Quantum network data transmission method and system for multi-quantum-bit communication

Abstract

The invention relates to a quantum network data transmission method and system for multi-quantum-bit communication, and belongs to the technical field of quantum communication and network. According to the method, an E2E entanglement generation mechanism and a specific noise interference mechanism of multi-quit communication in the quantum network are established, a multi-quit entanglement quality evaluation model, an E2E communication time delay model and a request data transmission model are modeled, and a centralized and distributed collaborative optimization network response model is designed. According to the system, a proposed multi-quit entanglement communication framework is trained and deployed on quantum network models of different scales, the system is used for processing transmission requirements of heterogeneous requests in batches, and stable and efficient data transmission service can be provided for concurrent response of batch requests in a large-scale quantum network. According to the invention, resource configuration and transmission quantification can be carried out on entanglement establishment and entanglement transmission, the response time delay of a large-scale quantum network to a multi-request transmission demand is reduced, and quantum resources are reasonably used.

Description

Technical Field

[0001] The present invention relates to a quantum network data transmission method and system for multi-qubit communication, belonging to the field of quantum communication and network technology. Background Art

[0002] Quantum mechanisms, with their inherent privacy protection and high-speed transmission properties, are revolutionizing traditional communications. With significant global support, unprecedented progress has been made in quantum applications such as quantum key distribution, distributed quantum computing, and quantum nuclear magnetic resonance. As a vehicle for quantum information transmission, quantum networks promise to efficiently transmit vast amounts of data, ensuring high-speed, large-scale, and latency-free communication, thereby promoting the widespread adoption of massive quantum applications. High-dimensional entanglement, thanks to its efficient data-carrying capacity and noise immunity, is seen as a potential solution.

[0003] In quantum networks, quantum bit (qubit) data transmission relies on quantum entanglement, a medium-free communication channel established between two quantum nodes. Long-distance end-to-end (E2E) entanglement is a key component of large-scale quantum networks that can respond to massive quantum requests. However, quantum decoherence, caused by the mutual interference noise between quantum states and their surroundings, introduces further uncertainty into the establishment of long-distance entanglement. Quantum decoherence degrades the quality of quantum entangled links and weakens their data transmission capabilities, resulting in unstable connections and numerous quantum information encoding errors, especially in long-distance E2E entanglement. Furthermore, the limited qubit resource storage of quantum nodes restricts the generation of large numbers of entangled links, making it even more difficult for quantum networks to meet the data transmission needs of batch requests.

[0004] Therefore, rationally routing E2E entangled paths and designing efficient qubit resource allocation strategies to alleviate the impact of quantum decoherence and thus meet the massive data transmission needs of large-scale quantum applications is a technical problem that needs to be urgently studied in quantum network communications, known as the Entanglement Routing and Qubit Provisioning (ERQP) problem.

[0005] Network modeling of quantum networks is the first step in addressing the ERQP problem. Currently, existing network modeling maximizes network throughput by establishing as many single-qubit end-to-end entanglements as possible. However, this single-qubit entangled state can only transmit one qubit at a time. Furthermore, single-qubit entangled states are more susceptible to noise interference from quantum decoherence. A type of quantum multi-qubit entanglement offers significant improvements in noise resistance. This type of entanglement forms multi-qubit entanglement between two network nodes through the weighted linear superposition of multiple qubit states and has been shown to possess even greater noise immunity. Compared to single-qubit entanglement, multi-qubit entanglement generated by the superposition of multiple qubit states has a higher-dimensional parameter space. When the entangled state is subject to decoherence noise, changes in some coefficients have a smaller impact on the overall entanglement. It offers greater compatibility with quantum coding errors such as qubit flips and phase flips, enabling more efficient data transmission than single-qubit entanglement. Therefore, how to consider this type of entanglement to resist the effects of decoherence and meet the needs of large-scale data transmission is an issue that needs to be considered in building a more efficient quantum network.

[0006] The second aspect is optimization methods. Existing research primarily focuses on optimizing network performance through centralized approaches, such as the extended Dijkstra algorithm, randomized rounding, and opportunistic methods. Running these algorithms, especially when running concurrent iterations or performing one-shot solutions in large-scale networks, incurs excessive communication overhead due to the frequent acquisition of global information. Other research has employed fully distributed approaches to reduce the communication burden and achieve local privacy protection, such as graph-based distance routing and neighbor-based control. However, these local-perspective algorithms often suffer from performance degradation due to the limited network information they input. Furthermore, there are differences in objectives between centralized and distributed control. Global objectives aim to serve quantum users and applications, ensuring that requested data transmission needs are quickly met, while local objectives focus on node benefits, such as resource utilization and energy consumption. Therefore, designing a flexible transmission framework that balances communication overhead and network performance while optimizing both global and local objectives is a pressing challenge in realizing an effective quantum network.

[0007] In summary, there is an urgent need to develop a quantum network data transmission technology for multi-qubit entangled transmission that can reasonably schedule and configure network resources and comprehensively consider optimization goals, so as to ensure the timely and low-cost response of quantum networks to massive quantum requests to the greatest extent possible. Summary of the Invention

[0008] This invention aims to address technical challenges in responding to multi-request transmissions and optimizing network resources in large-scale quantum networks. It innovatively proposes a quantum network data transmission method and system for multi-qubit communication, enabling resource allocation and transmission quantization for entanglement establishment and transmission. The invention aims to reduce the response latency of large-scale quantum networks to multi-request transmissions and rationally utilize quantum resources.

[0009] The technical solutions adopted by the present invention are as follows.

[0010] On the one hand, the present invention proposes a quantum network data transmission method for multi-qubit communication, including quantum network modeling, collaborative optimization algorithm design and construction of a transmission system framework.

[0011] Quantum network modeling introduces a multi-qubit high-dimensional entangled state representation through a physical experimental model that realizes multi-qubit entanglement. This high-dimensional state is used to define and quantify quantum network model components, including entanglement link quality, entanglement transmission delay, and entanglement transmission data volume. The following steps are involved:

[0012] Step 1: High-dimensional quantum state representation of multi-qubit entanglement.

[0013] A quantum node contains multiple quantum registers that are designed to generate multi-qubit entanglement. Each register contains d single-qubit (quantum bit) photon pairs, which are interconnected through internal entanglement. By packaging and sending these d qubits of entanglement simultaneously, the register completes an entanglement transmission and forms a d-dimensional entangled quantum state between the sending and receiving nodes.

[0014] In quantum experiments, due to the coupling entanglement between the entangled state and the surrounding particles, each transmission of the register may not successfully establish entanglement. In order to ensure the successful establishment of entanglement, the frequency of successful entanglement establishment is estimated through multiple entanglement establishment attempts. Assume that sending f times can ensure that the entangled state is successfully established at least once, and use the W quantum state form to represent the d-dimensional entangled quantum state

[0015]

[0016] Where f is the number of transmissions; d is the entanglement dimension; p is the probability of obtaining the state |1> when measuring a single qubit; |e ij > indicates that in the i·j-bit quantum state W, only the (i·d+j)th bit is in state |1>, while the rest are all in state |0>. This entangled quantum state consumes f·d qubits of the sending node's resources.

[0017] Step 2: High-dimensional quantum state representation of network model components.

[0018] Step 2.1: End-to-end entanglement quality quantification.

[0019] We introduce a collapse decoherence noise model affected by transmission distance and use the entanglement fidelity metric to characterize entanglement quality. In the pure state of quantum entanglement (unaffected by decoherence noise), the entanglement quality is 1, and the entanglement link is guaranteed to be successfully established. For a request with an initiating node, the fidelity of the end-to-end entanglement path is the product of the fidelities of each individual link in the path.

[0020] Step 2.2: Model end-to-end communication delay.

[0021]

[0022] The current entangled communication delay refers to the communication delay between the end-to-end entangled transmission quantum information (entangled link) and the information measurement result (traditional communication link) at the current time t. The historical cumulative delay refers to the cumulative communication delay to the previous time t-1.

[0023] Step 2.3 quantifies the amount of transmitted data.

[0024] The amount of data transmitted is quantified in the form of one entangled link in the end-to-end entangled path of the request. ij , the amount of data transmitted on the link for:

[0025]

[0026] Step 3: Joint optimization objective modeling.

[0027] Based on the network component model established in steps 2.1, 2.2, and 2.3, the efficiency optimization goal of the quantum network responding to multiple requests is further established.

[0028] The collaborative decision-making between the centralized controller and the distributed agents at each moment is achieved by defining a joint optimization goal, where:

[0029]

[0030] That is to maximize the information flow of the network at the current moment.

[0031]

[0032] The collaborative optimization algorithm utilizes a centralized and distributed approach to quantum resource allocation optimization. It assigns the decision-making of the quantum network's response to multiple transmission requests to a central controller and distributed agents, respectively, establishing an optimization dependency between the centralized controller and multiple agents. The centralized controller uses Integer Linear Programming (ILP) modeling and the Brand-and-Bound (B&B) approximation algorithm to handle the end-to-end entanglement path selection process for each request.

[0033] Step 1: ILP modeling of entangled path selection.

[0034] The entanglement path selection aims to maximize the data pre-transmission capacity of the network while considering the entanglement quality (fidelity), entanglement delay and node memory constraints of each request.

[0035] First, after the distributed agents initialize their respective policy networks, the agents on each node input the initial state, that is, the local observed network information. Output the qubit resources pre-allocated by this node for each request

[0036] After obtaining the pre-configured resources of each node for each request, the controller Each entangled connection in Calculate the amount of pre-transmitted data for request r through the link Then calculate the optimization goal of ILP Fidelity constraints Entanglement delay constraint and node memory constraints Where R represents the quantum request set, P r represents the candidate path set for quantum request r, represents the kth candidate path of quantum request r. is a Boolean variable, It means that during the t-th entanglement transmission process, r is requested to select its k-th candidate path to transmit data. If it is 0, it means no selection. It is a Boolean constant. A value of 1 indicates that the kth candidate path of request r passes through network node n, and a value of 0 indicates that it does not pass through network node n. represents the amount of data transmitted from node n to node n+1 on the selected candidate path for request r during the tth entanglement transmission. e represents the natural base. F thr Represents the fidelity threshold of the entanglement transmission path. Only when the fidelity of the path is greater than the threshold F thrThe transmission path is valid only when Represents the length of the k-th candidate path for request r. D represents the cumulative transmission delay of request r in the first t-1 entanglement transmissions. thr Indicates the transmission path delay threshold. It represents the amount of data stored in node n by request r during the tth entanglement transmission. C represents the amount of data transmitted from node n-1 to node n on the selected candidate path during the tth entanglement transmission request r. n Indicates the resource performance upper limit of network node n.

[0037] Step 2: Approximate solution of entanglement path selection.

[0038] Step 2.1: Linear relaxation of the ILP.

[0039] Relax the binary variable of the ILP problem modeled in step 1 (selected as 1, not selected as 0) into a continuous variable between [0, 1], and use the solver to solve it in linear time to obtain the optimal solution of the relaxed problem in the continuous space.

[0040] Step 2.2: B&B algorithm at link level.

[0041] Use the link-level B&B algorithm to approximate the relaxed fractional solution to the integer solution And prove the approximation ratio. The algorithm consists of three parts: direct approximation, path decomposition, and target constraint pruning.

[0042] First, The fractional solution of is directly approximated to the corresponding entanglement path k, and the remaining Push them into the processing queue in descending order. Out of the queue, the variable is applied to all candidate paths of request r To filter:

[0043] Whether the target path segment P satisfies the three constraints;

[0044] Does the target path segment P exist? middle.

[0045] If both of the above are true, then Enter the qualified queue P'.

[0046] After that, find the shortest overlapping path segment (starting from the first node) of all qualified candidate paths in the qualified queue P′, and record the last overlapping node as n i-1 , and record the first different node as n i ,n′ i ,n″i ,.... Add different nodes n to the target path segment P in sequence i ,n′ i ,n″ i , ..., forming a new target path segment P′. The next loop is recursively entered until the qualified candidate queue is 1 (the qualified path is the selected path) or the qualified candidate queue is empty (no path is selected). The link-level candidate path selection problem is a variant of the multi-product flow problem. Analysis of this variant of the multi-product flow problem in existing literature indicates that the approximation ratio of this approximation algorithm is 1.5.

[0047] The transmission system framework utilizes a qubit resource allocation algorithm based on multi-agent reinforcement learning (MARL). Each network node deploys an RL agent, each composed of a deep neural network. The observed local network state is input and the agent's action response is generated. All agents execute their output actions, interacting with the global network environment to transform the current network state into the next state and return the resulting reward. Agent design elements in the MARL algorithm include a state representation o, an action encoding a, a reward calculation rd, and a training cycle episode.

[0048] Step 1: State representation unit.

[0049] Local observation o of agent n n Including candidate path information, node memory usage, memory remaining and neighbor information. Candidate path information is composed of two values express, Indicates that the kth candidate path of request r passes through node n, otherwise it does not pass. The node memory usage is determined by Indicates that node n stores the data resources of request r. The remaining memory is expressed as Neighbor information is represented as a binomial value I n,n′ , I n,n′ =1 means that node n and node n′ are neighbors. Therefore, the size of the state representation space of agent n is |R|(|P r |+1)+N+1, where |R| is the number of requests, |P r | is the number of candidate paths, and N is the number of nodes.

[0050] Step 2: Action encoding unit.

[0051] Action a of agent n n Expressed as That is, the number of resources pre-configured by agent n for request r at the tth transmission is Setting up X thrThe upper bound of Because the transmission capacity of the entanglement link is determined by the physical properties of the communication link. In order to achieve the concurrent transmission of multiple requests, a "request network module" is designed to independently determine the pre-configured resources of the request r The size of the action space of each agent is |R|X thr .

[0052] Step 3: Reward calculation unit.

[0053] The local reward of agent n Total memory usage including node n Total provisioned resources And the global rewards issued by the global controller in, It means that r is requested to select the kth candidate path as the transmission path during the tth entanglement transmission. Indicates the number of requests that globally complete data needs until the tth entanglement transmission.

[0054] Step 4: Model training unit.

[0055] An episode represents a single RL training cycle. An episode consists of multiple steps, representing the number of times the agent interacts with the network environment. The maximum number of steps per episode is set to T. The agent stops interacting with the environment when: the network completes its transmission response to all requests, or the network reaches its maximum number of transmissions, T.

[0056] The core of this algorithm design is to use an approximate algorithm with low computational complexity to partition the problem solution space and embed it into the Markov process decision of reinforcement learning, making the problem solving faster in a large-scale network environment. Therefore, any reinforcement learning method can be applied to intelligent agent learning. The present invention uses a relatively simple Deep Q-learning Network (DQN) to implement policy learning.

[0057] In another aspect, the present invention proposes a quantum network data transmission system for multi-qubit communication. The system includes a quantum network model adapted for multi-qubit communication and employs conventional noise system modeling. Based on a simulated network model, the system implements a collaborative optimization framework between a centralized controller and distributed agents, achieving real-time, concurrent response to multiple transmission requests. Testing has shown that the system can process 200-1000 requests in real time on a quantum network with 18-108 network nodes, and exhibits scalability across a wide range of problems.

[0058] Specifically, the system consists of three implementation phases: qubit resource preconfiguration phase, routing phase, and actual transmission phase.

[0059] In the qubit resource pre-configuration phase, it is completed independently by the distributed agents deployed on the network nodes. Agent n inputs the network information o that it observes locally. n , and output its response action a n , that is, qubit resources pre-configured for each request In particular, during the 0th optimization, each request randomly selects an initial entangled path for the agent to collect local network states.

[0060] The routing phase involves selecting the E2E entanglement path for each request. During each entanglement transmission, the distributed agent's qubit resource provisioning decision is transmitted to the central controller. The central controller uses the provisioning policy, candidate path information, and global network information, comprehensively considering entanglement fidelity, transmission delay, and node memory capacity constraints to establish and solve a centralized optimization model. For each request, the centralized controller selects the candidate path k that satisfies multiple constraints and maximizes throughput, given the current provisioning policy and network status.

[0061] The actual transmission phase occurs within the entire network environment. The preconfigured strategies of the distributed agents are filtered by the constraints and optimization objectives of the centralized controller to determine the qubit resources that can actually participate in the network link transmission. Based on this decision information, the network executes the tth entanglement transmission, completes the state transition to the t+1th entanglement transmission, and returns the global benefit.

[0062] Modeling high-dimensional quantum network components provides a formulaic basis for describing multi-qubit communication. The system is initiated during the qubit resource preconfiguration phase. Each agent collects local network observations to generate a preconfigured strategy and transmits this strategy to the centralized controller. The centralized controller's routing phase involves screening the preconfigured strategies and selecting candidate paths to satisfy various quantum entanglement constraints. Based on qualified qubit resource configuration and path selection strategies, the actual transmission phase completes global network state migration.

[0063] Beneficial effects

[0064] Compared with the prior art, the present invention has the following advantages:

[0065] 1. The method of the present invention clarifies for the first time the E2E entanglement generation mechanism and specific noise interference mechanism of multi-qubit communication in quantum networks, models a multi-qubit entanglement quality assessment model, an E2E communication delay model, and a request data transmission model, and designs a network response model for centralized and distributed collaborative optimization.

[0066] 2. The system of the present invention trains and deploys the proposed multi-qubit entangled communication framework on quantum network models of different scales to batch process the transmission needs of heterogeneous requests, and can provide stable and efficient data transmission services for concurrent responses to batch requests in large-scale quantum networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is an overall schematic diagram of the present invention.

[0068] Figure 2 This is a diagram representing the high-dimensional entangled state of the method of the present invention.

[0069] Figure 3 This is the algorithm design diagram of the method of the present invention.

[0070] Figure 4 This is a deployment workflow diagram of the system of the present invention. DETAILED DESCRIPTION

[0071] The present invention is further described in detail below with reference to the accompanying drawings and examples.

[0072] Example

[0073] like Figure 1 As shown, a quantum network data transmission method for multi-qubit communication includes quantum network modeling, collaborative optimization algorithm and transmission system framework.

[0074] Quantum network modeling uses high-dimensional entangled states—quantum W states—to represent the formation process of multi-qubit quantum entanglement, quantum decoherence noise models, and the high-dimensional quantum state properties of fidelity, a quantum entanglement quality assessment indicator. Quantum network components also include the amount of entangled transmission data, end-to-end (E2E) entanglement quality quantification, and E2E entanglement delay modeling. The specific steps are as follows:

[0075] Step 1: Representation of high-dimensional quantum state properties.

[0076] Step 1.1: Multi-qubit entangled state representation.

[0077] When a request to establish a d-qubit entanglement link is received between two network nodes, the register in the sending node selects d qubits from the idle qubit resources to generate a transmitter, such as Figure 2 As shown. The d-qubit quantum state in the emitter is represented as:

[0078]

[0079] Here, p represents the probability that each qubit is measured as state |1>. represents the dimension migration operation of state expansion. represents the nitrogen vacancy ground state.

[0080] If there is no noise modeling, that is, the uncertainty of entanglement establishment cannot be quantified, then the register can be successfully generated through multiple entanglement establishment attempts. The high-dimensional W state is used to represent this entanglement establishment process as:

[0081]

[0082] Where f represents the number of times the system attempts to establish entanglement. i Represents a one-dimensional unit quantum state.

[0083] Formula 1 and Formula 2 can be written as:

[0084]

[0085] Among them, |e ij > is the unit quantum state, (i·d+j) means that only the position is in state |1>, and the rest are in state |0>.

[0086] Step 1.2: Quantum decoherence representation.

[0087] The collapse decoherence model is used, which occurs when the interaction force between the qubit particles in the register is much greater than the interaction force between other scattered particles. The collapse decoherence model is expressed as:

[0088]

[0089] Where τ represents the scattering constant and Δx represents the physical distance between the two entangled particles. Introducing collapse decoherence noise into the multi-qubit entangled communication model represented by the high-dimensional W state (Equation 3) yields the multi-qubit entanglement establishment model under decoherence noise:

[0090]

[0091] Among them, e ij represents a two-dimensional unit quantum state. e represents the natural base.

[0092] Step 1.3: Entanglement quality indicator representation.

[0093] Entanglement fidelity is an indicator of entanglement quality, with a value between (0, 1). It is used to measure the difference between the current quantum state and the initial quantum pure state. According to Equation 5, the fidelity of a quantum entanglement link is expressed as:

[0094]

[0095] Step 2: Quantum network component representation.

[0096] Step 2.1: Transmitted data volume representation.

[0097] At the tth entanglement transmission, node n stores resources and pre-configures the kth candidate E2E entanglement path selected by request r Resources. Therefore, during the tth entanglement transmission, the path The entangled link l on n,n+1 The amount of data transmitted is:

[0098]

[0099] Among them, node n-1 and node n+1 are path The predecessor and successor nodes of node n. Then, during the tth entanglement transmission, the terminal node d of the request r r The amount of data that has successfully arrived in is:

[0100]

[0101] when , request r is successfully responded to, and all resources occupied by request r in the network are released.

[0102] Step 2.2: Quantify the E2E entanglement quality.

[0103] The E2E entanglement quality quantitative indicator - E2E entanglement fidelity, is obtained by multiplying the link-level fidelity obtained in step 1.3, which means that the probability of successfully establishing the E2E entanglement path is:

[0104]

[0105] Considering the collapse decoherence model in step 1.2, the fidelity of the E2E entanglement path is calculated by the following formula:

[0106]

[0107] in,

[0108] Step 2.3: E2E entanglement delay modeling.

[0109] The E2E entanglement transmission delay of request r during the tth entanglement transmission is:

[0110]

[0111] in, is the cumulative sum of the previous t-1 entanglement transmission delays.

[0112] The collaborative optimization algorithm adopts the algorithm design process of multi-objective collaborative optimization between a centralized controller and a distributed RL agent. Figure 3 As shown in Figure 1, distributed RL agents optimize local objectives by collecting local network state information and pre-configuring the required qubit resources for each request. Using the pre-configured qubit resources as parameters, the centralized controller calculates the data transfer volume, transmission latency, resource usage, and throughput of the entanglement path, models the ILP solution, and selects the end-to-end entanglement path required for each request to optimize the global objective. The detailed implementation steps of the distributed agent are as follows:

[0113] Step 1: Initialize the agent model.

[0114] During the configuration phase, each network node aims to collaborate with neighboring nodes to complete the transfer task while fully utilizing their respective resources. The RL agent in each node inputs locally observed network information into the policy network to obtain actions and calculate local rewards. The policy network is trained through the gradual accumulation of interaction experience with the network environment. All trained agents adaptively transmit qubit data within the quantum network to update the request set.

[0115] Take DQN as an example, as an RL algorithm to learn distributed pre-configured policies. DQN consists of a policy network and a target network: the policy network learns the mapping between observations and Q values, while the target network stabilizes the learning process by periodically copying parameters from the policy network. The DQN agent is updated as follows:

[0116]

[0117] Among them, α is the update step size, γ is the reward factor, is the local reward of agent n at the tth entanglement transfer.

[0118] Step 2: Agent model improvement.

[0119] In order to configure resources for multiple requests simultaneously, the present invention designs an improved policy network architecture, which divides the output layer of the policy network into |R| request modules. Each request module contains two fully connected layers, which independently determine its resource configuration strategy. The improved policy network is trained through the following process:

[0120] Step 2.1: Training process of the improved policy network.

[0121] Sample a small batch of experience tuples (O, A, RD, O') with a capacity of bs from the entire experience pool. The policy network of agent n consists of two parts: public parameters Evaluation part network represented by and request module network like Figure 3 As shown, the output of the request module is represented as:

[0122]

[0123] Therefore, at the tth entanglement transfer, the policy network output of agent n is:

[0124]

[0125] The detailed implementation steps of the centralized controller are as follows:

[0126] Step 1: ILP modeling of path selection.

[0127] The goal of the routing phase is to, based on the pre-allocated resource strategy obtained in the configuration phase Select the best E2E path from the candidate path set of all requests. In the case of link-level data transmission It can be obtained by Equation 7. Due to the different transmission capabilities of entangled links, the present invention constructs an accurate integer linear programming (ILP) model to maximize the total amount of transmitted data:

[0128]

[0129]

[0130] Formula 15 maximizes the total amount of data transmitted for the tth entanglement transmission. Formula 16 ensures that the fidelity of each E2E entanglement must be greater than the threshold F thr Formula 17 stipulates that the transmission delay of each request must be less than the threshold D thr Equation 18 specifies the node capacity limit. Equation 19 ensures that the t-th E2E entanglement for each request can be established through at most one physical path. Here, T represents the entire network response period, i.e., the maximum number of entanglement transmissions.

[0131] Step 2: Approximate solution.

[0132] By adding an integer variable Relaxing to non-integer variables The implementation solves the above integer linear programming (ILP) problem in polynomial time complexity. The linear programming (LP) solver obtains the fractional solution These non-integer solutions can satisfy fidelity, latency, and memory constraints while maximizing global data throughput. Secondly, we improve the branch-and-bound algorithm by introducing local-objective filters and multi-constraint filters in each branch to select a near-optimal E2E entanglement path.

[0133] Step 2.1: Link-level B&B algorithm.

[0134] In the link-level B&B algorithm, all candidate paths are first filtered by the recorded global goals and constraints (Equations 16, 17, and 18) to select qualified candidate path branches. For request r, if its qualified candidate path set If the current number of paths is less than or equal to 1, the selection result of the request is determined (not selected at all / select one), then the optimization target obj of the global record is updated global , then enter the next request branch. If the number of stored paths is greater than 1, the link-level branch is entered and the sub-path growth branch is started from the first non-public point until

[0135] The transmission system framework deployment includes two deployment modes: fixed request set deployment and flexible request set deployment. Each agent is trained to cooperate in transmitting information to meet the transmission requirements of quantum requests under multiple constraints, while optimizing its own resource usage through local observation. The input of each agent is set as |R|(|P r |+1)+|N|+1, which is related to the topology scale and request pool capacity. The specific implementation of the two deployment modes is as follows:

[0136] Step 1: Fixed request set deployment.

[0137] Given a fixed-length request set The request set is divided into z request pools, where If the zth request pool is not full, the configuration decision in the qubit resource pre-configuration phase is Mask with 0. Each agent maintains z local environments, one for each request pool. All request pools share network node resources.

[0138] like Figure 4 As shown, after the resource pre-configuration phase, the system obtains resource configuration decisions Indicates that among all requests, the request The absolute order of the global routing is based on the pre-configured resources. Calculate link-level pre-transmission volume and select appropriate E2E entanglement paths for all requests Entering the transmission phase, all decided resource transmissions are executed concurrently, and it is determined whether the requests in each request pool are successfully responded to. If they are successfully responded to, the request is deleted, and a copy is randomly selected from the remaining unfinished requests in the request pool to fill the deleted vacant position, until all requests in the request pool are successfully responded to, and then the request pool is destroyed. Figure 4 As shown, Request pool copies and to fill the requested position.

[0139] Step 2: Flexible request set deployment.

[0140] Given an infinite request queue System from Select |R| requests to fill the initial request pool R and establish the initial local environment to start the agent. In the pre-configuration and routing phases, the agent only maintains a local environment to generate qubit resource configuration and E2E entanglement path selection decisions. In the actual transmission phase, when some requests in the request pool R are successfully responded to, unlike the fixed deployment, the vacant positions in the request pool are filled by The new request is filled in. Figure 4 , new request and The memory usage of each node when initially added is 0.

[0141] The method and system provided in this embodiment can be run on a self-made quantum network topology implemented using NetworkX and the Waxman model. The system algorithm can be implemented using Python 3.6 and Gurobi 9.1. The system can train and run intelligent agents on an Intel(R) Core(TM) i7 Windows 10 64-bit system with an NVIDIA GeForce GTX 1660SUPER graphics card.

Claims

1. A quantum network data transmission method for multi-qubit communication, characterized in that: Including quantum network modeling, collaborative optimization algorithm design and construction of transmission system framework; Quantum network modeling introduces a multi-qubit high-dimensional entangled state representation through a physical experimental model that realizes multi-qubit entanglement. This high-dimensional state is used to define and quantify quantum network model components, including entanglement link quality, entanglement transmission delay, and entanglement transmission data volume. The following steps are included: Step 1: High-dimensional quantum state representation of multi-qubit entanglement; A quantum node contains multiple quantum registrars that are ready to generate multi-qubit entanglement. Each registrar contains d single-qubit (qubit) photon pairs, which are interconnected through internal entanglement. By packaging and sending these d qubits of entanglement simultaneously, the registrar completes an entanglement transmission and forms a d-dimensional entangled quantum state between the sending and receiving nodes. The frequency of successful entanglement establishment is estimated through multiple entanglement establishment attempts. Assume that sending f times can ensure that the entangled state is successfully established at least once, and use the W quantum state form to represent the d-dimensional entangled quantum state. Where f is the number of transmissions; d is the entanglement dimension; p is the probability of obtaining the state |1> when measuring a single qubit; |e ij > indicates that in the i·j-bit W quantum state, only the (i·d+j) bit is in state |1>, and the rest are all in state |0>; this entangled quantum state consumes f·d qubits of the sending node; Step 2: High-dimensional quantum state representation of network model components; Step 2.1: Quantify the end-to-end entanglement quality; The entanglement fidelity metric is used to characterize the quality of entanglement. In the case of a pure state of quantum entanglement, the entanglement quality is 1, and the entanglement link can be successfully generated. For a request with a starting node, the fidelity of its end-to-end entanglement path is the product of the fidelity of each single link in the path. Step 2.2: Modeling end-to-end communication delay; End-to-end communication delay of quantum requests = = Current entanglement communication delay (selected path length ) + historical accumulated communication delay Among them, the current entanglement communication delay refers to the communication delay of the end-to-end entangled transmission of quantum information and information measurement results at the current time t; the historical cumulative delay refers to the cumulative communication delay at the previous time t-1; Step 2.3 quantifies the amount of transmitted data; The amount of data transmitted is quantified in the form of entangled links one by one in the end-to-end entangled path of the request; ij , the amount of data transmitted on the link for: (Node i decides the amount of data to be transmitted The amount of data stored in node i for this request +The amount of data flowing into node i for this request ) Step 3: Joint optimization objective modeling; Based on the network component model established in steps 2.1, 2.2, and 2.3, further establish the efficiency optimization goal of the quantum network responding to multiple requests; The collaborative decision-making between the centralized controller and the distributed agents at each moment is achieved by defining a joint optimization goal, where: The optimization goal of the centralized controller = Σ(request, path, node) (data volume transmitted by node n ), that is, maximizing the information flow of the network at the current moment; The local optimization goal of distributed agent n = Σ(request, path) (the resources stored in node n for request r) + resources pre-configured in node n for request r transmission ) + global reward returns the number of completed requests; The collaborative optimization algorithm is a quantum resource allocation optimization method that uses a combination of centralization and distribution. It delegates the decision-making of the quantum network's response to multi-request transmission needs to the central controller and distributed agents, respectively, establishing an optimization dependency between the centralized controller and multiple agents. The centralized controller uses integer linear programming (ILP) modeling and the B&B approximation algorithm to handle the E2E entanglement path selection process for requests. The transmission system framework adopts a qubit resource allocation algorithm based on multi-agent reinforcement learning (MARL). Each network node deploys an RL agent, each composed of a deep neural network, which inputs the observed local network state to obtain the agent's action response. All agents execute the output actions separately, interacting with the global network environment to transfer the network state at the current moment to produce the network state at the next moment, and return the resulting reward. The agent design elements in the MARL algorithm include state representation o, action encoding a, reward calculation rd, and training cycle episode.

2. A quantum network data transmission method for multi-qubit communication according to claim 1, characterized in that: The collaborative optimization algorithm includes the following steps: Step 1: ILP modeling of entanglement path selection; First, after the distributed agents initialize their respective policy networks, the agents on each node input the initial state, that is, the local observed network information. Output the qubit resources pre-allocated by this node for each request After obtaining the pre-configured resources of each node for each request, the controller Each entangled connection in Calculate the amount of pre-transmitted data for request r through the link Then calculate the optimization goal of ILP Entanglement mass fidelity constraint Entanglement delay constraint and node memory constraints Where R represents the quantum request set, P r represents the candidate path set for quantum request r, represents the kth candidate path of quantum request r; is a Boolean variable, Indicates that during the t-th entanglement transmission process, r is requested to select its k-th candidate path to transmit data. If it is 0, it means no selection; It is a Boolean constant. A value of 1 indicates that the kth candidate path of request r passes through network node n, and a value of 0 indicates that it does not pass through network node n. represents the amount of data transmitted from node n to node n+1 on the selected candidate path for request r during the tth entanglement transmission; e represents the natural base; F thr Represents the fidelity threshold of the entanglement transmission path. Only when the fidelity of the path is greater than the threshold F thr The transmission path is valid only when represents the length of the kth candidate path for request r; represents the cumulative transmission delay of request r in the first t-1 entanglement transmissions; D thr Indicates the transmission path delay threshold; represents the amount of data stored in node n by request r during the tth entanglement transmission; C represents the amount of data transmitted from node n-1 to node n on the selected candidate path by request r during the tth entanglement transmission; n Indicates the resource performance upper limit of network node n; Step 2: Approximate solution of entanglement path selection; Step 2.1: Linear relaxation of ILP; Relax the binomial variables of the ILP problem modeled in step 1 into continuous variables between [0,1] and use the solver to solve it in linear time to obtain the optimal solution of the relaxed problem in the continuous space. Step 2.2: B&B algorithm at link level; Use the link-level B&B algorithm to approximate the relaxed fractional solution to the integer solution Y k r,t , and prove the approximation ratio; including direct approximation, path decomposition, and target constraint pruning; First, The fractional solution of is directly approximated to the corresponding entanglement path k, and the remaining Push them into the processing queue in descending order; put the largest Out of the queue, the variable is applied to all candidate paths of request r To filter: Whether the target path segment P satisfies the three constraints; Does the target path segment P exist? middle; If both of the above are true, then Enter the qualified queue P′; After that, find the shortest overlapping path segments of all qualified candidate paths in the qualified queue P′, and record the last overlapping node as n i-1 , and record the first different node as n i ,n′ i ,n″ i ,...; add different nodes n to the target path segment P in sequence i ,n′ i ,n″ i ,..., forming a new target path segment P′, and recursively entering the next loop until the qualified candidate queue is 1 or the qualified candidate queue is empty.

3. A quantum network data transmission method for multi-qubit communication according to claim 1, characterized in that: The transmission system framework includes the following steps: Step 1: State representation unit; Local observation o of agent n n Including candidate path information, node memory usage, memory remaining and neighbor information; candidate path information is composed of two values express, Indicates that the kth candidate path of request r passes through node n, otherwise it does not pass; the node memory usage is determined by Indicates that node n stores the data resources of request r; the remaining memory is expressed as Neighbor information is represented as a binomial value I n,n′ , I n,n′ = 1 means that node n and node n′ are neighbors; therefore, the size of the state representation space of agent n is |R|(|P r |+1)+N+1, where |R| is the number of requests, |P r | is the number of candidate paths, N is the number of nodes; Step 2: Action encoding unit; Action a of agent n n Expressed as That is, the number of resources pre-configured by agent n for request r at the tth transmission is Setting up X thr The upper bound of Because the transmission capacity of the entanglement link is determined by the physical properties of the communication link; in order to achieve concurrent transmission of multiple requests, a "request network module" is designed to independently determine the pre-configured resources of the request r The size of the action space of each agent is |R|X thr ; Step 3: Reward calculation unit; The local reward of agent n Total memory usage including node n Total provisioned resources And the global rewards issued by the global controller in, It means that r is requested to select the kth candidate path as the transmission path during the tth entanglement transmission. Indicates the number of requests that globally complete data needs until the tth entanglement transmission; Step 4: Model training unit; An episode represents an RL training cycle. An episode consists of multiple steps, that is, the agent interacts with the network environment step times. The longest episode of the agent is set to contain T steps, that is, the agent stops interacting with the environment in two situations: the network completes the transmission demand response for all requests, and the network iteration reaches the maximum number of transmissions T.

4. A quantum network data transmission system for multi-qubit communication according to the method of claim 1, characterized in that: This includes a quantum network model adapted for multi-qubit communication with conventional noise system modeling. Based on the simulation network model, the system implements a collaborative optimization framework between a centralized controller and distributed agents to achieve real-time, concurrent response to multiple request transmission needs. The system consists of three implementation phases: qubit resource preconfiguration phase, routing phase, and actual transmission phase; In the qubit resource pre-configuration phase, it is completed independently by the distributed intelligent agents deployed on the network nodes; the intelligent agent n inputs the network information o that it observes locally. n , and output its response action a n , that is, qubit resources pre-configured for each request During the 0th optimization, each request randomly selects an initial entangled path for collecting the local network state of the agent; The routing phase involves selecting the requested E2E entanglement path. During each entanglement transmission, the distributed agent's qubit resource preconfiguration decision is transmitted to the central controller. The central controller uses the preconfiguration strategy, candidate path information, and global network information to comprehensively consider entanglement fidelity, transmission delay, and node memory capacity constraints, establishes and solves a centralized optimization model. For each request, the centralized controller selects the candidate path k that satisfies multiple constraints and maximizes throughput under the current preconfiguration strategy and network status. The actual transmission phase occurs within the entire network environment. The pre-configured strategies of the distributed agents are filtered by the constraints and optimization objectives of the centralized controller to determine the qubit resources that can actually participate in the network link transmission. Based on this decision information, the network executes the tth entanglement transmission, completes the state transition to the t+1th entanglement transmission, and returns the global benefit. Modeling of high-dimensional quantum network components provides a formulaic basis for describing multi-qubit communications; The system is started during the qubit resource preconfiguration phase. Each intelligent agent collects local network observations to generate a preconfiguration strategy and transmits the strategy to the centralized controller. The routing phase in the centralized controller involves screening the preconfiguration strategy and selecting candidate paths to satisfy various quantum entanglement constraints. Based on qualified qubit resource configuration and path selection strategies, the state migration of the global network is completed during the actual transmission phase.

Images