Generative AI endogenous communication network architecture and hierarchical collaborative scheduling method

Abstract

The invention discloses a generative AI endogenous communication network architecture and a hierarchical collaborative scheduling method, and relates to the technical field of communication networks, the communication network architecture comprises an intention extraction layer, a mapping adaptation layer and a network execution layer, the intention extraction layer processes multi-modal data by using a generative AI model to obtain a semantic demand representation of a user, and the mapping adaptation layer carries out mapping on the semantic demand representation of the user; the mapping adaptation layer maps the semantic demand representation into a service quality index by using a generative AI model, maps a real-time network state of a heterogeneous network in the network execution layer into a network performance index, and generates a transmission scheduling strategy based on the service quality index and the network performance index, and the network execution layer works based on the transmission scheduling strategy, so that the network performance of the heterogeneous network is improved. End-to-end deterministic transmission scheduling of data packets of a service is implemented. According to the invention, efficient deterministic transmission scheduling of intelligent computing services in a 6G network can be realized.

Description

Technical Field

[0001] The present application relates to the field of communication network technology, and in particular to a communication network architecture endogenous to a generative AI (Artificial Intelligence) and a hierarchical collaborative scheduling method. Background Art

[0002] As society's requirements for networks continue to increase, 6G communication systems have become the focus of research by international industry, academia, and standards groups. Compared with 5G networks, 6G networks not only have significant improvements in performance indicators such as bandwidth and latency, but more importantly, they have achieved ubiquitous intelligence and deep integration of computing and networks. Artificial intelligence technology will be deeply embedded in the 6G network architecture as a core driving force, especially in large-scale intelligent access scenarios such as the Internet of Vehicles. 6G networks need to process and transmit massive amounts of high-dimensional multimodal information for intelligent computing services, which poses a huge challenge to traditional optimization algorithms.

[0003] The deterministic transmission scheduling of intelligent computing services in 6G networks faces multiple technical challenges. First, at the information processing level, it is necessary to solve the problem of unified feature extraction of multimodal data, which directly affects the efficiency of perceiving the network status. At the same time, it is also necessary to accurately understand and express the semantic needs of users. Secondly, the frequent interactions between terminal nodes and computing centers often span multiple heterogeneous networks, which makes the deterministic transmission scheduling of multimodal data more complicated. In the face of these challenges, generative artificial intelligence, with its unique generation and generalization capabilities, provides a new solution to the problem of efficient deterministic transmission scheduling in 6G scenarios.

[0004] In response to the above challenges, there is an urgent need to design a new communication network architecture with generative AI endogenous and a hierarchical collaborative scheduling method to support emerging intelligent computing services in 6G networks. Summary of the invention

[0005] The purpose of this application is to provide a generative AI-endogenous communication network architecture and a hierarchical collaborative scheduling method, which can realize efficient deterministic transmission scheduling of intelligent computing services in 6G networks and support emerging intelligent computing services in 6G networks.

[0006] To achieve the above objectives, this application provides the following solutions:

[0007] In a first aspect, the present application provides a generative AI endogenous communication network architecture, wherein the generative AI endogenous communication network architecture includes:

[0008] The intent extraction layer is used to process multimodal data using a generative AI model to obtain the user's semantic demand representation; the multimodal data is the user's demand data during the business execution of intelligent computing services in the 6G network;

[0009] A mapping adaptation layer, which is connected to the intent extraction layer and the network execution layer respectively, and is used to map the semantic demand representation into the service quality indicator using the generative AI model, map the real-time network status of the heterogeneous network in the network execution layer into the network performance indicator, and generate a transmission scheduling strategy based on the service quality indicator and the network performance indicator;

[0010] The network execution layer is used to work based on the transmission scheduling strategy to achieve end-to-end deterministic transmission scheduling of service data packets.

[0011] Optionally, the intent abstraction layer includes:

[0012] A multimodal processing module is used to extract features from multimodal data to obtain user data;

[0013] A generation processing module is used to infer user behaviors and preferences based on user data and predict and generate synthetic data;

[0014] The semantic demand processing module is used to extract, expand and compress semantics by combining user data and synthetic data to obtain the user's semantic demand representation;

[0015] Among them, both the generation processing module and the semantic requirement processing module adopt generative AI models.

[0016] Optionally, the multimodal data includes text, audio, pictures and location information. In this case, the multimodal processing module includes a CNN model, a ViT model and a GNN model. The CNN model is used to perform feature extraction on the text to obtain a first feature, the CNN model is used to perform feature extraction on the audio to obtain a second feature, the ViT model is used to perform feature extraction on the picture to obtain a third feature, and the GNN model is used to perform feature extraction on the location information to obtain a fourth feature. The first feature, the second feature, the third feature and the fourth feature are combined to form user data.

[0017] Optionally, the mapping adaptation layer includes:

[0018] Service requirement knowledge base, used to map semantic requirement representations into service quality indicators; service quality indicators include: basic performance indicators, reliability indicators, resource indicators and business characteristic indicators. Basic performance indicators include bandwidth, latency and business priority. Reliability indicators include packet loss rate and jitter. Resource indicators include computing resources and network resources. Business characteristic indicators include data packet size, traffic pattern and service duration.

[0019] A network resource knowledge base is used to map the real-time network status of heterogeneous networks in the network execution layer into network performance indicators; network performance indicators include: link status indicators, cache resource indicators, timing performance indicators and device capability indicators. Link status indicators include available transmission bandwidth and link utilization, cache resource indicators include queue length and cache utilization, timing performance indicators include current link delay and jitter range, and device capability indicators include port status;

[0020] Adaptation decision module, used for dynamically matching demand and resources based on service quality indicators and network performance indicators, and generating transmission scheduling strategies; the transmission scheduling strategies include: routing decision matrix, bandwidth allocation matrix and heterogeneous network configuration, the routing decision matrix includes the next hop selection of each flow at each node, the bandwidth allocation matrix includes the bandwidth allocated to each flow on each link, and the nodes are network nodes in the heterogeneous network;

[0021] Among them, generative AI models are deployed in both the service demand knowledge base and the network resource knowledge base, and the adaptation decision module uses an end-to-end deterministic transmission scheduling algorithm based on the diffusion model enhanced D3QN to generate a transmission scheduling strategy.

[0022] Optionally, the network execution layer includes a heterogeneous network, which includes a first edge network, a core network, and a second edge network connected in sequence, the first edge network and the second edge network both adopt time-sensitive networks, and the core network adopts a deterministic network; wherein time-aware shapers are deployed in the first edge network and the second edge network, and a time-aware shaper and a credit-based shaper are deployed in the core network.

[0023] Optionally, the network execution layer also includes:

[0024] The terminal processing module is connected to the mapping adaptation layer and is used to obtain the transmission scheduling strategy from the mapping adaptation layer in real time;

[0025] The global processing module is connected to the terminal processing module and is used to store the transmission scheduling strategy obtained by the terminal processing module for a long time and send it to the heterogeneous network in the storage order.

[0026] Optionally, the configuration of the heterogeneous network includes a gating list of each node in the first edge network and the second edge network and a gating list, an idle rate, and a sending rate of each node in the core network.

[0027] In a second aspect, the present application provides a hierarchical collaborative scheduling method for a generative AI endogenous communication network architecture, which is applied to the generative AI endogenous communication network architecture described in any one of the above, and the hierarchical collaborative scheduling method for the generative AI endogenous communication network architecture includes:

[0028] The intent extraction layer uses a generative AI model to process multimodal data to obtain the user's semantic demand representation;

[0029] The mapping adaptation layer uses a generative AI model to map semantic demand representations into service quality indicators, maps the real-time network status of heterogeneous networks in the network execution layer into network performance indicators, and generates transmission scheduling strategies based on service quality indicators and network performance indicators;

[0030] The network execution layer works based on the transmission scheduling strategy to achieve end-to-end deterministic transmission scheduling of service data packets.

[0031] Optionally, a transmission scheduling strategy is generated based on the service quality indicator and the network performance indicator, specifically including:

[0032] Standardize the service quality indicators and network performance indicators to obtain standardized data; the standardized data includes network topology, link load, queue status, flow characteristics and delay requirements;

[0033] The standardized data is used as the state vector, and an action vector is generated based on the state vector, and the action vector is used as the transmission scheduling strategy.

[0034] Optionally, generating an action vector based on the state vector specifically includes: taking the state vector as input and determining the action vector using a trained policy network;

[0035] Among them, the reward function used when training the trained policy network is:

[0036] R(s,a)=w 1 ·R delay +w 2 ·R jit +w 3 ·R thr -w 4 ·C vio ;

[0037] Among them, R(s, a) is the reward; w 1 is the first weight; R delay is the delay; 2 is the second weight; R jit is jitter; w 3 is the third weight; R thr is the throughput; w 4 is the fourth weight; C vip Penalty for violating service quality indicators.

[0038] According to the specific embodiments provided in this application, this application has the following technical effects:

[0039] The present application provides a communication network architecture and a hierarchical collaborative scheduling method endogenous to generative AI, including an intent extraction layer, a mapping adaptation layer, and a network execution layer. The intent extraction layer uses a generative AI model to process multimodal data to obtain a semantic demand representation of the user. The mapping adaptation layer uses a generative AI model to map the semantic demand representation to a service quality indicator, and maps the real-time network status of the heterogeneous network in the network execution layer to a network performance indicator, and generates a transmission scheduling strategy based on the service quality indicator and the network performance indicator. The network execution layer works based on the transmission scheduling strategy to achieve end-to-end deterministic transmission scheduling of service data packets. By integrating the generative AI model into the intent extraction layer and the mapping adaptation layer, the present application can solve the problem of unified feature extraction of multimodal data, the need to accurately understand and express the semantic needs of users, and solve the problem that the deterministic transmission scheduling of multimodal data becomes more complicated, thereby achieving efficient deterministic transmission scheduling of intelligent computing services in 6G networks and supporting emerging intelligent computing services in 6G networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 A schematic diagram of a new communication network architecture with generative AI endogenous to 6G networks provided in Example 1 of the present application.

[0042] Figure 2 A flowchart of the semantic requirement representation extraction method provided in Example 1 of the present application.

[0043] Figure 3 A flowchart of the method for mapping requirements and resources provided in Example 1 of the present application.

[0044] Figure 4 A flow chart of an end-to-end deterministic transmission scheduling algorithm based on a diffusion model-enhanced D3QN (Dueling Double Deep Q-network) provided in Example 1 of the present application.

[0045] Figure 5 A flow chart of a hierarchical collaborative scheduling method for a generative AI-native communication network architecture provided in Example 2 of the present application. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0047] Example 1

[0048] This embodiment provides a communication network architecture that is endogenous to a generative AI, such as Figure 1 As shown, the communication network architecture inherent to the generative AI includes: an intent extraction layer, a mapping adaptation layer, and a network execution layer.

[0049] The intent extraction layer is used to process multimodal data using a generative AI model to obtain the user's semantic demand representation. The multimodal data is the user's demand data during the business execution of intelligent computing services in the 6G network.

[0050] The mapping adaptation layer is connected to the intent extraction layer and the network execution layer respectively, and is used to use the generative AI model to map the semantic demand representation into service quality indicators, map the real-time network status of the heterogeneous network in the network execution layer into network performance indicators, and generate transmission scheduling strategies based on the service quality indicators and network performance indicators.

[0051] The network execution layer is used to work based on the transmission scheduling strategy to achieve end-to-end deterministic transmission scheduling of service data packets.

[0052] In this embodiment, the intent extraction layer is used to process various multimodal data, obtain user data, and fuse the actual user data and the predicted synthetic data to extract the user's semantic demand representation, thereby realizing the user's multimodal data processing and semantic demand representation extraction. The intent extraction layer deploys a multimodal processing module, a generation processing module, and a semantic demand processing module to achieve efficient extraction of the user's semantic demand representation, which is convenient for subsequent service demand mapping.

[0053] Specifically, the intent extraction layer includes: a multimodal processing module, a generation processing module and a semantic demand processing module.

[0054] The multimodal processing module is used to extract features from multimodal data to obtain user data.

[0055] The generation processing module is used to infer user behaviors and preferences based on user data and predict and generate synthetic data.

[0056] The semantic demand processing module is used to extract, expand and compress semantics by combining user data and synthetic data to obtain the user's semantic demand representation. By fusing actual user data (as the main input) and predicted synthetic data (as the auxiliary input), semantic extraction, expansion and compression are achieved to form an accurate semantic demand representation.

[0057] Among them, both the generation processing module and the semantic requirement processing module adopt generative AI models. The generative AI model adopted by the generation processing module can be LLM (Large Language Model), and the generative AI model adopted by the semantic requirement processing module can be LLM.

[0058] Among them, the multimodal processing module is used to process the multimodal data generated in the 6G application scenario, such as text, audio, picture, and location information, and uses a deep learning model to uniformly map data of different modes to the feature space to form a standardized user data representation to obtain user data. That is, in this embodiment, the multimodal data may include text, audio, pictures and location information. At this time, the multimodal processing module includes a CNN (Convolutional Neural Networks) model, a ViT (Vision Transformer) model and a GNN (Graph Neural Networks) model. The CNN model is used to extract features from text to obtain a first feature, the CNN model is used to extract features from audio to obtain a second feature, the ViT model is used to extract features from pictures to obtain a third feature, and the GNN model is used to extract features from location information to obtain a fourth feature. The first feature, the second feature, the third feature and the fourth feature form user data.

[0059] This embodiment integrates a generative AI model into the intent extraction layer to solve the problem of unified feature extraction of multimodal data and the problem of accurately understanding and expressing users' semantic needs at the information processing level.

[0060] In this embodiment, the mapping and adaptation layer generates a transmission scheduling strategy by mapping and adapting the demand and resources. The mapping and adaptation layer deploys two knowledge bases and an adaptation decision module to achieve accurate mapping and adaptation of service demand and network resources and strategy generation. The semantic demand representation and real-time network status are mapped to specific QoS (Quality of Service) indicators and network performance indicators through the two knowledge bases, and then the demand and resources are adapted through the adaptation decision module to generate a transmission scheduling strategy. The service demand knowledge base maps the semantic demand representation extracted by the intent extraction layer to specific QoS indicators with the help of LLM. An effective search space is defined through LLM, and the mapping path from the semantic demand representation to the QoS indicator is specified. The network resource knowledge base maps the real-time network status uploaded by the network execution layer to specific network performance indicators, such as link utilization, queue length, etc., which are used to quantify network performance. The adaptation decision module dynamically matches demand and resources according to QoS indicators and network performance indicators, that is, adapts demand and resources. The end-to-end deterministic transmission scheduling algorithm based on the diffusion model to enhance D3QN generates the optimal transmission scheduling strategy and sends it to the network execution layer, that is, the transmission scheduling strategy is subsequently deployed to each network node (i.e., device) of the network execution layer. This embodiment calls the end-to-end deterministic transmission scheduling algorithm based on the diffusion model to enhance D3QN, and dynamically controls the transmission routing and configuration of the data packet according to the adaptation of resources and demand. In this way, the resource allocation in the 6G network can be optimized, and the deterministic transmission scheduling of intelligent computing services can be realized.

[0061] Specifically, the mapping adaptation layer includes: a service demand knowledge base, a network resource knowledge base and an adaptation decision module.

[0062] The service requirement knowledge base is used to map semantic requirement representations into service quality indicators. The service quality indicators include basic performance indicators, reliability indicators, resource indicators, and business characteristic indicators. Basic performance indicators include bandwidth, latency, and business priority. Reliability indicators include packet loss rate and jitter. Resource indicators include computing resources and network resources. Business characteristic indicators include data packet size, traffic pattern, and service duration.

[0063] The network resource knowledge base is used to map the real-time network status of heterogeneous networks in the network execution layer into network performance indicators. The network performance indicators include: link status indicators, cache resource indicators, timing performance indicators and equipment capability indicators. Link status indicators include available transmission bandwidth and link utilization. Cache resource indicators include queue length and cache utilization. Timing performance indicators include current link delay and jitter range. Equipment capability indicators include port status.

[0064] The adaptation decision module is used to dynamically match demand and resources based on service quality indicators and network performance indicators, and generate a transmission scheduling strategy. The transmission scheduling strategy includes: a routing decision matrix, a bandwidth allocation matrix and the configuration of a heterogeneous network. The routing decision matrix includes the next hop selection of each flow at each node, and the bandwidth allocation matrix includes the bandwidth allocated to each flow on each link. The node is a network node in a heterogeneous network.

[0065] Among them, generative AI models are deployed in both the service demand knowledge base and the network resource knowledge base. The generative AI models deployed in the service demand knowledge base and the network resource knowledge base can both be LLMs. The adaptation decision module uses an end-to-end deterministic transmission scheduling algorithm based on the diffusion model enhanced D3QN to generate a transmission scheduling strategy. The diffusion model belongs to a generative AI model.

[0066] In this embodiment, the network execution layer provides the required physical network equipment, and transmits the data packet deterministically according to the transmission scheduling strategy of the mapping adaptation layer, combined with the deterministic transmission mechanism, to achieve end-to-end deterministic transmission scheduling of the data packet. The network execution layer includes: terminal LLM, global LLM and heterogeneous network. The terminal user downloads the terminal LLM from the cache server deployed at the edge of the network, personalizes and initializes it, and the terminal LLM obtains the transmission scheduling strategy from the mapping adaptation layer. The global LLM stores the transmission scheduling strategy aggregated by the terminal LLM as long-term memory data, and optimizes the transmission scheduling strategy through text learning and model fine-tuning to obtain an optimized transmission scheduling strategy. The heterogeneous network operation is subsequently controlled based on the optimized transmission scheduling strategy.

[0067] Heterogeneous networks include edge networks and core networks. The edge networks use time-sensitive networking (TSN), and the core networks use deterministic networking (DetNet). TSN deploys a unified time synchronization mechanism and scheduling list in the local area network, enabling all devices in the network to send and receive data packets according to the preset timing, thereby achieving controllability of data transmission delay in the local area network. DetNet reserves dedicated network resources, including bandwidth and queues, for critical business traffic in the wide area network, thereby ensuring its end-to-end deterministic transmission performance and reliability requirements.

[0068] Deterministic transmission mechanisms, such as TimeAwareness Shaper (TAS) and Credit-based Shaping (CBS), are deployed in edge networks and core networks. The gating mechanism is used to meet the transmission delay and jitter requirements of high real-time services at a smaller time granularity. TAS adopts a time-sensitive gating scheduling mechanism to control the transmission state of the packet queue through the time dimension. According to the pre-configured gate control list (GCL), the queue is periodically switched on and off: when the queue is in the on state, the packet is allowed to be transmitted; when the queue is in the off state, the packet will remain waiting in the queue until the state is switched to on. After the gate control list is determined, the scheduling mechanism will be executed cyclically according to the set time period. CBS controls the transmission scheduling of data by assigning a credit value to each queue. The credit value of each queue will change with time, and the rate of increase and decrease has corresponding parameter constraints. The credit-based control mode is used to reserve bandwidth for different priority flows and ensure the upper limit of transmission delay.

[0069] TSN achieves deterministic transmission within the local area network through the TAS mechanism. DetNet ensures the end-to-end transmission performance of key business traffic in the wide area network through the TAS and CBS fusion mechanism, and meets high real-time business needs through refined gating scheduling in the time dimension.

[0070] Specifically, the network execution layer includes a heterogeneous network, and the heterogeneous network includes a first edge network, a core network, and a second edge network connected in sequence. The first edge network and the second edge network both adopt a time-sensitive network, and the core network adopts a deterministic network. Among them, a time-aware shaper is deployed in the first edge network and the second edge network, and a time-aware shaper and a credit-based shaper are deployed in the core network.

[0071] At this time, in the transmission scheduling strategy, the configuration of the heterogeneous network includes the gating list of each node in the first edge network and the second edge network and the gating list, idle rate and sending rate of each node in the core network.

[0072] The network execution layer also includes: a terminal processing module and a global processing module.

[0073] Terminal processing module (i.e. Figure 1 The terminal LLM in the mapping adaptation layer is connected to the mapping adaptation layer to obtain the transmission scheduling strategy from the mapping adaptation layer in real time.

[0074] Global processing module (i.e. Figure 1The global LLM in the terminal processing module is connected to the terminal processing module and is used to store the transmission scheduling strategy obtained by the terminal processing module for a long time and send it to the heterogeneous network in the storage order.

[0075] Based on the above-mentioned new communication network architecture endogenous to generative AI, this embodiment further provides a hierarchical collaborative scheduling method, which is a hierarchical collaborative intelligent method, including: semantic demand representation extraction, demand and resource mapping, and end-to-end deterministic transmission scheduling algorithm based on diffusion model enhanced D3QN.

[0076] (1) Extraction of semantic requirement representation.

[0077] Semantic requirement representation extraction is the function performed by the intent extraction layer, such as Figure 2 As shown, the method includes: (a) using a distributed data acquisition system to collect text, audio, picture and location information data in 6G application scenarios in real time to obtain the user's original multimodal data; (b) establishing a deep learning model, including a CNN model for processing time series data, a ViT model for processing image data, and a GNN model for processing graph structure data, and using different deep learning models to extract features from the generated multimodal data: the CNN model extracts features of time series data (such as text and audio), the ViT model extracts features of image data (such as pictures), and the GNN model extracts features of graph structure data (such as location information). The extracted features are used as unified tags, that is, the collected multimodal data such as text, audio, picture and location information are input into different deep learning models for feature extraction. The feature extraction process is specifically as follows: the time series data is input into the CNN model to extract time series pattern features through convolution and pooling operations, and the image data is input into the ViT The model extracts visual features through patch segmentation and multi-head self-attention layer, inputs graph structure data into the GNN model, extracts spatial relationship features through the graph attention layer, generates a unified serialized token representation, and obtains user data; (c) Based on user data, LLM predicts a part of user behavior data and preference data as synthetic data. When predicting synthetic data, multimodal data is unified into serialized token representation, the distribution of serialized token representation is analyzed, and then new samples that conform to this distribution are generated, thereby realizing the prediction of user behavior data and preference data, so as to generate synthetic data based on user data to assist in extracting user semantic needs; (d) Synthetic data is combined with user data and input into LLM trained by local database to reason about the data, that is, user data and synthetic data are directly input into LLM to realize semantic reasoning and obtain semantic demand representation. LLM not only extracts the semantics of data, but also expands and compresses semantics according to context.

[0078] (2) Mapping of requirements and resources.

[0079] The mapping of demand and resources is part of the functions completed by the mapping adaptation layer, including: (a) The mapping of demand and resources is completed by a dual knowledge base architecture consisting of a service demand knowledge base and a network resource knowledge base. Both knowledge bases define effective search spaces through LLM and indicate precise mapping paths; (b) The service demand knowledge base uses LLM to accurately map semantic business requirements (i.e., semantic demand representation) into specific QoS indicators, such as the quantitative representation of bandwidth requirements, the numerical mapping of delay constraints, the hierarchical definition of business priorities, etc.; (c) The network resource knowledge base uses LLM to map real-time network status into network performance indicators, such as key standardized resource status representation indicators such as link utilization and queue length.

[0080] like Figure 3 As shown, the process of constructing two databases in this embodiment is as follows: (a) Service demand knowledge base construction: Initially standardize the service demand into four indicators: basic performance indicators (bandwidth, delay, service priority), reliability indicators (packet loss rate, jitter), resource indicators (computing resources, network resources), and service feature indicators (packet size, traffic pattern, service duration); (b) Network resource knowledge base construction: Initially standardize the resource status into four indicators: link status indicators (available transmission bandwidth, link utilization), cache resource indicators (queue length, cache utilization), timing performance indicators (current link delay, jitter range), and device capability indicators (port status); (c) Knowledge base fine-tuning and updating: On the basis of defining the initial standardized indicators, fine-tune them in combination with different demand scenarios (such as high real-time services, large bandwidth services) and network types (edge ​​networks, core networks). The data mainly consists of historical data and real-time feedback data, and the indicator content of the knowledge base is continuously updated based on this. Fine-tuning means using some specific data to train the LLM in the knowledge base on the basis of the initial standardized indicators, so as to expand the indicators of the knowledge base. For the service demand knowledge base, its specific data refers to tasks with different requirements. For the network resource knowledge base, its specific data refers to different networks, such as TSN networks and DetNet networks. The fine-tuning principles in different scenarios are the same, but the data used are different; (d) Accurate mapping of requirements and resources: Based on the fine-tuned knowledge base, the semantic requirement representation is mapped into quantitative QoS indicators, and the real-time network status is mapped into standardized network performance indicators to achieve accurate mapping of requirements and resources. It should be noted that after the knowledge base is established, the mapping is like looking up a dictionary. For example, if the user requires smoothness, the knowledge base directly gives the precise requirement delay of 50ms.

[0081] (3) Enhance the end-to-end deterministic transmission scheduling algorithm of D3QN based on the diffusion model.

[0082] The end-to-end deterministic transmission scheduling algorithm based on the diffusion model enhanced D3QN, that is, part of the functions completed by the mapping adaptation layer, is used to generate the transmission scheduling strategy, including: (a) the data abstraction module receives QoS indicators and network performance indicators, and standardizes them into a specific data structure format to obtain standardized data; (b) the standardized data is input into the diffusion agent, and the QoS indicators and network performance indicators are integrated into a state vector; (c) the diffusion agent regards the current network resource state as noise, iteratively denoises according to the QoS indicators, and generates an action vector (i.e., a deterministic transmission scheduling strategy); (d) the data abstraction module configures the transmission scheduling strategy to the data plane, completes the deterministic transmission scheduling of data packets, and obtains corresponding rewards based on delay, jitter and throughput.

[0083] like Figure 4 As shown, based on QoS indicators and network performance indicators, the training process of the end-to-end deterministic transmission scheduling algorithm of D3QN enhanced by the diffusion model specifically includes the following steps.

[0084] (1) Data structure construction: The data abstraction module receives QoS indicators and network performance indicators and standardizes them into a specific data structure format, namely, the nth state vector o n , to prepare for the subsequent generation of the state vector set S.

[0085] (2) State vector integration: The diffusion agent integrates the state vectors and defines the state vector set as S = {o 1 , o 2 , ..., o N}, where o n is the nth state vector, and N is the number of state vectors. n =(G, L, Q, F, T), G is the network topology, L is the link load, Q is the queue state, F is the flow feature, T is the delay requirement, G is represented by a graph structure, L and T can be represented by a single digit respectively, Q and F are in matrix form, G, Q are network performance indicators, F and T are service quality indicators, and the queue state matrix represents the state (i.e., length and priority) of the jth queue of node i, where I is the number of nodes and J is the number of queues. The queue state matrix consists of two parameters: length and priority. For example, if the current length of queue 2 of node 1 is 80 and the priority is 2, then the matrix element q 1，2 =(80,2), flow feature matrix f k is the flow feature of the kth flow, K is the number of flows, f k =(s k , d k ,qos k ) represents the source node sk , destination node d k and the QoS requirement of the kth flow qos k .

[0086] (3) Scheduling strategy generation: The diffusion agent generates an action vector based on each state vector. Specifically, for a state vector, the current network resource state is regarded as noise, and iterative denoising is performed according to service requirements to generate configuration actions, that is, generate action vectors. The action space is defined as A = {a 1 , a 2 , ..., a M}, where a m is the mth action vector, and M is the number of action vectors. m =(R, B, C tsn , C det ), is the routing decision matrix, represents the next hop selection of flow k at node i, is the bandwidth allocation matrix, represents the bandwidth allocated to flow k on link c, C is the number of links, It is the configuration of TSN domain, GCL i represents the gating list of node i, It is the configuration of DetNet domain, ID i and SD i They represent the idleslope (i.e., idle rate) and sendslope (i.e., sending rate) of node i respectively.

[0087] The action vector is defined by the reinforcement learning algorithm. The state vector is input, and the algorithm learns the optimal strategy to generate the action vector. The above matrix is ​​the action generated by the algorithm according to the state, which is used to configure the core network and edge network. The function is to realize the deterministic transmission and forwarding of data packets and control the routing and forwarding order of data packets. That is, the action vector is the configuration to realize the deterministic transmission scheduling of data packets. For example, GCL is the gate list of TAS, which can control the switch of each gate and determine whether the data packet is transmitted. The transmission of data packets is controlled by various tables. The routing decision matrix controls the next node of the data packet, and the gate list GCL controls the transmission order of the data packet.

[0088] (4) Configuration deployment evaluation: The data abstraction module configures the scheduling decision to the data plane, completes the deterministic transmission scheduling of data packets, and obtains corresponding rewards based on delay, jitter and throughput. The reward function is defined as R(s, a) = w i ·R delay +w 2 ·R jit +w3 ·R thr -w 4 ·C vio , w l , The weights to be assigned are considered in terms of delay, jitter, and throughput. At the same time, the QoS indicators are ensured to be met by constraining violation penalties. delay is the delay, R jit is jitter, R thr is the throughput, C vio Penalty for violating service quality indicators.

[0089] For the generated state vector and action vector, the corresponding rewards are calculated, and the state vector, action vector and reward are used to update the policy network to obtain a trained policy network. The action vector can be generated by directly inputting the state vector. That is, the transmission scheduling strategy can be generated by using the end-to-end deterministic transmission scheduling algorithm based on the diffusion model to enhance D3QN. It acts on the core network and edge network in the network execution layer to realize the scheduling, transmission and routing planning of data packets in the network.

[0090] The present embodiment discloses a new communication network architecture endogenous to generative AI (abbreviated as GenNet) and a hierarchical collaborative scheduling method. The new communication network architecture endogenous to generative AI includes: an intent extraction layer, a mapping adaptation layer and a network execution layer. The intent extraction layer processes various multimodal data to realize the extraction of user semantic demand representation. The mapping adaptation layer generates a transmission scheduling strategy through mapping and adaptation of demand and resources. The network execution layer provides the required physical network equipment. According to the transmission scheduling strategy of the mapping adaptation layer, combined with the deterministic transmission mechanism, end-to-end deterministic transmission scheduling of data packets is realized. The hierarchical collaborative scheduling method includes: semantic demand representation extraction, demand and resource mapping, and end-to-end deterministic transmission scheduling algorithm based on diffusion model enhanced D3QN. The new communication network architecture and hierarchical collaborative scheduling method provided in this embodiment improve computing efficiency, ensure transmission delay, effectively promote intelligent resource management, and support emerging intelligent computing services in 6G networks.

[0091] In the first aspect, this embodiment uses the idea of ​​ubiquitous intelligence to propose a new communication network architecture with generative AI endogenous to 6G networks to support the low latency requirements of 6G network intelligent computing services. The architecture includes three layers: intent extraction layer, mapping adaptation layer, and network execution layer. The layers collaborate to promote intelligent resource management and support emerging intelligent computing services in 6G networks. In the second aspect, a layered collaborative scheduling method is provided, including semantic demand representation extraction, demand and resource mapping, and an end-to-end deterministic transmission scheduling algorithm based on diffusion model-enhanced D3QN, which supports more efficient intelligent computing service scheduling in 6G networks and has the following effects.

[0092] (1) This embodiment provides a new communication network architecture for 6G network generative AI endogenous, so that each layer and each function has independent or collaborative deterministic guarantee capabilities. Through the demand transmission and information sharing between the three layers of the intent extraction layer, mapping adaptation layer and network execution layer, the generation and generalization capabilities of generative AI are fully utilized to improve computing efficiency, ensure transmission delay, promote intelligent resource management, and support emerging intelligent computing services in 6G networks.

[0093] (2) The hierarchical collaborative scheduling method provided in this embodiment serves the three-layer network architecture mentioned in the new communication network architecture. The mapping of demand and resources solves the problem of accurate matching of user semantic demand and network resources. Through the dual knowledge base architecture and the deep semantic understanding ability of LLM, dynamic adaptation of service demand to network resources can be achieved, ensuring the efficiency of resource allocation and the reliability of service quality. The end-to-end deterministic transmission scheduling algorithm based on the diffusion model-enhanced D3QN realizes intelligent scheduling decisions in a cross-domain network environment, ensuring the end-to-end deterministic transmission performance of traffic in heterogeneous networks.

[0094] Example 2

[0095] This embodiment provides a hierarchical collaborative scheduling method for a generative AI endogenous communication network architecture, which is applied to the generative AI endogenous communication network architecture described in Example 1, such as Figure 5 As shown, the hierarchical collaborative scheduling method of the communication network architecture endogenous to the generative AI includes:

[0096] S1: The intent extraction layer uses a generative AI model to process multimodal data to obtain the user's semantic demand representation.

[0097] S2: The mapping adaptation layer uses a generative AI model to map semantic demand representations into service quality indicators, maps the real-time network status of heterogeneous networks in the network execution layer into network performance indicators, and generates transmission scheduling strategies based on service quality indicators and network performance indicators.

[0098] S3: The network execution layer works based on the transmission scheduling strategy to achieve end-to-end deterministic transmission scheduling of service data packets.

[0099] Among them, generating a transmission scheduling strategy based on service quality indicators and network performance indicators specifically includes: standardizing the service quality indicators and network performance indicators to obtain standardized data, the standardized data includes network topology, link load, queue status, flow characteristics and delay requirements; using the standardized data as a state vector, and generating an action vector based on the state vector, and using the action vector as a transmission scheduling strategy.

[0100] The action vector is generated based on the state vector, specifically including: taking the state vector as input and using the trained policy network to determine the action vector. The reward function used when training the trained policy network is:

[0101] R(s,a)=w 1 ·R delay +w 2 ·R jit +w 3 ·R thr -w 4 ·C vio ;

[0102] Among them, R(s, a) is the reward; w 1 is the first weight; R delay is the delay; 2 is the second weight; R jit is jitter; w 3 is the third weight; R thr is the throughput; w 4 is the fourth weight; C vip Penalty for violating service quality indicators.

[0103] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0104] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0105] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A communication network architecture endogenous to generative AI, characterized in that: The communication network architecture inherent to the generative AI includes: The intent extraction layer is used to process multimodal data using a generative AI model to obtain the user's semantic demand representation; the multimodal data is the user's demand data during the business execution of intelligent computing services in the 6G network; A mapping adaptation layer, which is connected to the intent extraction layer and the network execution layer respectively, and is used to map the semantic demand representation into the service quality indicator using the generative AI model, map the real-time network status of the heterogeneous network in the network execution layer into the network performance indicator, and generate a transmission scheduling strategy based on the service quality indicator and the network performance indicator; The network execution layer is used to work based on the transmission scheduling strategy to achieve end-to-end deterministic transmission scheduling of service data packets.

2. The generative AI endogenous communication network architecture according to claim 1, characterized in that: The intent abstraction layer includes: A multimodal processing module is used to extract features from multimodal data to obtain user data; A generation processing module is used to infer user behaviors and preferences based on user data and predict and generate synthetic data; The semantic demand processing module is used to extract, expand and compress semantics by combining user data and synthetic data to obtain the user's semantic demand representation; Among them, both the generation processing module and the semantic requirement processing module adopt generative AI models.

3. The generative AI endogenous communication network architecture according to claim 2, characterized in that: Multimodal data includes text, audio, pictures and location information. At this time, the multimodal processing module includes a CNN model, a ViT model and a GNN model. The CNN model is used to extract features from the text to obtain a first feature, the CNN model is used to extract features from the audio to obtain a second feature, the ViT model is used to extract features from the picture to obtain a third feature, and the GNN model is used to extract features from the location information to obtain a fourth feature. The first feature, the second feature, the third feature and the fourth feature are combined to form user data.

4. The generative AI endogenous communication network architecture according to claim 1, characterized in that: The mapping adaptation layer includes: Service requirement knowledge base, used to map semantic requirement representations into service quality indicators; service quality indicators include: basic performance indicators, reliability indicators, resource indicators and business characteristic indicators. Basic performance indicators include bandwidth, latency and business priority. Reliability indicators include packet loss rate and jitter. Resource indicators include computing resources and network resources. Business characteristic indicators include data packet size, traffic pattern and service duration. A network resource knowledge base is used to map the real-time network status of heterogeneous networks in the network execution layer into network performance indicators; network performance indicators include: link status indicators, cache resource indicators, timing performance indicators and device capability indicators. Link status indicators include available transmission bandwidth and link utilization, cache resource indicators include queue length and cache utilization, timing performance indicators include current link delay and jitter range, and device capability indicators include port status; Adaptation decision module, used for dynamically matching demand and resources based on service quality indicators and network performance indicators, and generating transmission scheduling strategies; the transmission scheduling strategies include: routing decision matrix, bandwidth allocation matrix and heterogeneous network configuration, the routing decision matrix includes the next hop selection of each flow at each node, the bandwidth allocation matrix includes the bandwidth allocated to each flow on each link, and the nodes are network nodes in the heterogeneous network; Among them, generative AI models are deployed in both the service demand knowledge base and the network resource knowledge base, and the adaptation decision module uses an end-to-end deterministic transmission scheduling algorithm based on the diffusion model enhanced D3QN to generate a transmission scheduling strategy.

5. The generative AI endogenous communication network architecture according to claim 4, characterized in that: The network execution layer includes a heterogeneous network, which includes a first edge network, a core network, and a second edge network connected in sequence. The first edge network and the second edge network both adopt time-sensitive networks, and the core network adopts a deterministic network; wherein time-aware shapers are deployed in the first edge network and the second edge network, and a time-aware shaper and a credit-based shaper are deployed in the core network.

6. The generative AI endogenous communication network architecture according to claim 5, characterized in that: The network execution layer also includes: The terminal processing module is connected to the mapping adaptation layer and is used to obtain the transmission scheduling strategy from the mapping adaptation layer in real time; The global processing module is connected to the terminal processing module and is used to store the transmission scheduling strategy obtained by the terminal processing module for a long time and send it to the heterogeneous network in the storage order.

7. The generative AI endogenous communication network architecture according to claim 5, characterized in that: The configuration of the heterogeneous network includes a gating list of each node in the first edge network and the second edge network and a gating list, an idle rate, and a sending rate of each node in the core network.

8. A hierarchical collaborative scheduling method for a generative AI endogenous communication network architecture, applied to the generative AI endogenous communication network architecture described in any one of claims 1-7, characterized in that: The hierarchical collaborative scheduling method of the communication network architecture endogenous to the generative AI includes: The intent extraction layer uses a generative AI model to process multimodal data to obtain the user's semantic demand representation; The mapping adaptation layer uses a generative AI model to map semantic demand representations into service quality indicators, maps the real-time network status of heterogeneous networks in the network execution layer into network performance indicators, and generates transmission scheduling strategies based on service quality indicators and network performance indicators; The network execution layer works based on the transmission scheduling strategy to achieve end-to-end deterministic transmission scheduling of service data packets.

9. The hierarchical collaborative scheduling method of the generative AI endogenous communication network architecture according to claim 8, characterized in that: Generate a transmission scheduling strategy based on service quality indicators and network performance indicators, including: Standardize the service quality indicators and network performance indicators to obtain standardized data; the standardized data includes network topology, link load, queue status, flow characteristics and delay requirements; The standardized data is used as the state vector, and an action vector is generated based on the state vector, and the action vector is used as the transmission scheduling strategy.

10. The hierarchical collaborative scheduling method of the generative AI endogenous communication network architecture according to claim 9, characterized in that: Generating an action vector based on the state vector specifically includes: taking the state vector as input and using a trained policy network to determine the action vector; Among them, the reward function used when training the trained policy network is: R(s,a)=w1·R delay +w2·R jit +w3·R thr -w4·C vio ; Among them, R(s,a) is the reward; w1 is the first weight; R delay is the delay; w2 is the second weight; R jit is jitter; w3 is the third weight; R thr is the throughput; w4 is the fourth weight; C vio Penalty for violating service quality indicators.

Images