Knowledge base synchronization method and system for space-based semantic communication, and computer readable medium

Abstract

The application relates to a knowledge base synchronization method, system and computer readable medium for space-based semantic communication, the knowledge base synchronization method comprising: initializing a general semantic knowledge base; in a first communication stage, training the general semantic knowledge base according to a first quantity of communication data to obtain a first semantic knowledge base, and transmitting the first semantic knowledge base to a sending end node and a receiving end node; judging whether the semantic knowledge base of the receiving end node needs to be updated, if the semantic knowledge base of the receiving end node needs to be updated, updating the semantic knowledge base of the receiving end node according to the first data transmitted by the sending end node to obtain a synchronized first communication semantic knowledge base; and the sending end node and the receiving end node communicating according to the synchronized first communication semantic knowledge base. The application can reduce the semantic knowledge base updating frequency of both communication parties in a satellite communication system, and save the time and bandwidth resources required for semantic knowledge base synchronization.

Description

Technical Field

[0001] This invention relates primarily to the field of satellite communications, and more specifically to a knowledge base synchronization method, system, and computer-readable medium for space-based semantic communications. Background Technology

[0002] Satellite mobile communication systems have become an important part of modern communication methods due to their wide coverage and long communication distance. However, compared with terrestrial communication, satellite communication has lower bandwidth and longer latency. With the increasing volume of communication data, satellite communication urgently needs a new communication paradigm to improve its transmission capacity. Since semantic communication can significantly reduce transmission rate requirements, integrating semantic communication into satellite communication has become a new research hotspot. The core of semantic communication is a semantic knowledge base; both the sending and receiving users in semantic communication need to possess the same or similar semantic knowledge bases.

[0003] In existing semantic communication methods, the definition of semantics is limited by a predefined tag database, resulting in a single and fixed semantic knowledge base. When encountering unknown knowledge that cannot be encoded or decoded, communication between the two parties becomes impossible. To address this issue, patent document CN114490920A proposes an automatic semantic knowledge base construction and update method. This method uses an automatic model learning and updating mechanism to simultaneously and collaboratively update the semantic knowledge base using a semantic codec, achieving highly dynamic semantic communication. However, this method requires the receiver to frequently transmit evaluation scores to the sender via a semantic evaluator during the simultaneous collaborative updating of the semantic codec. The sender then updates the model based on these scores. This process necessitates frequent communication between the two parties, which is impractical for satellite communication. Satellites typically move at high speeds relative to the ground in their orbits. For a ground-based communication terminal, the visibility time of a satellite is short. If an update to the semantic knowledge base is detected within the satellite's visible range, a significant amount of time will be spent on model updates, contradicting the purpose of applying semantic communication to space-based communication scenarios. In addition, the method proposed in patent document CN114490920A requires the updated model to be sent to the receiving end after the model is updated at the sending end until it converges in order to continue communication. This adds extra overhead to space-based semantic communication, and the receiving end loses its original knowledge base, increasing the probability that the knowledge base needs to be updated when communicating with other sending ends.

[0004] Existing methods for updating semantic knowledge bases include federated learning. Based on data sources, federated learning can be categorized into horizontal federated learning, vertical federated learning, and federated transfer learning. Horizontal federated learning is suitable for situations where data sample spaces are similar but feature spaces differ; vertical federated learning is suitable for situations where data sample spaces differ but feature spaces are similar; and federated transfer learning is suitable for situations where both data sample spaces and feature spaces are dissimilar. Based on network architecture, federated learning can be divided into centralized federated learning and decentralized federated learning. For federated learning scenarios involving multiple users, a centralized federated architecture is generally used, with each enterprise acting as a server to coordinate the global model. For scenarios involving multiple enterprises facing data silos for model training, a decentralized federated architecture is generally suitable because it is difficult to select a coordinating server from among multiple enterprises. Current research on federated learning mainly focuses on centralized federated architectures; however, because there are no nodes in a satellite constellation that can be permanently connected to all satellites, a centralized federated architecture cannot be used to update the semantic knowledge base.

[0005] Furthermore, due to the large number of satellites and ground terminals involved in communication due to their orbits around the Earth, each communication node needs to store multiple semantic knowledge bases to overcome semantic communication problems between nodes with different semantic knowledge backgrounds. The more nodes involved, the more semantic knowledge bases need to be stored. However, satellite storage capacity is limited, making it impossible to store a large number of semantic knowledge bases. Therefore, in current space-based semantic communication, both parties face numerous problems, including high frequency of semantic knowledge base updates, high communication overhead, frequent exchange of semantic knowledge bases, long update times, and low universality of semantic knowledge bases. Summary of the Invention

[0006] The technical problem to be solved by this application is to provide a method, system and computer-readable medium for synchronizing the knowledge base of space-based semantic communication, which can reduce the frequency of updating the semantic knowledge base of the two communicating parties in the satellite communication system and save the time and bandwidth resources required for semantic knowledge base synchronization.

[0007] The technical solution adopted in this application to solve the above-mentioned technical problems is a knowledge base synchronization method for space-based semantic communication, including: initializing a general semantic knowledge base; in the first communication stage, training the general semantic knowledge base according to a first amount of communication data to obtain a first semantic knowledge base, and transmitting the first semantic knowledge base to the sending node and the receiving node; determining whether it is necessary to update the semantic knowledge base of the receiving node itself, and if it is necessary to update, updating the semantic knowledge base of the receiving node itself according to the first data transmitted by the sending node to obtain the synchronized first communication semantic knowledge base; and the sending node and the receiving node communicating according to the synchronized first communication semantic knowledge base.

[0008] In one embodiment of this application, after the step of the sending node and the receiving node communicating based on the synchronized first communication semantic knowledge base, the method further includes: determining whether the amount of communication data reaches a preset threshold; if the preset threshold is reached, in the second communication stage, training a general semantic knowledge base based on a second amount of communication data to obtain a second semantic knowledge base, and transmitting the second semantic knowledge base to the sending node and the receiving node, wherein the second amount is greater than the first amount; the sending node updates its own semantic knowledge base according to a federated learning method to obtain a federated semantic knowledge base; transmitting the federated semantic knowledge base to the receiving node; and the sending node and the receiving node communicating based on the federated semantic knowledge base.

[0009] In one embodiment of this application, the ground training terminal continuously collects communication data during the communication process between the sending node and the receiving node. The ground training terminal trains a general semantic knowledge base. The ground training terminal controls the size of the general semantic knowledge base through a mask. During the training of the general semantic knowledge base from the first communication stage to the second communication stage, the proportion of 0 in the mask is gradually reduced to gradually increase the general semantic knowledge base.

[0010] In one embodiment of this application, the step of determining whether the semantic knowledge base of the receiving node needs to be updated includes: the sending node transmitting k pieces of first data to the receiving node, where k is a positive integer greater than or equal to 1, and the first data includes source data and the semantic representation of the source data after semantic encoding; decoding the semantic knowledge base of the receiving node according to the semantic representation, obtaining k decoded data for each semantic knowledge base of the receiving node, and calculating the average similarity score (average_similarity) between the k decoded data and the source data for the i-th knowledge base version. i The maximum score of the average similarity score is calculated using the following formula:

[0011] score=max(average_s imilarity i (i∈[1,m])

[0012] Where m represents the number of versions of the semantic knowledge base of the receiving node itself;

[0013] The version number (index) of the semantic knowledge base corresponding to the maximum average similarity score is calculated using the following formula:

[0014] index=argmax(average_s imilarity i (i∈[1,m])

[0015] Determine whether the maximum value of the average similarity score has reached the preset similarity threshold. If it has not reached the preset similarity threshold, the semantic knowledge base of the receiving node itself needs to be updated; if it has reached the preset similarity threshold, communication is carried out based on the semantic knowledge base of the version number index.

[0016] In one embodiment of this application, the step of updating the semantic knowledge base of the receiving node based on the first data transmitted by the sending node includes: the receiving node performing transfer training on the semantic knowledge base with version number index using k pieces of the first data.

[0017] In one embodiment of this application, the step of determining whether the semantic knowledge base of the receiving node needs to be updated includes: calculating the update magnitude α of the sending node. T And calculate the update magnitude α of the receiving node. R Determine the update magnitude α of the sending node respectively. T and the update magnitude α of the receiving node R Is it greater than the preset update threshold? If α T and α R All values ​​are not greater than the preset update threshold, and communication is based on the existing semantic knowledge base; if α T and / or α R If the value exceeds the preset update threshold, the semantic knowledge base of the receiving node needs to be updated; the sending node transmits k pieces of first data to the receiving node, where k is a positive integer greater than or equal to 1. The first data includes the source data and the semantically encoded representation of the source data.

[0018] In one embodiment of this application, the update magnitude α of the sending node is calculated. T The steps include: encoding the data to be sent using a copy of the semantic knowledge base from the first communication phase to obtain a semantic representation s of the data to be sent. old Input semantic representation s old The source data and its corresponding data are fed into the virtual decoder, and the average value of the output of the softmax layer in the virtual decoder is taken as the semantic representation s. old The update magnitude α of the sending node is calculated using the following formula, based on the similarity score (similarity_T) of the corresponding source data. T :

[0019]

[0020] Where x is a hyperparameter;

[0021] And, calculate the update magnitude α of the receiving node. R The steps include: randomly storing the received semantic representation c every preset time interval during the communication process. tConstruct test set C = {c t , t=0,...,j}, where, T_total represents the duration of a single communication, where T is the preset duration; a semantic representation c is randomly selected from the test set C. t Decode the semantic representation c using a dynamically updated semantic knowledge base. t Obtain semantic representation c t The source data c new Decode the semantic representation c using a copy of the semantic knowledge base from the first communication phase. t Obtain semantic representation c t The source data c old ; Calculate source data c new and source data c old The similarity score, similarity_R, is used to calculate the update magnitude α of the receiving node using the following formula. R :

[0022]

[0023] In one embodiment of this application, the step of updating the semantic knowledge base of the receiving node based on the first data transmitted by the sending node includes: calculating the training learning rate α using the following formula:

[0024] α = max(α) T α R )

[0025] Where, α T α represents the update magnitude of the sending node. R Indicates the update magnitude α of the receiving node. R The receiving node takes k source data as labels and k semantic representations as input, and performs transfer training on the semantic knowledge base according to the training learning rate α.

[0026] In one embodiment of this application, the step of training a general semantic knowledge base based on a first amount of communication data includes: constructing an end-to-end training framework including a transmitter encoder and a receiver decoder, wherein the end-to-end training framework uses the Transformer mechanism; inputting training source data s to the transmitter encoder to obtain the semantic representation corresponding to the training source data s; and decoding the semantic representation corresponding to the training source data s by the receiver decoder and outputting the decoded training source data. And obtain the trained end-to-end training framework; input the first amount of communication data into the trained end-to-end training framework, and use the cross-entropy of the initial source data and the decoded source data as the Loss function to update the semantic knowledge base parameters used by the transmitter encoder and receiver decoder respectively.

[0027] In one embodiment of this application, the data structure of the second quantity of communication data includes: a start symbol, a communication party name, a separator, data, and an end symbol, wherein the communication party name includes the name of the sending node and the name of the receiving node, and the data is the source data without semantic encoding.

[0028] In one embodiment of this application, the step of updating its own semantic knowledge base according to the federated learning method includes: all sending nodes exchanging their respective semantic knowledge bases with each other until each sending node obtains the semantic knowledge bases of other sending nodes after y iterations, where y is a positive integer greater than or equal to 1; each sending node merging all its own semantic knowledge bases using an aggregation algorithm.

[0029] In one embodiment of this application, the federated semantic knowledge base of the sending node i needs to be updated for t rounds, where i and t are positive integers greater than or equal to 1. During the t-th round of federated semantic knowledge base update, the sending node i obtains the model parameters of the semantic knowledge base at time t+1. The steps include: Step S1: The sending node i iterates the model parameters of its own semantic knowledge base at time t. This includes: constructing an end-to-end semantic knowledge base self-updating framework comprising virtual senders and virtual receivers; calculating model parameters using the following formula after y iterations of the end-to-end semantic knowledge base self-updating framework.

[0030]

[0031] Where η represents the update rate, The gradient operator is represented by l(·), the loss function is represented by l(·), and the training dataset is represented by b; Step S2: The sending node i adds its own ID i to its own semantic knowledge base name list list. i And transmit a list of semantic knowledge base names to other sending nodes. i Step S3: Sending node i receives the list of semantic knowledge base names transmitted by other sending node j. j , j∈N i N i This represents the set of other sending nodes; and a list of semantic knowledge base names transmitted to sending node j. i The list of semantic knowledge base names is removed. j This includes all semantic knowledge base information other than the semantic knowledge base itself. The data structure of the semantic knowledge base information includes: knowledge base name l, and the amount of data b used by the semantic knowledge base in this iteration. l Semantic knowledge base parameters Step S4: When other sending node j switches or the semantic knowledge base name list is accessed...i When the content changes, repeat steps S2 to S3 until the list of semantic knowledge base names is reached. i All sending nodes are included; sending node i adds its own ID i to its own acknowledgment list list_ACK. i And transmit the acknowledgment list list_ACK to other sending node j. i ; Receive ACK information transmitted by other sending node j j And store it in the acknowledgment information list list_ACK i Step S5: When the confirmation message list list_ACK is received... i When acknowledgment information from all sending nodes is included, the following formula is used to aggregate all semantic knowledge bases to obtain the federated semantic knowledge base z:

[0032]

[0033] Where n represents the total number of semantic knowledge bases.

[0034] To address the aforementioned technical problems, this application also proposes a knowledge base synchronization system for space-based semantic communication, comprising: a memory for storing instructions executable by a processor; and a processor for executing the instructions to implement the aforementioned knowledge base synchronization method for space-based semantic communication.

[0035] To address the aforementioned technical problems, this application also proposes a computer-readable medium storing computer program code, which, when executed by a processor, implements the knowledge base synchronization method for space-based semantic communication as described above.

[0036] The technical solution of this application generates a custom-sized general semantic knowledge base through initialization. In the first communication phase, the general semantic knowledge base is trained based on a first amount of communication data to obtain a first semantic knowledge base, which is then transmitted to the sending node and the receiving node. The two communicating parties can communicate based on this semantic knowledge base. The update frequency is determined by whether the receiving node's own semantic knowledge base needs to be updated. If an update is required, the receiving node updates its own semantic knowledge base based on the first data transmitted by the sending node, and the two communicating parties communicate based on the synchronized first semantic knowledge base. If an update is not required, the two communicating parties communicate based on their respective existing semantic knowledge bases. This configuration, while ensuring communication, reduces the update frequency of the semantic knowledge bases of both communicating parties, reduces communication overhead, and saves the time and bandwidth resources required for semantic knowledge base synchronization. The receiving node updates its own semantic knowledge base, avoiding the need for the two communicating parties to exchange semantic knowledge bases. The update time of the semantic knowledge base is relatively low, allowing for longer data communication time within the communication period. Attached Figure Description

[0037] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the specific embodiments of this application will be described in detail below with reference to the accompanying drawings, wherein:

[0038] Figure 1 This is an exemplary flowchart of a knowledge base synchronization method for space-based semantic communication according to an embodiment of this application;

[0039] Figure 2 This is an exemplary flowchart of a knowledge base synchronization method for space-based semantic communication according to another embodiment of this application;

[0040] Figure 3 This is an exemplary architecture diagram of communication terminals using a semantic knowledge base according to an embodiment of this application;

[0041] Figure 4 This is an exemplary schematic diagram of an end-to-end training framework according to an embodiment of this application;

[0042] Figure 5 This is an exemplary schematic diagram of a self-updating framework for the semantic knowledge base of the sending node according to an embodiment of this application;

[0043] Figure 6 This is an exemplary communication flowchart of an embodiment of this application when the sending node and the receiving node do not need to synchronize the semantic knowledge base;

[0044] Figure 7 This is an exemplary communication flowchart of an embodiment of this application when the sending node and the receiving node need to synchronize the semantic knowledge base;

[0045] Figure 8 This is an exemplary flowchart of a knowledge base synchronization method for space-based semantic communication according to another embodiment of this application;

[0046] Figure 9 This is a system block diagram of a knowledge base synchronization system for space-based semantic communication according to an embodiment of this application. Detailed Implementation

[0047] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the specific embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0048] Many specific details are set forth in the following description in order to provide a full understanding of this application. However, this application may also be implemented in other ways different from those described herein, and therefore this application is not limited to the specific embodiments disclosed below.

[0049] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0050] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, various steps can be processed in reverse order or simultaneously. Furthermore, other operations may be added to these processes, or one or more steps may be removed from these processes.

[0051] This application proposes a knowledge base synchronization method for space-based semantic communication, applied to a satellite communication system based on semantic communication. This system includes a transmitting node and a receiving node, which can be a satellite or a ground station, etc. For example, the transmitting node can be a satellite, and the receiving node can be a satellite; or the transmitting node can be a satellite, and the receiving node can be a ground station, etc. This application does not impose any restrictions on the transmitting and receiving nodes.

[0052] Figure 1 This is an exemplary flowchart of a knowledge base synchronization method for space-based semantic communication according to an embodiment of this application.

[0053] refer to Figure 1 As shown, the knowledge base synchronization method for space-based semantic communication in this embodiment includes the following steps:

[0054] Step S110: Initialize the general semantic knowledge base.

[0055] Step S120: In the first communication phase, a general semantic knowledge base is trained based on a first amount of communication data to obtain a first semantic knowledge base, and the first semantic knowledge base is transmitted to the sending node and the receiving node.

[0056] Step S130: Determine whether the semantic knowledge base of the receiving node needs to be updated. If it needs to be updated, update the semantic knowledge base of the receiving node according to the first data transmitted by the sending node to obtain the synchronized first communication semantic knowledge base.

[0057] Step S140: The sending node and the receiving node communicate according to the synchronized first communication semantic knowledge base.

[0058] For example, the first communication phase is the initial stage of communication between the communicating parties in the satellite communication system, and only a small amount (i.e., the first quantity) of communication data can be used to train a small semantic knowledge base (i.e., the first semantic knowledge base). This application does not limit the duration of the first communication phase, the amount of the first quantity of communication data, or the size of the first semantic knowledge base.

[0059] Figure 2 This is an exemplary flowchart of a knowledge base synchronization method for space-based semantic communication according to another embodiment of this application. (See reference...) Figure 2 As shown, the knowledge base synchronization method for space-based semantic communication in this embodiment further includes, after step S140 described above:

[0060] Step S150: Determine whether the amount of communication data has reached a preset threshold. If it has reached the preset threshold, proceed to step S160.

[0061] Step S160: In the second communication phase, a general semantic knowledge base is trained based on a second quantity of communication data to obtain a second semantic knowledge base. The second semantic knowledge base is then transmitted to the sending node and the receiving node, wherein the second quantity is greater than the first quantity.

[0062] Step S170: The sending node updates its own semantic knowledge base according to the federated learning method to obtain the federated semantic knowledge base.

[0063] Step S180: Transmit the federated semantic knowledge base to the receiving node.

[0064] Step S190: The sending node and the receiving node communicate according to the federated semantic knowledge base.

[0065] For example, the second communication phase is the later stage of communication between the communicating parties in the satellite communication system, during which a large amount (i.e., the second quantity) of communication data is accumulated and can be used to train a large semantic knowledge base (i.e., the second semantic knowledge base). This application does not limit the duration of the second communication phase, the amount of the second quantity of communication data, or the size of the second semantic knowledge base.

[0066] It should be noted that steps S110 to S140 mentioned above are the main steps of this application, while steps S150 to S190 are optional steps. In practical applications, it is possible to choose to perform only steps S110 to S140, or to perform steps S110 to S190. This application does not impose any restrictions.

[0067] In some embodiments, the ground training end continuously collects communication data during the communication process between the sending node and the receiving node, and trains a general semantic knowledge base. The ground training end controls the size of the general semantic knowledge base through a mask. During the training of the general semantic knowledge base from the first communication stage to the second communication stage, the proportion of 0 in the mask is gradually reduced to gradually increase the general semantic knowledge base.

[0068] For example, since the aforementioned large semantic knowledge base (i.e., the second semantic knowledge base) can reuse the knowledge learned from the small semantic knowledge base (i.e., the first semantic knowledge base), in step S110 above, when initializing the general semantic knowledge base, the structure of the general semantic knowledge base can be designed as a large semantic knowledge base, and a mask can be used to control the size of the general semantic knowledge base. During the first communication phase, the mask is used to perform an intersection operation on the general semantic knowledge base to obtain a small semantic knowledge base, which is then pre-trained on the ground training end using a small amount of communication data. As the amount of communication data gradually increases, the proportion of 0s in the mask can be gradually reduced during the second communication phase to gradually increase the general semantic knowledge base. Before each increase in the general semantic knowledge base, pre-training using collected data is required on the ground training end.

[0069] For example, the general semantic knowledge base used in the first communication stage and the second communication stage is essentially the same semantic knowledge base, but the size of the semantic knowledge base is controlled by using different proportions of 0 in the mask in different communication stages. In the first communication stage, the proportion of 0 in the mask is larger, so the general semantic knowledge base is smaller, and the trained first semantic knowledge base is also smaller; as the training of the general semantic knowledge base progresses from the first communication stage to the second communication stage, the proportion of 0 in the mask gradually decreases, so the general semantic knowledge base gradually increases, and the trained second semantic knowledge base also gradually increases. This configuration in this application gradually enhances the universality of the semantic knowledge base.

[0070] The following details steps S110 to S190 described above:

[0071] In step S110, the general semantic knowledge base is initialized. Exemplarily, this step involves designing the structure of the general semantic knowledge base and creating the initial general semantic knowledge base.

[0072] Figure 3 This is an exemplary architecture diagram of communication between various communication terminals using a semantic knowledge base, according to an embodiment of this application. Exemplarily, in the first communication phase, only a small amount of data is available for training a small semantic knowledge base, and each communication terminal stores several copies of the semantic knowledge base to adapt to the semantic knowledge bases of other communication terminals. (See reference...) Figure 3As shown, satellite A, satellite N, and ground station B each store n semantic knowledge bases. Satellite A and ground station B communicate using semantic knowledge base 1; satellite A and satellite N communicate using semantic knowledge base 2; and satellite N and ground station B communicate using semantic knowledge base 3. Figure 3 The communication architecture shown is only an example. This application does not limit the communication architecture of each communication terminal and can be set independently according to the actual situation.

[0073] In step S120, during the first communication phase, a general semantic knowledge base is trained based on a first amount of communication data to obtain a first semantic knowledge base, which is then transmitted to the sending node and the receiving node. The two communicating parties can then perform semantic communication based on the first semantic knowledge base.

[0074] Figure 4 This is an exemplary schematic diagram of an end-to-end training framework according to an embodiment of this application. In some embodiments, step S120, which involves training a general semantic knowledge base based on a first amount of communication data, includes:

[0075] Step S1201: Construct an end-to-end training framework including a transmitter encoder and a receiver decoder, the end-to-end training framework using the Transformer mechanism; for example, refer to Figure 4 As shown, the end-to-end training framework includes a transmitter encoder 403, a channel simulator 404, a receiver decoder 405, a transmitter semantic knowledge base 401, and a receiver semantic knowledge base 402. The transmitter semantic knowledge base 401 guides the transmitter encoder 403 in encoding the source data, and the receiver semantic knowledge base 402 guides the receiver decoder 405 in decoding the source data. The channel simulator 404 simulates channel conditions, incorporating channel noise into the semantic coding process and improving its noise resistance. The transmitter encoder 403 and receiver decoder 405 are implemented based on the Transformer mechanism. The transmitter encoder 403 receives and encodes the source data S, and its learnable parameters are the encoder's key vector weights. Encoder query vector weights and encoder value vector weights The semantic representation is obtained through the encoder in the Transformer. The receiver decoder 405 receives the semantic representation and decodes it. Its learnable parameters are the key vector weights of the decoder. Decoder query vector weights and the value vector weights of the decoder The decoder in the Transformer obtains the decoded source data.

[0076] Step S1202: Input the training source data s to the transmitter encoder to obtain the semantic representation corresponding to the training source data s;

[0077] Step S1203: The receiver decoder decodes the semantic representation corresponding to the training source data s and outputs the decoded training source data. And obtain a pre-trained end-to-end training framework;

[0078] Step S1204: Input the first amount of communication data into the trained end-to-end training framework, and use the cross-entropy of the initial source data and the decoded source data as the Loss function to update the semantic knowledge base parameters used by the transmitter encoder and receiver decoder respectively.

[0079] Figure 5 This is an exemplary schematic diagram of a self-updating framework for the semantic knowledge base of a sending node according to an embodiment of this application. Exemplarily, after step S120 described above, all communication nodes run their own semantic knowledge bases, and when a sending node receives new unknown primitive information, it automatically updates its semantic knowledge base. During the process of the sending node self-updating its semantic knowledge base, a self-updating framework for the knowledge base needs to be constructed.

[0080] For example, refer to Figure 5 As shown, the semantic knowledge base 500 of the sending end constructs a virtual sending end 503 and a virtual receiving end 504, and uses... Figure 4 The end-to-end training method shown updates the semantic knowledge base of the sending end itself. During self-updating, the sending end needs to imagine a receiving end communicating with it in order to use the knowledge. Figure 4 The end-to-end training method shown trains (i.e., self-updates) the semantic knowledge base of the sending end. The trained model is the virtual sending end semantic knowledge base 501. This model does not need to output anything. After the sending end completes its self-update, it uses this semantic knowledge base to communicate with other communication nodes.

[0081] In step S130, it is determined whether the receiving node's own semantic knowledge base needs to be updated. If an update is needed, the receiving node's own semantic knowledge base is updated according to the first data transmitted by the sending node, resulting in a synchronized first communication semantic knowledge base. For example, the receiving node only updates and synchronizes its own semantic knowledge base according to the first data sent by the sending node when it is determined that an update is needed; if it is determined that an update is not needed, the communicating parties use the existing semantic knowledge base for communication. This configuration reduces the frequency of semantic knowledge base updates between the communicating parties, thus reducing the resource and time consumption for synchronizing the semantic knowledge base.

[0082] This application proposes two implementation methods for synchronizing semantic knowledge bases between communicating parties, based on the number of nodes participating in the communication. Method 1 is suitable for situations with a large number of participating nodes, such as hundreds of communication nodes; Method 2 is suitable for situations with a small number of participating nodes, such as dozens of communication nodes. This application does not limit the number of communication nodes, and either Method 1 or Method 2 can be selected according to the actual situation. Method 1 will be introduced first, and Method 2 will be introduced later.

[0083] Method 1: Applicable when there are many nodes participating in the communication, the number of semantic knowledge base copies stored by each receiving node can be customized. In some embodiments, the step S130, determining whether the receiving node's own semantic knowledge base needs to be updated, includes:

[0084] Step S1311: The sending node transmits k pieces of first data to the receiving node, where k is a positive integer greater than or equal to 1. The first data includes the source data and the semantic representation of the source data after semantic encoding. For example, k is less than the total number of information to be sent in this communication.

[0085] Step S1312: Decode the semantic knowledge base of the receiving node based on the semantic representation. For each semantic knowledge base of the receiving node, obtain k decoded data. For the i-th knowledge base version, calculate the average similarity score (average_similarity) between the k decoded data and the source data. i For example, after receiving the k first data entries, the receiving node uses all its stored semantic knowledge base versions (assuming the receiving node currently has m versions of semantic knowledge base) to decode the received semantic representation. Each semantic knowledge base version obtains k decoded data entries. For the i-th knowledge base version, using the Transformer decoder mechanism, the source data and its semantic representation are used as the input to the decoder. The average of the outputs of the softmax layer in the decoder is taken as the similarity score of that source data entry. The average of the similarity scores of these k data entries is taken as the average similarity score. i Subsequent communication will use the semantic knowledge base version with the highest average similarity score.

[0086] Step S1313: Calculate the maximum value of the average similarity score using the following formula (1):

[0087] score=max(average_similarity i (i∈[1,m]) (1)

[0088] Where m represents the number of versions of the semantic knowledge base of the receiving node itself;

[0089] Step S1314: Calculate the version number (index) of the semantic knowledge base corresponding to the maximum average similarity score using the following formula (2):

[0090] index=argmax(average_similarity i (2)

[0091] Step S1315: Determine whether the maximum value of the average similarity score (score) reaches the preset similarity threshold. If it does not reach the preset similarity threshold, the semantic knowledge base of the receiving node itself needs to be updated; if it reaches the preset similarity threshold, communication is performed based on the semantic knowledge base of the version number (index). For example, the similarity threshold can be set independently, and this application does not impose any restrictions.

[0092] In some embodiments, step S130, which involves updating the semantic knowledge base of the receiving node based on the first data transmitted by the sending node, includes: the receiving node performing transfer training on the semantic knowledge base with version number index using k pieces of the first data. After the transfer training is completed, the receiving node uses the transferred-trained semantic knowledge base for communication. For example, when communication is possible, the receiving node confirms to the sending node that communication is ready by sending an acknowledgment (ACK) message. The ACK message is a confirmation message after the receiving node has correctly received the information. The sending node will only proceed with subsequent communication after receiving the ACK message from the receiving node.

[0093] For example, when the receiving node needs to perform transfer training, the transfer training will generate a new version of the semantic knowledge base, at which point the receiving node needs to save the new semantic knowledge base. Since satellite storage resources are limited, and the update process requires calculating the similarity between all versions of the semantic knowledge base and the source data, the number of semantic knowledge base versions should not be too large.

[0094] This application addresses the storage method for a new semantic knowledge base. It presupposes that a total of l versions of the semantic knowledge base can be stored. If the number m of existing semantic knowledge bases is less than the storage capacity l, then the new version of the semantic knowledge base is stored, and its corresponding number is m+1. If the number m of existing semantic knowledge bases is greater than or equal to the storage capacity l, then there are two further cases: (1) If the maximum value of the average similarity score is greater than or equal to the preset storage value, then the new version of the semantic knowledge base is not stored; (2) If the maximum value of the average similarity score is less than the preset storage value, then the new version of the semantic knowledge base is stored, and its corresponding number is index. This application does not impose any restrictions on the preset storage value and allows for independent setting.

[0095] Method 2: Applicable when the number of participating nodes is small, the number of semantic knowledge base copies stored by each receiving node is fixed. For example, if there are p communication nodes, each communication node stores p-1 semantic knowledge base copies. The semantic knowledge base copies are encoded according to the name of the communication partner, and communication is performed using the semantic knowledge base copy corresponding to the communication partner. In some embodiments, step S130, determining whether to update the receiving node's own semantic knowledge base, includes:

[0096] Step S1321: Calculate the update magnitude α of the sending node. T And calculate the update magnitude α of the receiving node. R In some embodiments, in step S1321, the update magnitude α of the sending node is calculated. T The steps include: encoding the data to be sent using a copy of the semantic knowledge base from the first communication phase to obtain a semantic representation s of the data to be sent. old Input semantic representation s old The source data and its corresponding data are fed into the virtual decoder, and the average value of the output of the softmax layer in the virtual decoder is taken as the semantic representation s. old The update magnitude α of the sending node is calculated using the following formula (3) based on the similarity score (similarity_T) of the corresponding source data. T :

[0097]

[0098] Where x is a hyperparameter;

[0099] And, calculate the update magnitude α of the receiving node. R The steps include: randomly storing the received semantic representation c every preset time interval during the communication process. t Construct test set C = {c t , t=0,...,j}, where, T_total is the duration of a single communication, and T is the preset duration (i.e., the storage duration c). t (Time interval); Randomly select a semantic representation c from the test set C. t Decode the semantic representation c using a dynamically updated semantic knowledge base. t Obtain semantic representation c t The source data c new Decode the semantic representation c using a copy of the semantic knowledge base from the first communication phase. t Obtain semantic representation c t The source data c old ; Calculate source data c new and source data c oldThe similarity score is similarity_R. For example, the definition of similarity differs depending on the type of original information. When the original information is text data, similarity can be defined as the Euclidean distance of the text vectors, word shift distance, etc.; when the original information is image data, similarity can be defined as the Structure Similarity Index Measure (SSIM), hash distance, etc. The update magnitude α of the receiving node is calculated using the following formula (4). R :

[0100]

[0101] Where x is a hyperparameter.

[0102] Figure 6 This is an exemplary communication flowchart of an embodiment of this application when the sending node and the receiving node do not need to synchronize their semantic knowledge bases. (See reference) Figure 6 As shown, in this communication process, the sending node 601 transmits sending synchronization information to the receiving node 602 in step S6001; the receiving node 602 transmits receiving synchronization information to the sending node 601 in step S6002; the sending node 601 transmits semantic representation to the receiving node 602 in step S6003, and the two communicating parties do not need to synchronize their semantic knowledge bases; the receiving node 602 transmits ACK to the sending node 601 in step S6004, and the two parties then begin communication. Here, ACK is an acknowledgment message after the receiving node 602 has correctly received the information, and the sending node 601 will only proceed with the next communication process after receiving the ACK from the receiving node 602.

[0103] Figure 7 This is an exemplary communication flowchart illustrating how a sending node and a receiving node need to synchronize a semantic knowledge base, according to an embodiment of this application. (See reference...) Figure 7As shown, in this communication process, the sending node 601 transmits sending synchronization information to the receiving node 602 in step S6001; the receiving node 602 transmits receiving synchronization information to the sending node 601 in step S6002; the sending node 601 transmits semantic representation to the receiving node 602 in step S6003, and both parties need to synchronize their semantic knowledge bases; the receiving node 602 transmits ACK to the sending node 601 in step S6004; the sending node 601 transmits source data to the receiving node 602 in step S6005; the receiving node 602 transfers and trains the semantic knowledge base in step S60051, and transmits "receiving ready" to the sending node 601 in step S6006; the sending node 601 transmits semantic representation to the receiving node 602 in step S6007; the receiving node 602 transmits ACK to the sending node 601 in step S6008, and then the two parties begin communication. The “Receiver Ready” message is a confirmation message from the receiver node 602 after it has completed the semantic knowledge base transfer training. The sender node 601 will only proceed with the next communication process after receiving the “Receiver Ready” message from the receiver node 602.

[0104] For example, in the process of determining whether to synchronize the semantic knowledge base, both parties send synchronization information to each other based on whether their respective update magnitude exceeds the threshold to determine whether to synchronize, and store the learning rate and the other party's name in the synchronization information sent by the other party.

[0105] Step S1322: Determine the update magnitude α of the sending node respectively. T and the update magnitude α of the receiving node R Is it greater than the preset update threshold? If α T and α R If neither value exceeds the preset update threshold, then communication is based on the existing semantic knowledge base, and synchronization of the semantic knowledge base is not required. The communication process between the two parties is as follows: Figure 6 As shown; if α T and / or α R If the update magnitude of at least one of the two parties exceeds the preset update threshold, then the semantic knowledge base of the receiving node itself needs to be updated. The communication process between the two parties is as follows: Figure 7 As shown. This application determines whether to update the semantic knowledge base of the receiving node by exchanging synchronization information once between the two communicating parties. During the communication process between the satellite and the ground, it avoids the bandwidth consumption caused by repeatedly updating the semantic knowledge base of both parties at the same time on both the space and ground sides, and shortens the available communication time between the two parties.

[0106] Step S1323: The sending node transmits k pieces of first data to the receiving node, where k is a positive integer greater than or equal to 1. The first data includes the source data and the semantic representation of the source data after semantic encoding.

[0107] In some embodiments, step S130, updating the semantic knowledge base of the receiving node based on the first data transmitted by the sending node, includes: calculating the training learning rate α using the following formula (5):

[0108] α = max(α) T α R (5)

[0109] Where, α T α represents the update magnitude of the sending node. R Indicates the update magnitude α of the receiving node. R The receiving node uses k source data as labels and k semantic representations as input, and performs transfer training on the semantic knowledge base according to the training learning rate α. After transfer training, the receiving node saves the new version of the semantic knowledge base and overwrites the semantic knowledge base before transfer training. This new version of the semantic knowledge base is also known as the first communication semantic knowledge base.

[0110] This application may employ either method 1 or method 2 described above during the synchronization of the semantic knowledge base. When semantic knowledge base synchronization is required, the sending node sends source data and its semantic representation to the receiving node, which then uses the received data to perform transfer training on its own semantic knowledge base. This avoids the high communication resource overhead caused by the direct transmission of the entire semantic knowledge base between the communicating parties, thereby achieving the goal of eliminating the need to exchange semantic knowledge bases and increasing the data communication duration within the available communication period.

[0111] In step S140, the sending node and the receiving node communicate according to the synchronized first communication semantic knowledge base.

[0112] For example, after a period of communication, the ground training end of a satellite communication system accumulates a large amount of data for training a large semantic knowledge base. This data contains information about the receiving and sending ends during the communication process, thus enabling the training of a larger semantic knowledge base. All subsequent communication nodes use the same semantic knowledge base, avoiding the resource and time consumption associated with synchronizing the semantic knowledge base. In step S150, it is determined whether the amount of communication data reaches a preset threshold. If it does, step S160 is executed. This application does not impose any restrictions on this preset threshold; it can be set independently.

[0113] In step S160, during the second communication phase, a general semantic knowledge base is trained based on a second quantity of communication data to obtain a second semantic knowledge base. This second semantic knowledge base is then transmitted to the sending and receiving nodes, where the second quantity is greater than the first quantity. It should be noted that the method for training the general semantic knowledge base in the second communication phase is essentially the same as the method described above for training the general semantic knowledge base in the first communication phase, and will not be repeated here. The difference lies in the specific data structure of the second quantity of communication data used to input the end-to-end training framework in the second communication phase.

[0114] In some embodiments, the data structure of the second amount of communication data in step S160 is shown in Table 1 below:

[0115] Table 1. Data Structure of the Second Quantity of Communication Data

[0116] start symbol Name of the communicating party separator data End symbol

[0117] Table 1 includes: a start symbol, the name of the communicating party, a separator, the data, and an end symbol. The name of the communicating party includes the name of the sending node and the name of the receiving node, and the data is the source data without semantic encoding. For example, during communication, the receiving node stores not only the received data but also the name of the sending node it communicates with. When constructing the dataset, the name of the receiving node and the name of the sending node that sent the data are encoded as the name of the communicating party, and the start symbol, separator, and end symbol are used to separate the name of the communicating party and the data. To fully utilize the large amount of accumulated data, the semantic knowledge base trained in the second communication stage has a larger number of parameters.

[0118] In step S170, the sending node updates its own semantic knowledge base according to the federated learning method to obtain a federated semantic knowledge base. For example, considering that the sending and receiving ends are clusters, and each node within the cluster receives new data with different feature spaces, which aligns with the application scenario of horizontal federated learning, this application uses the method of horizontal federated learning to exchange semantic knowledge bases among the sending nodes, jointly updating the semantic knowledge base within the sending cluster. Subsequently, the decoder in the updated semantic knowledge base is broadcast to all receiving nodes to complete the synchronization of the semantic knowledge base.

[0119] For example, each sender node in the sender cluster may encounter unknown data during runtime and update its own semantic knowledge base. All sender nodes will update the federated model asynchronously and on an equal footing. After multiple iterations, the decoder in the federated semantic knowledge base that has completed the current round of updates will be broadcast to all receiver nodes. Then, the two communicating parties use the federated semantic knowledge base for semantic communication.

[0120] For example, a peer-to-peer asynchronous update federated model means that, for instance, n sending nodes would have n semantic knowledge bases. During the operation of a satellite communication system, some sending nodes would update their own semantic knowledge bases upon encountering unknown data, thus creating n different semantic knowledge bases. Subsequently, the communicable sending nodes continuously exchange their semantic knowledge bases until each sending node has the semantic knowledge bases of all other sending nodes. Each sending node then uses the same method to aggregate these semantic knowledge bases, forming a new "federated model." This process is the peer-to-peer asynchronous update federated model.

[0121] For example, inter-satellite links exist within the satellite constellation of the transmitting node cluster to exchange semantic knowledge bases between satellites. In some embodiments, step S170, in which the transmitting node updates its own semantic knowledge base according to the federated learning method, includes: all transmitting nodes exchanging their respective semantic knowledge bases with each other until each transmitting node obtains the semantic knowledge bases of all other transmitting nodes after y iterations, where y is a positive integer greater than or equal to 1; each transmitting node merging all its own semantic knowledge bases using an aggregation algorithm.

[0122] In some embodiments, the federated semantic knowledge base of the sending node i needs to be updated for t rounds, where i and t are positive integers greater than or equal to 1. During the t-th round of federated semantic knowledge base update, the sending node i obtains the model parameters of the semantic knowledge base at time t+1. The steps include:

[0123] Step S1: The sending node i iterates the model parameters of its own semantic knowledge base at time t. This includes: constructing an end-to-end semantic knowledge base self-updating framework comprising a virtual sender and a virtual receiver; the end-to-end semantic knowledge base self-updating framework calculates model parameters using the following formula (6) after y iterations.

[0124]

[0125] Where η represents the update rate, The gradient operator is represented by l(·), the loss function is represented by l(·), and the training dataset is represented by b.

[0126] Step S2: The sending node i adds its own ID i to its own semantic knowledge base name list list. i And transmit a list of semantic knowledge base names to other sending nodes. i ;

[0127] Step S3: Sending node i receives the list of semantic knowledge base names transmitted by other sending node j. j , j∈Ni N i This represents the set of other sending nodes; and a list of semantic knowledge base names transmitted to sending node j. i The list of semantic knowledge base names is removed. j This includes all semantic knowledge base information other than the semantic knowledge base itself. The data structure of the semantic knowledge base information includes: knowledge base name l, and the amount of data b used by the semantic knowledge base in this iteration. l Semantic knowledge base parameters

[0128] Step S4: When other sending node j switches or the semantic knowledge base name list is accessed... i When the content changes, repeat steps S2 to S3 until the list of semantic knowledge base names is reached. i All sending nodes are included; sending node i adds its own ID i to its own acknowledgment list list_ACK. i And transmit the acknowledgment list list_ACK to other sending node j. i ; Receive ACK information transmitted by other sending node j j And store it in the acknowledgment information list list_ACK i ;

[0129] Step S5: When the confirmation message list list_ACK is received i When the confirmation information of all sending nodes is included, the following formula (7) is used to aggregate all semantic knowledge bases to obtain the federated semantic knowledge base z:

[0130]

[0131] Where n represents the total number of semantic knowledge bases.

[0132] In step S180, the sending node transmits the federated semantic knowledge base to the receiving node, and the sending node broadcasts the updated decoder in the federated semantic knowledge base to all communicable receiving nodes.

[0133] In step S190, the sending node and the receiving node communicate based on the federated semantic knowledge base. For example, referring to Table 1 above, during communication, the sending node encodes its own name and the receiving node's name as "communicator name," and uses the federated semantic knowledge base to semantically encode the source data. At this point, the "data" in Table 1 is the semantically encoded data. After receiving the data shown in Table 1, the receiving node decodes the obtained data using the federated semantic knowledge base.

[0134] Figure 8This is an exemplary flowchart of a knowledge base synchronization method for space-based semantic communication according to another embodiment of this application. (See reference...) Figure 8 As shown in the flowchart, the process includes a first communication stage and a second communication stage. In the second communication stage, the proportion of 0 in the mask, alpha, is continuously reduced until alpha is 0. Alpha and gamma are natural numbers not less than 0. Alpha, gamma, the first preset threshold in step S805, and the second preset threshold in step S811 can be set independently, and this application does not impose any restrictions.

[0135] For example, refer to Figure 8 As shown, in step S801, a general semantic knowledge base is designed and initialized; in step S802, a mask operation with a ratio of 0 to alpha is performed on the general semantic knowledge base, and the general semantic knowledge base under this mask is trained on the ground training end; in step S803, all communication nodes run their own semantic knowledge base and update their semantic knowledge bases; in step S804, the two communicating parties communicate after synchronizing their semantic knowledge bases; in step S805, the ground training end determines whether the amount of collected communication data has reached the first preset threshold. If it has, step S803 is executed; if it has, step S806 is executed.

[0136] Continue to refer to Figure 8 As shown, in step S806, the proportion of 0 in the mask is set to alpha = alpha - gamma; in step S807, a masking operation with a proportion of 0 of alpha is performed on the general semantic knowledge base, and the general semantic knowledge base under this mask is trained at the ground training end; in step S808, all sending nodes update the semantic knowledge base through federated learning; in step S809, the sending nodes transmit the updated federated semantic knowledge base to the receiving nodes; in step S810, the two communicating parties communicate using the federated semantic knowledge base; in step S811, the ground training end determines whether the amount of collected communication data has reached the second preset threshold. If yes, otherwise step S808 is executed; if yes, step S812 is executed; in step S812, the ground training end determines whether alpha is 0. If yes, otherwise step S806 is executed; if yes, step S813 is executed; in step S813, the two communicating parties continue to communicate using the current semantic knowledge base.

[0137] The knowledge base synchronization method for space-based semantic communication proposed in this application enables semantic communication to be established even when both communicating parties update their own semantic knowledge bases. In the first communication phase, since only a small amount of data is used for training, a small semantic knowledge base (first semantic knowledge base) is employed. The communicating parties only need to exchange synchronization information before each communication, eliminating the need for repeated exchange of new data and saving time and bandwidth resources required for semantic knowledge base synchronization. In the second communication phase, a large amount of data has accumulated, allowing the training of a large semantic knowledge base (second semantic knowledge base). This enables each communicating node to use the same semantic knowledge base, avoiding the resource and time consumption associated with semantic knowledge base updates. This application proposes different semantic knowledge base synchronization methods for different stages of space-based semantic communication. While ensuring communication, the semantic knowledge base update frequency for both communicating parties is low, reducing communication overhead. Furthermore, the semantic knowledge bases of the receiving and sending ends are updated independently, avoiding the need for semantic knowledge base exchange. Additionally, the semantic knowledge base update time is low, allowing for longer data communication time within the communication period.

[0138] This application also includes a knowledge base synchronization system for space-based semantic communication, comprising a memory and a processor. The memory stores instructions executable by the processor; the processor executes these instructions to implement the aforementioned knowledge base synchronization method for space-based semantic communication.

[0139] Figure 9 This is a system block diagram of a knowledge base synchronization system for space-based semantic communication according to an embodiment of this application. (Reference) Figure 9As shown, the space-based semantic communication knowledge base synchronization system 900 may include an internal communication bus 901, a processor 902, a read-only memory (ROM) 903, a random access memory (RAM) 904, and a communication port 905. When applied to a personal computer, the space-based semantic communication knowledge base synchronization system 900 may also include a hard disk 906. The internal communication bus 901 enables data communication between the components of the space-based semantic communication knowledge base synchronization system 900. The processor 902 can perform judgments and issue prompts. In some embodiments, the processor 902 may consist of one or more processors. The communication port 905 enables data communication between the space-based semantic communication knowledge base synchronization system 900 and external systems. In some embodiments, the space-based semantic communication knowledge base synchronization system 900 can send and receive information and data from a network through the communication port 905. The space-based semantic communication knowledge base synchronization system 900 may also include different forms of program storage units and data storage units, such as a hard disk 906, a read-only memory (ROM) 903, and a random access memory (RAM) 904, capable of storing various data files used for computer processing and / or communication, as well as possible program instructions executed by the processor 902. The processor executes these instructions to implement the main part of the method. The results of the processor processing are transmitted to the user equipment through the communication port and displayed on the user interface.

[0140] The aforementioned knowledge base synchronization method for space-based semantic communication can be implemented as a computer program, stored in hard disk 906, and loaded into processor 902 for execution to implement the knowledge base synchronization method for space-based semantic communication of this application.

[0141] This application also includes a computer-readable medium storing computer program code that, when executed by a processor, implements the knowledge base synchronization method for space-based semantic communication described above.

[0142] When the knowledge base synchronization method for space-based semantic communication is implemented as a computer program, it can also be stored as an article of art in a computer-readable storage medium. For example, computer-readable storage media can include, but are not limited to, magnetic storage devices (e.g., hard disks, floppy disks, magnetic stripes), optical discs (e.g., compact discs (CDs), digital multifunction discs (DVDs)), smart cards, and flash memory devices (e.g., electrically erasable programmable read-only memory (EPROM), cards, sticks, key drives). Furthermore, the various storage media described herein can represent one or more devices and / or other machine-readable media used for storing information. The term "machine-readable medium" can include, but is not limited to, wireless channels and various other media (and / or storage media) capable of storing, containing, and / or carrying code and / or instructions and / or data.

[0143] It should be understood that the embodiments described above are merely illustrative. The embodiments described herein may be implemented in hardware, software, firmware, middleware, microcode, or any combination thereof. For hardware implementation, the processor may be implemented within one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, and / or other electronic units designed to perform the functions described herein, or combinations thereof.

[0144] Some aspects of this application can be executed entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The aforementioned hardware or software may be referred to as a "data block," "module," "engine," "unit," "component," or "system." The processor may be one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DAPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, or combinations thereof. Furthermore, aspects of this application may manifest as computer products residing in one or more computer-readable media, including computer-readable program code. For example, computer-readable media may include, but are not limited to, magnetic storage devices (e.g., hard disks, floppy disks, magnetic tapes, etc.), optical discs (e.g., compressed CDs, digital multifunction DVDs, etc.), smart cards, and flash memory devices (e.g., cards, sticks, key drives, etc.).

[0145] A computer-readable medium may contain a propagated data signal containing computer program code, for example, on baseband or as part of a carrier wave. This propagated signal may take various forms, including electromagnetic, optical, and so on, or suitable combinations thereof. A computer-readable medium can be any computer-readable medium other than a computer-readable storage medium, which can be connected to an instruction execution system, apparatus, or device to enable communication, propagation, or transmission of a program for use. The program code located on the computer-readable medium can be propagated through any suitable medium, including radio, cable, fiber optic cable, radio frequency signals, or similar media, or any combination of the above media.

[0146] The basic concepts have been described above. Obviously, for those skilled in the art, the above disclosure is merely illustrative and does not constitute a limitation of this application. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements, and corrections to this application. Such modifications, improvements, and corrections are suggested in this application, and therefore remain within the spirit and scope of the exemplary embodiments of this application.

[0147] Furthermore, this application uses specific terms to describe embodiments of the application. For example, "an embodiment," "one embodiment," and / or "some embodiments" refer to a particular feature, structure, or characteristic related to at least one embodiment of the application. Therefore, it should be emphasized and noted that "an embodiment," "one embodiment," or "an alternative embodiment" mentioned twice or more in different locations in this specification do not necessarily refer to the same embodiment. In addition, certain features, structures, or characteristics in one or more embodiments of the application can be appropriately combined.

[0148] In some embodiments, numbers describing the quantity of components and attributes are used. It should be understood that such numbers used in the description of embodiments are modified in some examples with the terms "approximately," "approximately," or "generally." Unless otherwise stated, "approximately," "approximately," or "generally" indicates that the numbers are allowed to vary by ±20%. Accordingly, in some embodiments, the numerical parameters used in the specification and claims are approximate values, which may be changed depending on the characteristics required by individual embodiments. In some embodiments, numerical parameters should take into account specified significant digits and employ a general method of digit reservation. Although the numerical ranges and parameters used to confirm their breadth of scope in some embodiments of this application are approximate values, in specific embodiments, such values ​​are set as precisely as feasible.

Claims

1. A knowledge base synchronization method for space-based semantic communication, characterized in that, include: Initialize the general semantic knowledge base; In the first communication phase, the general semantic knowledge base is trained based on a first amount of communication data to obtain a first semantic knowledge base, and the first semantic knowledge base is transmitted to the sending node and the receiving node. Determine whether the semantic knowledge base of the receiving node needs to be updated. If it needs to be updated, update the semantic knowledge base of the receiving node according to the first data transmitted by the sending node to obtain the synchronized first communication semantic knowledge base. The sending node and the receiving node communicate according to the synchronized first communication semantic knowledge base; Determine whether the amount of communication data has reached a preset threshold. If it has, then... In the second communication phase, the general semantic knowledge base is trained based on a second quantity of communication data to obtain a second semantic knowledge base, and the second semantic knowledge base is transmitted to the sending node and the receiving node, wherein the second quantity is greater than the first quantity.

2. The knowledge base synchronization method as described in claim 1, characterized in that, Following the step of the sending node and the receiving node communicating according to the synchronized first communication semantic knowledge base, the method further includes: The sending node updates its own semantic knowledge base according to the federated learning method to obtain a federated semantic knowledge base. Transmit the federated semantic knowledge base to the receiving node; The sending node and the receiving node communicate based on the federated semantic knowledge base.

3. The knowledge base synchronization method as described in claim 2, characterized in that, During the communication process between the sending node and the receiving node, the ground training terminal continuously collects the communication data and trains the general semantic knowledge base. The ground training terminal controls the size of the general semantic knowledge base through a mask. During the training of the general semantic knowledge base from the first communication stage to the second communication stage, the proportion of 0 in the mask is gradually reduced to gradually increase the general semantic knowledge base.

4. The knowledge base synchronization method as described in claim 1, characterized in that, The steps for determining whether the semantic knowledge base of the receiving node needs to be updated include: The sending node transmits k pieces of the first data to the receiving node, where k is a positive integer greater than or equal to 1. The first data includes source data and the semantically encoded semantic representation corresponding to the source data. The semantic knowledge base of the receiving node is decoded according to the semantic representation. For each semantic knowledge base of the receiving node, k decoded data are obtained. For the i-th knowledge base version, the average similarity score between the k decoded data and the source data is calculated. The maximum value of the average similarity score is calculated using the following formula. : Where m represents the number of versions of the semantic knowledge base of the receiving node itself; The maximum value of the average similarity score is calculated using the following formula. Version number of the corresponding semantic knowledge base : Determine the maximum value of the average similarity score. Whether a preset similarity threshold has been reached; if not, the semantic knowledge base of the receiving node itself needs to be updated; if the preset similarity threshold has been reached, it is determined according to the version number. Semantic knowledge base communication.

5. The knowledge base synchronization method as described in claim 4, characterized in that, The steps of updating the semantic knowledge base of the receiving node based on the first data transmitted by the sending node include: The receiving node uses k pieces of the first data to number the version. Transfer training is performed on the semantic knowledge base.

6. The knowledge base synchronization method as described in claim 1, characterized in that, The steps for determining whether the semantic knowledge base of the receiving node needs to be updated include: Calculate the update magnitude of the sending node. And calculate the update magnitude of the receiving node. ; Determine the update magnitude of the sending node respectively. and the update magnitude of the receiving node Is it greater than the preset update threshold? and All values ​​are not greater than the preset update threshold, and communication is based on the existing semantic knowledge base; if and / or If the value exceeds the preset update threshold, the semantic knowledge base of the receiving node itself needs to be updated. The sending node transmits k pieces of the first data to the receiving node, where k is a positive integer greater than or equal to 1. The first data includes source data and the semantically encoded semantic representation corresponding to the source data.

7. The knowledge base synchronization method as described in claim 6, characterized in that, Calculate the update magnitude of the sending node. The steps include: The data to be sent is encoded using a copy of the semantic knowledge base from the first communication phase to obtain a semantic representation of the data to be sent. Input the semantic representation The corresponding source data is sent to the virtual decoder, and the data from the virtual decoder is retrieved. The average value of the layer's output is used as a semantic representation. Similarity score of the corresponding source data The update magnitude of the sending node is calculated using the following formula. : in, For hyperparameters; And, calculate the update magnitude of the receiving node. The steps include: During the communication process, the received semantic representation is randomly stored every preset time interval. Build test set ,in, , The duration of a single communication. The preset duration; in the test set Randomly select a semantic representation The semantic representation is decoded using a dynamically updated semantic knowledge base. To obtain the semantic representation Source data Decode the semantic representation using a copy of the semantic knowledge base from the first communication phase. To obtain the semantic representation Source data ; Calculate the source data and the source data similarity score The update magnitude of the receiving node is calculated using the following formula. : 。 8. The knowledge base synchronization method as described in claim 6, characterized in that, The steps of updating the semantic knowledge base of the receiving node based on the first data transmitted by the sending node include: Calculate the training learning rate using the following formula. : in, This indicates the update magnitude of the sending node. Indicates the update magnitude of the receiving node. ; The receiving node takes k source data items as labels and k semantic representations as input, and then applies the training learning rate. Transfer training is performed on the semantic knowledge base.

9. The knowledge base synchronization method as described in claim 1, characterized in that, The steps for training the general semantic knowledge base based on a first quantity of communication data include: An end-to-end training framework is constructed, comprising a transmitter encoder and a receiver decoder, wherein the end-to-end training framework uses the Transformer mechanism; Input training source data The training source data is obtained from the encoder at the transmitting end. Corresponding semantic representation; The receiver decoder decodes the training source data. The corresponding semantic representation outputs the decoded training source data. And obtain a pre-trained end-to-end training framework; Input the first number of communication data into the trained end-to-end training framework, and use the cross-entropy of the initial source data and the decoded source data as the Loss function to update the semantic knowledge base parameters used by the sending encoder and the receiving decoder, respectively.

10. The knowledge base synchronization method as described in claim 2, characterized in that, The data structure of the second quantity of communication data includes: a start symbol, a communication party name, a separator, data, and an end symbol, wherein the communication party name includes the name of the sending node and the name of the receiving node, and the data is the source data without semantic encoding.

11. The knowledge base synchronization method as described in claim 2, characterized in that, The steps for the sending node to update its own semantic knowledge base according to the federated learning method include: All sending nodes exchange their respective semantic knowledge bases until each sending node obtains the semantic knowledge bases of all other sending nodes after y iterations, where y is a positive integer greater than or equal to 1. Each sending node uses an aggregation algorithm to merge all of its own semantic knowledge bases.

12. The knowledge base synchronization method as described in claim 11, characterized in that, The federated semantic knowledge base of sending node i needs to be updated for t rounds, where i and t are positive integers greater than or equal to 1. During the t-th round of federated semantic knowledge base update, sending node i obtains the model parameters of the semantic knowledge base at time t+1. The steps include: Step S1: The sending node i iterates the model parameters of its own semantic knowledge base at time t. This includes: constructing an end-to-end semantic knowledge base self-updating framework comprising a virtual sender and a virtual receiver; the end-to-end semantic knowledge base self-updating framework calculates the model parameters using the following formula after y iterations. : in, Indicates the update rate. The operator for calculating the gradient. Represents the Loss function. This represents the training dataset; Step S2: The sending node i adds its own ID i to its own semantic knowledge base name list. And transmit the list of semantic knowledge base names to other sending nodes. ; Step S3: The sending node i receives data from other sending nodes. List of semantic knowledge base names transmitted , This represents the set of other sending nodes; and the destination node... Transmit the list of names of the semantic knowledge base Remove the semantic knowledge base name list from the list. This includes all semantic knowledge base information other than the semantic knowledge base itself. The data structure of the semantic knowledge base information includes: knowledge base name. The amount of data used in this iteration of the semantic knowledge base Semantic knowledge base parameters ; Step S4: When other sending nodes A switch occurs or the list of semantic knowledge base names is mentioned. When the content changes, repeat steps S2 to S3 until the semantic knowledge base name list is reached. All sending nodes are included; the sending node i adds its own number i to its own acknowledgment information list. and to other sending nodes Transmit the confirmation information list ; Receive from other sending nodes Confirmation message of transmission And store it in the confirmation information list. ; Step S5: When the confirmation information list When acknowledgment information from all sending nodes is included, the following formula is used to aggregate all semantic knowledge bases to obtain the federated semantic knowledge base z: Where n represents the total number of semantic knowledge bases.

13. A knowledge base synchronization system for space-based semantic communication, characterized in that, include: Memory is used to store instructions that can be executed by the processor; A processor for executing the instructions to implement the knowledge base synchronization method as described in any one of claims 1-12.

14. A computer-readable medium storing computer program code, characterized in that, The computer program code implements the knowledge base synchronization method as described in any one of claims 1-12 when executed by a processor.

Images