of each user's preferences for different worker nodes. Therefore, unmatched users will be processed according to their own user preference list. Service requests are sent to worker nodes in descending order of priority, such as user requests. u 1. First, go to your preferred worker node. w 2. Send a service request if the worker node If the user's capacity is not full, the request will be accepted directly; if the user's capacity is full, further comparisons will be made. Among the currently matched users, the one least favored by this worker. If right If the user's preference is higher, replace that user; otherwise, replace the user. Next to the next worker node Apply.

[0049] Step 3.2: Provided that the worker node does not exceed its own service capacity limit, it accepts users with higher preferences and rejects other users according to the second preference function.

[0050] Specifically, the second preference function is first used to calculate the worker nodes' preferences for users, then the preferences are sorted from high to low, and a corresponding list of worker node preferences is built according to the sort order. The worker node preference list The system records the preference order of each worker node for different users. Therefore, if a worker node has not yet reached its service capacity limit, when it receives a user request, it will compare the requesting user with currently paired users (if any) in its worker node preference list. The position of the newly applying user in the worker node preference list. If a new user is ranked higher than some of the already paired users (better), and the worker node has remaining capacity, then the worker node will accept the new user's application and reject the relatively worse already paired users. If the new user's application is worse than the already paired users, or if the worker node's capacity is full, then the worker node will reject the application.

[0051] Step 3.3: Iteratively execute steps 3.1 and 3.2 until there are no more blocking pairs, and obtain a stable matching set.

[0052] Specifically, the blocking pair is defined as the existence of a pair of users. and worker nodes ,user For worker nodes The worker node's preference is higher than its current matched objects, and the worker node For users Their preference is also higher than their current matched partner, at this time this pair ( The blocking pairs are (wj) and (wj) are the blocking pairs. By iterating through steps 3.1 and 3.2, all blocking pairs are gradually eliminated. When no blocking pairs exist, the matching relationship reaches stability—all user and worker nodes are satisfied with their matching. The set of matching results at this point is denoted as the stable matching set Ψ, which records several pairs of mutually matching user and worker node pairs.

[0053] Step 4: Based on the stable matching set, the UCB algorithm is used to select the optimal path from the available candidate path set for semantic communication, and the reconstructed semantic result is obtained.

[0054] Specifically, the UCB algorithm is used to select the optimal path for each pair of user and worker nodes in the stable matching set Ψ from the available candidate path set and perform semantic communication to obtain the reconstructed semantic result.

[0055] Step 4.1: Determine the average empirical reward value and the number of times each path in the set of available candidate paths is selected.

[0056] Specifically, the set of all currently available candidate paths is set as the available candidate path set, and the average experience reward value and the number of times each path in the available candidate path set is maintained. The average experience reward is obtained by adding up all the rewards that a path has received when it was selected in the past and dividing by the number of times it was selected. The number of times it was selected is the sum of the total number of times a path has been selected.

[0057] Step 4.2: Calculate the upper confidence bound for each path in the available candidate path set based on the average reward value and the number of times it has been selected.

[0058] Here, the upper confidence threshold is represented as:

[0059] in, For users via path In the time slot The upper confidence threshold for the transmission task. For users In the path Time slot Average experience reward For path By user Number of selections.

[0060] Step 4.3: Select the path with the largest upper confidence threshold as the optimal path for semantic communication with the current user to obtain the reconstructed semantic results.

[0061] Here, the path with the largest upper confidence threshold is:

[0062] in, For users With worker nodes The path with the largest upper confidence bound after matching. For users At worker nodes Time slot The set of candidate paths below.

[0063] Step 4.31: Map the user's information to be transmitted into a semantic symbol vector using a semantic encoder and a channel encoder.

[0064] Specifically, the user is encoded using a semantic encoder and a channel encoder. Information to be transmitted Mapped to semantic symbol vectors semantic symbol vector Represented as:

[0065] in, For semantic encoders, For channel encoder, This is the parameter set of the semantic encoder. This is the set of parameters for the channel encoder.

[0066] Step 4.32: Transmit the semantic symbol vector through the optimal path to obtain the received signal.

[0067] Specifically, the encoded semantic symbol vector is transmitted along the selected optimal path. The user at the receiving end Received signal Receive signal Represented as:

[0068] in, For complex channel gain, It is noise.

[0069] Step 4.33: The received signal is recovered step by step using the channel decoder and semantic decoder to obtain the reconstructed semantic result.

[0070] Specifically, on the user side, the reconstructed semantic result is obtained through step-by-step recovery using a channel decoder and a semantic decoder. The reconstructed semantic results Represented as:

[0071] in, For channel decoder, For semantic decoders, This is the parameter set for the channel decoder. This is the set of parameters for the semantic decoder.

[0072] Step 5: Calculate the actual SC-QoS reward for the current semantic communication based on the calculation formula of the comprehensive semantic service quality index. And update the average reward value of the optimal path and the number of times it was selected.

[0073] Specifically, following the method proposed in step 1, the calculation formula for the comprehensive semantic service quality index is used to calculate the actual SC-QoS reward for the current semantic communication. It also updates the average reward value of the optimal path and the number of times it was selected.

[0074] Step 6: Multi-timescale closed-loop update.

[0075] Specifically, at the micro-timescale of a single time slot, the path quality distribution is learned through UCB path selection to optimize short-term SC-QoS; at the macro-timescale of multiple time slots, updates are made based on cumulative rewards and handover costs. , By using statistical measures, the preference lists of users and worker nodes are updated, and stable matching is periodically re-executed to ensure long-term matching stability and system efficiency.

[0076] This multi-timescale mechanism constitutes the overall framework of the MTUCB algorithm, enabling collaborative optimization of user-worker node matching and path selection.

[0077] The effects of this invention can be further illustrated by the following simulation experiments.

[0078] 1. Simulation conditions This invention was simulated on a computer with a 13th Gen Intel(R) Core(TM) i9-13900HX 2.20 GHz processor and a 64-bit Windows 11 operating system. The algorithm was written in Python 3.8, numerical calculations were performed using the NumPy 1.21.0 library, and results visualization was performed using the Matplotlib 3.4.2 and Seaborn 0.11.1 libraries. The code was edited and debugged in the PyCharm integrated development environment. The orbit and link scenarios of the integrated air-space-ground network were modeled using STK 11 to obtain the geometric relationships and path loss ranges of typical air-space-ground links. In the simulation program, the path quality was normalized to a random time-varying parameter of [0.4, 0.9] to reflect the fluctuation of link quality in a real integrated air-space-ground network.

[0079] In a basic simulation without considering network scaling, 15 user nodes and 10 worker nodes are set up. Each worker node provides 5 candidate transmission paths for each user, and the number of simulation time slots is 1000 to ensure sufficient algorithm convergence. Key algorithm parameters are set as follows: path quality weight coefficient α = 0.3, used to balance the impact of path physical quality and semantic compatibility; UCB exploration parameter ζ = 0.2, used to control the exploration intensity; switching cost weight ω = 0.1, used to suppress frequent switching of service nodes; the maximum number of users served by each worker node is limited to 2 to simulate the computational and link resource constraints in a real system. The elements of the knowledge base fit matrix are randomly initialized within the interval [0.5, 1.0], and the user's willingness to pay is uniformly distributed within the interval [0.3, 1.0].

[0080] To verify the effectiveness of the method of this invention, the following comparison algorithm was selected in the experiment: (1) UCB algorithm: The UCB multi-armed bandit problem proposed by Auer et al. in “Finite-time Analysis of the Multi-armed Bandit Problem”, Machine Learning, 47(2–3): 235–256, 2002 is adopted. UCB decision is only performed at the level of “user-worker node-path” triplet, without considering stable matching and switching costs.

[0081] (2) ε-Greedy algorithm: In each time slot, only the "user-worker node-path" combination with the highest estimated average service quality is selected without any exploration, and it is used as the baseline for pursuing only the instantaneous optimum.

[0082] (3) Stable matching algorithm: The stable matching algorithm proposed by Gale and Shapley in "College Admissions and the Stability of Marriage", American Mathematical Monthly, 69(1): 9–15, 1962 is adopted to match users and worker nodes. However, in order to simplify the comparison, each user uses a preset path after the matching is completed, and no further path-level learning and optimization are performed.

[0083] (4) Random algorithm: Randomly select worker nodes and transmission paths under the premise of satisfying worker node capacity constraints, as a reference for performance lower bound.

[0084] (5) Theoretical optimal algorithm: Assuming that the complete state information of all time slots is known in advance, the optimal "user-worker node-path" combination for each time slot is obtained by offline exhaustive search or global optimization, which is used as the upper bound of performance. It is only used for comparison and is not feasible to implement online.

[0085] 2. Simulation Content According to a specific embodiment of the present invention, under the above simulation conditions, the MTUCB algorithm proposed in this invention, as well as the UCB algorithm, ε-Greedy algorithm, stable matching algorithm, random algorithm, and theoretically optimal algorithm, are used to statistically compare the average service quality index and the overall system benefit index of the system within 1000 time slots. The average service quality comprehensively reflects semantic similarity, task completion, and link quality scores, while the overall system benefit reflects the comprehensive revenue score of providing semantic services to all users under resource constraints. The comparison results are summarized in Table 1.

[0086] Table 1 Algorithm Performance Comparison

[0087] As shown in Table 1, under the same network size and parameter settings, the MTUCB algorithm proposed in this invention significantly outperforms the comparative methods such as pure UCB, greedy matching, stable matching only, and random matching in both average service quality and overall system efficiency. Its performance is only slightly lower than the theoretically optimal algorithm, and it has a fast convergence speed and small performance curve fluctuations. This indicates that while ensuring matching stability, this invention can effectively balance exploration and utilization, fully explore high-quality transmission paths, and thus significantly improve the overall service quality of the integrated air-space-ground semantic communication network, verifying the advanced nature and effectiveness of the technical solution of this invention.

[0088] This invention proposes a semantic communication framework for integrated air-space-ground systems. Through a three-layer coupled architecture of "semantics-task-resources," combined with dynamic adaptive strategies and multi-timescale matching algorithms, it achieves a unified approach to semantic optimization and resource allocation. This scheme features task adaptability, long-term stable matching, and theoretical performance guarantees, effectively supporting various services such as disaster relief and remote sensing monitoring, and possesses good scalability.

[0089] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0090] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings, disclosure, and appended claims in carrying out the claimed invention. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0091] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, any modifications made without departing from the inventive concept should be considered within the scope of protection of the present invention.

Claims

1. A task-adaptive semantic communication method applied to an integrated space-ground network, characterized in that, include: Step 1: Construct a multi-dimensional semantic service quality evaluation system to generate a comprehensive semantic service quality index for quantifying the quality of semantic communication services based on path transmission quality indicators and knowledge base compatibility. Step 2: Based on the comprehensive semantic service quality index, construct the first preference function of users for worker nodes and the second preference function of worker nodes for users; Step 3: Based on the first preference function and the second preference function, run the stable matching algorithm with capacity constraints to realize dynamic service matching between users and worker nodes, and obtain a stable matching set; Step 4: Based on the stable matching set, the UCB algorithm is used to select the optimal path from the available candidate path set for semantic communication to obtain the reconstructed semantic result.

2. The task-adaptive semantic communication method according to claim 1, characterized in that, Step 1 includes: The semantic similarity of users on the path is obtained by fitting a function using channel CNR and semantic symbol number; Based on the minimum number of semantic symbols under the constraint of task semantic accuracy, the semantic rate of the user on the path is obtained. The path transmission quality index is constructed based on the semantic rate and the semantic similarity. Obtain the knowledge base compatibility between users and worker nodes, where the knowledge base compatibility is represented as: in, For users With worker nodes The compatibility of knowledge bases between them For knowledge similarity function, For users semantic knowledge base For worker nodes The semantic knowledge base; The path transmission quality index and the knowledge base compatibility are weighted and fused to obtain the comprehensive semantic service quality index.

3. The task-adaptive semantic communication method according to claim 2, characterized in that, The path transmission quality index is expressed as: in, For users via path In the time slot Path transmission quality metrics for transmission tasks For users via path In the time slot Semantic similarity of transmission tasks For users via path In the time slot The semantic rate of the transmission task. For users To the worker nodes The One transmission path, For users In the time slot The minimum semantic similarity, For users in time slots The lowest rate, , This is a function in the Sigmoid form.

4. The task-adaptive semantic communication method according to claim 3, characterized in that, The comprehensive semantic service quality indicators include: in, For users via path In the time slot The overall semantic service quality index after transmission These are the weighting coefficients.

5. The task-adaptive semantic communication method according to claim 1, characterized in that, The first preference function is expressed as: in, For users In the time slot To worker nodes Preference for transmission tasks user With worker nodes Historical average comprehensive semantic service quality index of interactions To explore the coefficient, For users With worker nodes Number of selections, As weight, For indicator functions, user The previously selected worker node; The second preference function is expressed as: in, For worker nodes For users In the time slot preference For users With worker nodes The compatibility of knowledge bases between them Based on historical average service quality, As weight.

6. The task-adaptive semantic communication method according to claim 1, characterized in that, Step 3 includes: Step 3.1: Unmatched users submit service requests to worker nodes according to their first preference function and in preference order; Step 3.2: Provided that the worker node does not exceed its own service capacity limit, it accepts users with higher preferences and rejects other users according to the second preference function; Step 3.3: Iteratively execute steps 3.1 and 3.2 until there are no more blocking pairs, and obtain the stable matching set.

7. The task-adaptive semantic communication method according to claim 1, characterized in that, Step 4 includes: Step 4.1: Determine the average empirical reward value and the number of times each path in the set of available candidate paths is selected; Step 4.2: Calculate the upper confidence bound for each path in the set of available candidate paths based on the average reward value and the number of times it has been selected; Step 4.3: Select the path with the largest upper confidence threshold as the optimal path for semantic communication with the current user, and obtain the reconstructed semantic result.

8. The task-adaptive semantic communication method according to claim 7, characterized in that, The upper confidence threshold is represented as: in, For users via path In the time slot The upper confidence threshold for the transmission task. For users In the path Time slot Average experience reward For path By user Number of selections To explore coefficients.

9. The task-adaptive semantic communication method according to claim 7, characterized in that, Step 4.3 includes: The user's information to be transmitted is mapped into a semantic symbol vector through a semantic encoder and a channel encoder; The semantic symbol vector is transmitted through the optimal path to obtain the received signal; The received signal is recovered step by step by a channel decoder and a semantic decoder to obtain the reconstructed semantic result.

10. The task-adaptive semantic communication method according to claim 7, characterized in that, Following step 4.3, the following is also included: The actual SC-QoS reward for current semantic communication is calculated based on the formula for calculating the comprehensive semantic service quality index. And update the average reward value of the optimal path and the number of times it was selected.

Images