Multi-satellite beam switching method based on CU-DU architecture and related equipment

Abstract

The invention provides a multi-satellite beam switching method and related equipment based on a CU-DU architecture, and the method comprises the steps: receiving a switching request sent by a DU end, and obtaining the user data of a UE end in the switching request; predicting the user data through an LSTM network to obtain prediction data of the UE end in the next time slot; determining an optimization algorithm matched with the current computing power of the CU end, and solving a corresponding satellite beam switching problem under the CU-DU architecture according to the optimization algorithm and the prediction data to obtain a switching strategy; and sending the switching strategy to the DU end, wherein the switching strategy is used for indicating the UE end to switch a satellite beam. By adopting the method provided by the invention, the accuracy of satellite beam switching can be improved, and the switching time delay and unnecessary frequent switching can be effectively reduced. The problems of low satellite beam switching accuracy, long time delay and the like in the prior art are effectively solved.

Description

Technical Field

[0001] The present invention relates to the field of satellite communication technology, and in particular to a multi-satellite beam switching method based on a CU-DU architecture and related equipment. Background Art

[0002] With the rapid development of wireless communication technology, low-orbit satellite communication systems have been widely used around the world due to their advantages such as low latency, high bandwidth and wide coverage. In order to improve network performance and resource utilization, modern satellite communication systems generally use multi-beam technology, which enables a single satellite to cover multiple ground areas at the same time and provide users with more flexible communication services.

[0003] In a multi-beam satellite communication system, the user equipment (UE) needs to switch between different satellite beams based on its own location and signal quality. At present, satellite communication systems usually adopt an architecture in which the central unit (CU) and the distributed unit (DU) are deployed separately, in which the CU is responsible for centralized control and resource management, and the DU realizes wireless access and user data transmission. This CU-DU separation architecture can improve system flexibility and scalability, but it also brings new challenges to satellite beam switching.

[0004] However, the existing satellite beam switching methods mainly make decisions based on information from the DU side, which makes it difficult to ensure the accuracy of satellite beam switching under the CU-DU separation architecture. Problems such as extended switching time and high switching frequency are prone to occur, affecting communication quality. Summary of the invention

[0005] The present invention provides a multi-satellite beam switching method and related equipment based on the CU-DU architecture, which are used to solve the problem in the prior art that satellite communication systems are difficult to achieve fast and accurate beam switching in a complex and changeable satellite communication environment, thereby improving system performance and user experience.

[0006] In a first aspect of the invention, a multi-satellite beam switching method based on a CU-DU architecture is provided, which is applied to a CU end, wherein the CU end establishes a communication connection with a DU end, and the DU end establishes a communication connection with a UE end through a satellite beam, and the UE end, the CU end, and the DU end constitute a CU-DU architecture, comprising the following steps: Receiving a handover request sent by the DU, and obtaining user data of the UE in the handover request; Predicting the user data through the LSTM network to obtain predicted data of the UE in the next time slot; Determine an optimization algorithm that matches the current computing capability of the CU end, and solve the corresponding satellite beam switching problem under the CU-DU architecture according to the optimization algorithm and the prediction data to obtain a switching strategy; The switching strategy is sent to the DU end, and the switching strategy is used to instruct the UE end to switch the satellite beam.

[0007] In a second aspect of the invention, another multi-satellite beam switching method based on a CU-DU architecture is provided, which is applied to a DU end, wherein the DU end establishes a communication connection with a CU end, and the DU end establishes a communication connection with a UE end through a satellite beam, and the UE end, the CU end, and the DU end constitute a CU-DU architecture, including the following steps: Sending a switching request to the CU end; Receive a switching strategy fed back by the CU end, where the switching strategy is obtained by solving the corresponding satellite beam switching problem under the CU-DU architecture using an optimization algorithm and prediction data that match the current computing power of the CU end, wherein the prediction data is prediction data of the UE end in the next time slot obtained by predicting the user data through an LSTM network; The UE is instructed to switch the satellite beam according to the switching strategy.

[0008] In a third aspect of the invention, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, a multi-satellite beam switching method based on the CU-DU architecture as described above is implemented.

[0009] In a fourth aspect of the invention, a non-transitory computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, a multi-satellite beam switching method based on a CU-DU architecture as described in any one of the above is implemented.

[0010] In a fifth aspect of the invention, a computer program product is provided, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the multi-satellite beam switching methods based on the CU-DU architecture as described above.

[0011] In summary, one or more technical solutions provided in the present invention have at least the following technical effects or advantages: The present invention obtains user data on the UE side at the CU side and predicts the predicted data of the next time slot, selects a suitable optimization algorithm in combination with the current computing power of the CU side, comprehensively considers the predicted data and the characteristics of the CU-DU architecture to solve the satellite beam switching problem, and finally obtains the switching strategy and sends it to the DU side to guide the UE side to switch. This method makes full use of the computing resources of the CU side and the ability to predict the status of the UE side, so that the switching decision is no longer solely dependent on the information on the DU side, but is based on more comprehensive data and more advanced algorithms. Through predictive decision-making and computing power adaptive optimization algorithms, this solution can significantly improve the accuracy of satellite beam switching, effectively reduce switching delays and unnecessary frequent switching, thereby achieving more efficient and reliable satellite communication services under the CU-DU separation architecture, and effectively solves the problems of low switching accuracy, long delays and high frequencies in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0013] Figure 1 It is a schematic diagram of the architecture of a CU-DU provided by the present invention.

[0014] Figure 2 This is one of the flow charts of a multi-satellite beam switching method based on a CU-DU architecture provided by the present invention.

[0015] Figure 3 This is a schematic diagram of an LSTM network structure provided by the present invention.

[0016] Figure 4 It is a schematic diagram of the long short-term memory deep reinforcement learning principle provided by the present invention.

[0017] Figure 5 It is a schematic diagram of a network architecture and switching classification scenario under a CU-DU separation architecture provided by the present invention.

[0018] Figure 6 This is a pre-switching flowchart of different DUs of the same CU under a CU-DU separation architecture provided by the present invention.

[0019] Figure 7 It is a pre-switching flow chart of different CU ends and different DU ends under a CU-DU separation architecture provided by the present invention.

[0020] Figure 8This is the second flow chart of a multi-satellite beam switching method based on a CU-DU architecture provided by the present invention.

[0021] Fig. 9 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] Before introducing the embodiments of the present invention, some terms involved in the embodiments of the present invention are first defined and explained.

[0024] The CU end establishes a communication connection with the DU end, and the DU end establishes a communication connection with the UE end through a satellite beam. The UE end, the CU end and the DU end constitute a CU-DU architecture.

[0025] Centralized Unit (CU): The CU is a functional entity in the 5G network architecture, responsible for non-real-time high-level protocol processing and control functions. It is usually deployed in a centralized location and can manage multiple DUs. The CU mainly handles non-real-time functions such as radio resource control (RRC) and packet data convergence protocol (PDCP), and is responsible for tasks such as resource scheduling and mobility management.

[0026] Distributed Unit (DU): The DU is another important functional entity in the 5G network architecture, responsible for real-time underlying protocol processing. The DU is usually deployed close to the antenna and directly connected to the radio unit (RU). The DU mainly handles real-time functions such as media access control (MAC), radio link control (RLC), and the physical layer.

[0027] User Equipment (UE): UE refers to the terminal equipment used by users, such as smartphones, tablet computers, IoT devices, etc. In satellite communication networks, UE can be ground mobile terminals, vehicle-mounted equipment, ship-mounted equipment, and other terminal equipment that can receive satellite signals.

[0028] The CU-DU architecture is a functional separation architecture used in 5G networks, especially suitable for satellite communication network environments. Figure 1 , Figure 1 A schematic diagram of the architecture of a CU-DU provided by an embodiment of the present invention is shown. The architecture divides the traditional base station functions into the CU end and the DU end. The CU end is responsible for non-real-time high-level protocol processing and control functions, such as RRC and PDCP, and mainly handles tasks such as resource scheduling and mobility management. The CU end is usually deployed in a centralized location and can manage multiple DUs. The DU end connected to the CU end is responsible for real-time underlying protocol processing, such as MAC, RLC and physical layer functions, and is usually deployed close to the antenna. In a satellite communication network, the DU end establishes a wireless connection with the user equipment (UE end) through a satellite beam.

[0029] Based on the above examples, please refer to Figure 2 , Figure 2 The invention provides a flowchart of a multi-satellite beam switching method based on a CU-DU architecture. The method is applied to a CU end and may further include the following steps: Step 101: Receive a handover request sent by the DU, and obtain user data of the UE in the handover request.

[0030] Specifically, in a satellite communication network with a CU-DU architecture, the UE needs to periodically send measurement reports to the currently serving DU to evaluate its current communication quality and the surrounding available satellite beams. After receiving these measurement reports, the DU will analyze them according to the preset switching trigger conditions. When the DU finds that the current communication quality of the UE drops below a certain threshold, or there is a candidate satellite beam with better signal quality, it will be considered that the UE needs to switch satellite beams.

[0031] In the above case, the DU will send a handover request to the CU. The handover request carries the UE's measurement report data, the current serving satellite beam information, and a list of possible candidate satellite beams. After receiving the handover request, the CU starts the decision-making process for satellite beam handover.

[0032] Step 102: predict the user data through the LSTM network to obtain the predicted data of the UE in the next time slot.

[0033] After receiving the handover request sent by the DU, the CU needs to predict the future state of the UE in order to make a more accurate handover decision. To this end, the CU inputs the user data of the UE carried in the handover request into the LSTM network and outputs the predicted data of the UE in the next time slot. By analyzing and learning historical data, the state of the UE at a future moment can be effectively predicted, thereby improving the accuracy and efficiency of handover decisions.

[0034] Specifically, step 102 may further include the following steps: Step 201: The center position and channel quality of n historical time slots are state aggregated through the LSTM network to obtain the center position and channel quality of the UE end in the next time slot.

[0035] Among them, the center position and channel quality are two key parameters for describing the UE status. The center position refers to the geographical location coordinates of the UE on the surface of the earth, which can be understood as the longitude and latitude coordinates of the UE in the embodiment of the invention. For a mobile UE, its center position may be represented as a two-dimensional coordinate sequence that changes over time.

[0036] Channel quality refers to the communication quality index of the wireless link between the UE and the current service satellite, which can be understood as a set of parameters reflecting factors such as signal strength and signal-to-noise ratio in the embodiments of the invention. This may include reference signal received power, signal-to-interference-plus-noise ratio, and reference signal received quality.

[0037] The above two parameters can be used as inputs of the LSTM network to characterize the communication status of the UE at different times. By analyzing the historical change trends of these two parameters, the LSTM network can predict the possible status of the UE at future times. Based on these prediction results, the CU can determine whether switching is required and select the best switching target. After the switching is executed, these parameters can also be used to evaluate the effect of the switching decision, such as the degree of improvement in channel quality after switching. In addition, the long-term accumulated center location and channel quality data can be used to optimize the coverage and resource allocation strategy of satellite beams to further improve network performance.

[0038] By considering both the center position and the channel quality, the present invention can comprehensively evaluate the status of the UE in terms of both space and communication quality. The change in the center position can help predict whether the UE is about to enter or leave the coverage of a satellite beam, while the change in channel quality directly reflects the communication status of the UE. The above multi-dimensional status evaluation enables the CU to make more accurate and effective switching decisions.

[0039] Specifically, the CU inputs the center position and channel quality of the UE in tn time slots carried in the switching request into the LSTM network for state aggregation, and outputs the center position and channel quality prediction value of the UE in t+1 time slots. This step is introduced to make full use of the historical data of the UE, and to capture the long-term and short-term dependencies in the time series through the characteristics of the LSTM network, so as to more accurately predict the state of the UE in the next time slot and provide a reliable basis for subsequent switching decisions.

[0040] In specific implementation, the location information and channel information between the satellite and the user in several historical time slots can be used as input and sent to the LSTM network for processing. The above input data includes: a position matrix representing the center point under all beams in several historical time slots, where is the position of the center point under all beams in time slot tn; a position matrix representing the user in several historical time slots; and a channel quality matrix representing the user and each beam node in several historical time slots, where is the channel quality of the user with each beam node at time slot tn. The above multi-dimensional input enables the LSTM network to simultaneously consider the movement of satellite beams, the movement of users, and the changes in channel quality, thereby comprehensively analyzing the system status.

[0041] For example, the location information and channel information between the satellite and the user in several historical time slots are As input, it is input into the LSTM network for state aggregation. At the same time, by extracting historical information from the continuous state, the terminal can understand the hidden change rules in the trajectory of the nodes.

[0042] in, represents the position of the center point of all beams in time slot tn, Represents the position matrix of the center points of all beams under several historical time slots; represents the user's position matrix under several historical time slots; represents the channel quality between the user and each beam node at time slot tn, Represents the channel quality matrix between the user and each beam node in several historical time slots.

[0043] Finally, the LSTM network outputs the predicted and extracted satellite and user location information for the next time slot. , as the pre-operation support of the algorithm. Among them, Indicates the position of the center point of all beams in the next time slot t+1; Indicates the position of the user in the next time slot t+1; It represents the channel quality between the user and each beam node at the next time slot t+1.

[0044] like Figure 3 and Figure 4 As shown, Figure 3 A schematic diagram of an LSTM network structure provided by the present invention, Figure 4 A schematic diagram of the long short-term memory deep reinforcement learning principle provided by the present invention. Specifically, the LSTM network is mainly composed of a forget gate, a memory gate, a transmission belt unit state update, and an output gate.

[0045] (1) Forget gate: For the unit state of LSTM, too much memory information will affect the neural network's processing of the current input information. Therefore, after a period of time, some "outdated" information needs to be selectively abandoned by the forget gate. and , output a Vectors of the same size ,in . Forget Gate The calculation formula is: ; in, is the state vector at the previous moment, Input vector for the current moment, is the parameter matrix, represents the activation function sigmoid, is the bias term.

[0046] (2) Memory gate: The memory gate is responsible for extracting and filtering valid information from the input vector at the current moment, and rating each component in the vector from 0 to 1. The higher the rating, the more memory that enters the unit state. The input gate consists of two activation functions: the tanh function and the sigmoid function. First, the tanh function is used to extract valid information, and then the sigmoid function is used to rate the information and control the memory placed in the unit state. The calculation formulas of the memory gate are: ; ; in, represents the generated input value vector, is the output of the input gate, represents the state vector at the previous moment, Input vector for the current moment, , is the parameter matrix, , is the bias term.

[0047] Transmission belt unit state update: get the forget gate output vector , input gate output vector , the generated input value vector After the state vector of the transmission belt at the previous moment is obtained, the information on the transmission belt is updated, which specifically includes the following process: (a) Transmit the state vector of the previous moment and Multiply the elements of , forgetting some information; (b) the generated input value vector and Multiply the elements of to get the input information; (c) , get the new transmission belt state vector at the current moment .

[0048] (3) Output gate: The output gate is a neural layer used to calculate the output value at the current moment. The output gate first uses the sigmoid function to extract information from the vector that is the integration of the current moment’s input vector and the previous moment’s output vector, and then uses the tanh function to compress the current unit state, mapping each element to The calculation formula of the output gate is:

[0049] in, is the output vector of the output gate, represents the state vector at the previous moment, Input vector for the current moment, is the sigmoid activation function, is the parameter matrix, is the bias term.

[0050] Finally, the results of the two activation functions are element-wise multiplied to obtain the output of the LSTM network at the current moment .

[0051] Step 202: The center position and channel quality of the UE in the next time slot are used as prediction data of the UE in the next time slot.

[0052] Step 103: determine an optimization algorithm that matches the current computing capability of the CU end, and solve the corresponding satellite beam switching problem under the CU-DU architecture according to the optimization algorithm and the prediction data to obtain a switching strategy.

[0053] Among them, the optimization algorithm refers to a mathematical method or calculation process for finding the optimal solution under given constraints. In the embodiment of the present invention, it can be understood as an algorithm for solving the satellite beam switching problem under the CU-DU architecture, specifically including a greedy algorithm and a deep reinforcement learning algorithm. These algorithms are used to select the optimal target satellite beam for the UE side while considering multiple objectives, thereby achieving efficient satellite beam switching.

[0054] Correspondingly, the satellite beam switching problem refers to the decision-making process of how to select the most suitable satellite beam for a mobile user device in a multi-satellite communication system to maintain a continuous communication connection. In the embodiments of the present invention, it can be understood as how to select the optimal target satellite beam for the UE based on the current network status and future prediction information under the CU-DU separation architecture to achieve efficient and stable communication services. The satellite beam switching problem can be used to solve challenges such as frequent switching and fluctuations in communication quality caused by high-speed movement of satellites and users in satellite communication networks.

[0055] Specifically, in satellite communication networks, the complexity of the satellite beam switching problem mainly stems from the high dynamics of the system. Due to the rapid movement of low-orbit satellites and the potential high-speed movement of ground users, the coverage area of ​​the satellite beam and the user location are constantly changing. This requires that the switching decision should not only consider the current communication quality, but also predict the future network status. At the same time, different types of services have different requirements for communication performance. For example, real-time services are more sensitive to delays, while non-real-time services pay more attention to throughput. In addition, although frequent switching may maintain high signal quality, it will also increase system overhead and may cause short service interruptions. Therefore, the satellite beam switching problem is essentially a complex decision-making process that needs to balance multiple objectives.

[0056] Based on the above embodiment, as an optional embodiment, the process of defining the satellite beam switching problem may specifically include the following steps: Step 301: Acquire the interaction information of the CU-DU architecture to build a system model and a channel model.

[0057] Among them, the system model includes a user model and a dynamic model. The user model is used to characterize the real-time performance of user data on the UE side, and the dynamic model is used to characterize the relative position relationship between the UE side and the satellite. Among them, the channel model includes a communication model and a switching cost model. The communication model is used to characterize the signal instructions and transmission efficiency on the UE side, and the cost model is used to characterize the switching delay and signaling cost of switching satellite beams on the UE side.

[0058] Specifically, the construction of the system model includes two parts: the user model and the dynamic model. The user model focuses on the user data characteristics of the UE, especially the real-time requirements of the data. Different types of services have different requirements for communication performance. For example, real-time services such as video calls are more sensitive to delays, while non-real-time services such as file downloads focus more on throughput. By establishing a user model, the priority of the switching strategy can be adjusted according to different service types to better meet the specific needs of users.

[0059] In the specific implementation, for the construction of user models, users generate real-time data With non-real-time data , services are transmitted in the form of data packets, each of which contains the specific data information of the service and the timestamp when the service is generated. Among them, real-time data services have strict time constraints, and data packets must be transmitted within specific time constraints. Non-real-time data services focus on the accuracy and efficiency of data transmission.

[0060] The dynamic model focuses on describing the relative position relationship between the UE and the satellite and its changes. Considering that both low-orbit satellites and ground users may be in a high-speed mobile state, accurately modeling the relative motion between them is crucial for predicting future communication quality and formulating forward-looking switching strategies.

[0061] In the specific implementation, it is necessary to dynamically analyze the satellites and users respectively, deduce the relationship between the relative position between the user and the node and the channel quality and transmission rate, and thus establish a dynamic model. The data services generated by the user can be processed by any node in the network. Construct a 5G network scenario with a ground layer and a space layer. The ground layer consists of ground stations, which are deployed ground stations, the set is represented by The space layer consists of high-speed mobile users and LEO satellites. LEO satellites, the set is represented by , , each satellite has the same number of beams, and the beam set is For convenience, a new set is defined for each satellite: , then the new two-dimensional vector is .

[0062] Secondly, the construction of the channel model includes the communication model and the switching cost model.

[0063] The communication model is mainly used to characterize the signal quality and transmission efficiency of the UE, involving multiple parameters such as signal strength, signal-to-noise ratio, available bandwidth, etc. By establishing an accurate communication model, the communication quality provided by different satellite beams can be evaluated, providing an important basis for switching decisions.

[0064] The switching cost model focuses on the impact of the switching process itself, including switching delay and signaling overhead. Switching delay directly affects user experience, especially the impact on real-time services; while signaling overhead is related to the utilization efficiency of network resources. By establishing a switching cost model, the benefits and costs of switching can be weighed in the decision-making process to avoid too frequent switching.

[0065] In specific implementation, the communication model is constructed based on the satellite coverage information of the user, taking into account the pitch angle fading, path loss, atmospheric fading and Richter small-scale fading. is a binary indicator, Indicates whether the user requests access to the satellite beam in time slot t , that is, when When the user requests to access the satellite beam , in other cases . represents the satellite accessed by the user in time slot t, represents the beam accessed by the user in time slot t.

[0066] The communication model is mainly composed of elevation angle fading, path loss, atmospheric fading and Richter small-scale fading. The specific expression is as follows: ; In the formula, is the carrier frequency, is the speed of light, is the propagation distance between the transmitter and the receiver. Rician small-scale fading .

[0067] Pitch angle fading It is expressed as: ; in, is the antenna aperture efficiency, The antenna factor determined by the coverage radius is the distance between the user and the beam. The pitch angle between.

[0068] Atmospheric decay The expression is as follows: ; in, is the attenuation through clouds and rain, expressed in dB / km.

[0069] Furthermore, the transmission rate can be derived from the communication model; When in time slot t, the user chooses to access the beam , then the beam in time slot t The signal-to-noise ratio (SNR) at the user is: ; in, is the transmission power, is the channel model, is the Gaussian white noise power.

[0070] According to Shannon's formula, we can get the beam The achievable transmission rate to the user is: ; Where B is the bandwidth of the shared spectrum, is the beam in time slot t The signal-to-noise ratio (SNR) at the user.

[0071] Furthermore, in order to ensure that real-time services can be transmitted on time, the time delay function is modeled: the switching delay consists of waiting time and propagation time. ;in Represents the waiting time from the time the switch occurs to the time the currently accessed satellite beam retrieves the requested data frame from the ground station. is the propagation time necessary for the signal interaction.

[0072] Furthermore, in order to avoid frequent switching, a switching cost model is modeled. If the user's goal is to maximize the instantaneous throughput in each time slot, the user may change the selection of satellite beams in each time slot, resulting in frequent switching. In order to avoid frequent switching, the switching signaling overhead function can be introduced to construct a switching cost model.

[0073] ; in, is the switching overhead of the user in time slot t, represents the satellite accessed by the user in time slot t, represents the satellite accessed by the user in time slot t-1, represents the beam accessed by the user in time slot t, Represents the beam accessed by the user in time slot t-1.

[0074] Step 302: Define the satellite beam switching problem and constraints at the UE end according to the system model and the channel model.

[0075] The satellite beam switching problem includes maximizing the user throughput, minimizing the switching delay and minimizing the signaling cost; the constraints include that the UE has at least one connectable satellite beam in any time slot, and the UE has one satellite beam connected in a time slot.

[0076] Specifically, after building the system model and channel model, the satellite beam switching problem and its constraints at the UE end need to be clearly defined based on these models. This converts the complex practical problem into a formal expression that can be solved by mathematical methods, providing clear goals and boundaries for subsequent optimization algorithms. By accurately defining the problem, multiple performance indicators can be better balanced while ensuring the practicality and feasibility of the solution.

[0077] Furthermore, when defining the satellite beam switching problem, multiple key performance indicators need to be considered comprehensively. The first is to maximize the user's throughput, which is directly related to the user's service quality experience, especially for data-intensive non-real-time services. The second is to minimize the switching delay, which is crucial to maintaining service continuity, especially for real-time services such as voice calls or video streaming. Finally, the signaling cost needs to be minimized, because frequent switching will increase the signaling burden of the network, which may lead to a decrease in the overall system efficiency. These three goals may conflict with each other. For example, frequent switching may increase throughput but increase latency and signaling costs. Therefore, the definition of the problem needs to seek a balance between these goals.

[0078] When implementing, define represents the maximization of the long-term throughput of all users, and defines Indicates the delay of signaling allocation during switching, defined as Minimization of switching overhead when switching on behalf of all users, Represents the weighted total reward function.

[0079]

[0080] in, Indicates whether the user requests access to the satellite beam in time slot t , Indicates that in time slot t, from beam The achieved transmission rate to the user, is the switching signaling overhead of the user in time slot t.

[0081] In addition, setting appropriate constraints is a key step to ensure the practicality and feasibility of the solution. The two core constraints proposed in the present invention, namely, the existence of at least one connectable satellite beam at the UE end in any time slot and the corresponding satellite beam connected by the UE end in one time slot, are crucial for building a satellite communication system that meets practical needs and is easy to manage.

[0082] First, the UE is required to have at least one connectable satellite beam in any time slot. This constraint is set to ensure the continuity of communication services. In satellite communication networks, the coverage area of ​​satellites is constantly changing due to the high-speed movement of satellites and the rotation of the earth. Without this constraint, the UE may be temporarily unable to connect to any satellite beam, resulting in communication interruption. By introducing the above constraints, the optimization algorithm can be forced to consider maintaining continuous connection when making switching decisions, thereby avoiding service interruptions and improving user experience.

[0083] Secondly, the second constraint requires that the UE can only connect to one satellite beam in a time slot. The setting of the above constraints is mainly based on the following considerations: First, from the perspective of technical implementation, most UE devices are usually equipped with only a single communication module, which makes it difficult to maintain connections with multiple satellite beams at the same time. Second, even if the device can support multiple connections, connecting to multiple satellite beams at the same time will greatly increase the complexity and resource consumption of the system. Finally, a single connection can simplify the network management and resource allocation process and improve the overall efficiency of the system.

[0084] Exemplarily, constraint condition 1 is defined: the user must select a satellite beam for access.

[0085] ; Define constraint 2: The user can only select one satellite beam for access: .

[0086] Based on the above satellite beam switching problem, in step 103, a matching optimization algorithm is determined according to the current computing power. According to the optimization algorithm and the prediction data, the corresponding satellite beam switching problem under the CU-DU architecture is solved to obtain a switching strategy, which can be specifically divided into the following two cases: If the current computing power is less than or equal to the threshold, the optimization algorithm is determined to be a greedy algorithm; if the current computing power is greater than the threshold, the optimization algorithm is determined to be a deep learning algorithm.

[0087] Specifically, based on the satellite beam switching problem, the present invention adopts an adaptive optimization algorithm selection mechanism to determine the most suitable optimization algorithm according to the current computing power of the CU end. The purpose is to ensure the timeliness and effectiveness of the switching decision under different hardware environments and network load conditions.

[0088] The advantage of the greedy algorithm is that it is fast in computation, simple to implement, and can make decisions quickly when computing resources are limited. However, it may not be able to find the global optimal solution, especially in scenarios where long-term benefits need to be considered. The advantage of the deep learning algorithm is that it can learn complex decision-making strategies, take long-term benefits into account, and can adapt to dynamically changing environments. But it also requires more computing resources and training time.

[0089] In general, by dynamically selecting the optimization algorithm according to the computing power, the present invention achieves a balance between efficiency and performance. When resources are limited, the system can still guarantee basic service quality through the greedy algorithm; when resources are sufficient, it can pursue higher performance through the deep learning algorithm.

[0090] Based on the above embodiment, as an optional embodiment, when the optimization algorithm is a greedy algorithm, step 103 may specifically include the following steps: Step 401: Calculate the instantaneous benefit of the UE accessing each candidate satellite relative to the optimization objective function based on the prediction data.

[0091] Specifically, under the condition of limited computing resources, the present invention adopts a greedy algorithm to solve the satellite beam switching problem. The above method can make decisions quickly within a limited time and ensure the responsiveness of the system. The core idea of ​​the greedy algorithm is to select the current optimal option at each decision point. Although it may not be able to achieve the global optimal solution, it can achieve a good balance between computational complexity and decision quality.

[0092] Firstly, according to the prediction data such as the position of the UE in the next time slot and the channel quality predicted by the LSTM network, the instant benefit of the UE accessing each candidate satellite relative to the optimization objective function is calculated.

[0093] In specific implementation, since the movement trajectories of satellites and users are predictable, the states between different time slots have temporal dependencies. The LSTM network can be used to extract the temporal dependency features, which is conducive to finding the optimal solution to the switching problem. First, the state matrix satellite and user position information and channel state of each historical time slot are taken as input: ; After being input into the LSTM network for state aggregation, the terminal can understand the hidden change rules in the trajectory of the nodes by extracting historical information from the continuous state. Finally, the LSTM network outputs the predicted and extracted satellite and user location information aggregation for the next time slot, which is supported by the algorithm's pre-operation expression as follows: ; After that, based on the output Ht+1 of the LSTM network and the status of the connected nodes at the current time slot t, the instantaneous benefit of each candidate satellite beam can be calculated. The instantaneous benefit function R(b) comprehensively considers the throughput, switching delay and number of switching times:

[0094] Aggregate state extracted by LSTM network It can be seen that there are many possible beams that can be accessed by users. A greedy algorithm is used to screen and access the maximum optimized target node in each time slot.

[0095] Furthermore, the state matrix of satellite and user position information and channel quality of several historical time slots can be Input to LSTM network. Output from LSTM network and the connected nodes in this time slot , find the candidate node that can maximize the optimization target as the candidate satellite with the largest immediate benefit.

[0096] Step 402: If there is a candidate satellite with the largest instant benefit, the candidate satellite with the largest instant benefit is determined as the first target satellite beam.

[0097] Step 403: If there is no candidate satellite with the maximum instant benefit, a candidate satellite is randomly determined as the first target satellite beam.

[0098] Specifically, after calculating the instantaneous benefits of all candidate satellites, the system will check whether there is a candidate satellite with the maximum instantaneous benefit. If there is a unique maximum value, the corresponding satellite beam is determined as the first target satellite beam. The above selection reflects the essence of the greedy strategy, that is, selecting the option that currently appears to be the best. However, in actual situations, there may be multiple candidate satellites with the same maximum instantaneous benefit, or the instantaneous benefits of all candidate satellites may differ very little. In order to deal with the above situation, the present invention introduces a random selection mechanism: if there is no unique maximum instantaneous benefit, the system will randomly select one from the candidate satellites with the maximum benefit as the first target satellite beam.

[0099] Step 404: Generate a switching strategy for switching to the first target satellite beam.

[0100] Specifically, after determining the first target satellite beam, the system will generate a corresponding switching strategy. The above strategy includes the target satellite beam information for switching, the expected switching time, and the resource reservation request that may be required. If the first target satellite beam is the same as the currently connected satellite beam, the strategy may be to keep the current connection unchanged; if they are different, the strategy will include specific switching instructions.

[0101] Based on the above embodiment, as another optional embodiment, when the optimization algorithm is a deep learning algorithm, step 103 may specifically include the following steps: Step 501: Define the state space, action space and reward function of the UE end through the Markov decision process model, the state space includes the center position and channel quality of the next time slot in the predicted data, the action space includes at least one candidate satellite beam that the UE end can connect to, and the reward function includes throughput reward, switching delay penalty and switching overhead penalty.

[0102] In the case of sufficient computing resources, the present invention adopts a deep learning algorithm to solve the satellite beam switching problem. The above method can better capture the long-term dynamic characteristics of the system, thereby making better switching decisions. In order to effectively apply the deep learning algorithm, it is necessary to first formalize the problem through the Markov decision process (MDP) model to define the state space, action space and reward function of the UE. The above modeling method can not only accurately describe the dynamic characteristics of the satellite communication system, but also provide a theoretical basis for the subsequent deep reinforcement learning algorithm.

[0103] In the specific implementation, a Markov decision process is constructed to gradually solve the optimization problem. In the reinforcement learning framework, the machine that learns and implements decisions is called an agent. The agent perceives the state from the environment at time t. After that, perform an action To interact with the environment. Updated to And get reward value . Define the state space, action space and reward function of user i.

[0104] For the state space : Define the state space under time slot t as: ;in, ; represents the satellite beam accessed by the user at time slot t, Indicates whether the user requests access to the satellite beam in time slot t .in, ; represents the prediction information output by the LSTM network; where, ; represents the position of the center point of all beams in the next time slot t+1; where, is the beam in time slot t+1 location; Indicates the position of the user in the next time slot t+1; ; represents the channel quality between the user and each beam node at time slot t+1, is the beam in time slot t+1 The signal-to-noise ratio at the user.

[0105] Furthermore, for the action space : In each time slot, the user dynamically adjusts the satellite beam to which it is connected to better ensure the quality of service. Select one node from the beams as the connection target, and set is the action of the user at time slot t, then the action space is defined as: ; in, Indicates whether the user requests access to the satellite beam in time slot t .

[0106] For the reward function :set up Indicates taking action After that, the user’s reward at time t. Since the goal is to maximize long-term throughput, minimize switching latency, and avoid frequent handovers, the reward function is divided into three parts.

[0107] (a) Throughput: For non-real-time data services, focus on the accuracy and efficiency of data transmission. Define the user's throughput reward as: ;in, Indicates whether the user requests access to the satellite beam in time slot t , Indicates that in time slot t, from beam The achieved transfer rate to the user. Reward function It represents the total throughput of the system in the current time slot. The larger the throughput, the higher the reward value of the agent and the higher the efficiency of data transmission. Conversely, the smaller the throughput, the smaller the reward value, the lower the efficiency of data transmission and the worse the user experience.

[0108] (b) Switching delay: For real-time data services, the real-time nature of data processing is important. The switching delay of a user is defined as: ;in, represents the waiting time from the time the switch occurs to the time the currently connected satellite beam retrieves the requested data frame from the ground station, represents the propagation time of the necessary signal interaction. Reward function Indicates the total switching delay of the system in the current time slot. The longer the delay, the lower the reward value obtained by the agent and the worse the real-time performance of data processing. Conversely, the shorter the delay, the greater the reward value obtained, the higher the real-time performance of data processing, and the better the user experience.

[0109] (c) Switching penalty: In order to minimize the frequent switching caused by improving data transmission performance, a switching cost reward function is defined: ; in, represents the satellite accessed by the user in time slot t, represents the satellite accessed by the user in time slot t-1, represents the beam accessed by the user in time slot t, represents the beam accessed by the user at time slot t-1. Reward function represents the switching cost of the user in the current time slot. If the satellite beam accessed by the user in the current time slot is the same as that in the previous time slot, the switching cost of the current time slot t is 0; if the user accesses the same satellite but different beams in the current time slot as in the previous time slot, that is, beam switching is performed, the switching cost of the current time slot t is ; If the user accesses a different satellite in the current time slot than in the previous time slot, i.e., an inter-satellite handover is performed, then the handover cost in the current time slot t is .

[0110] Step 502: Build a deep learning network, receive the state of the current time slot through the input layer of the deep learning network, extract state features through the convolution layer and the pooling layer, and generate action value through the fully connected layer and the output layer.

[0111] Specifically, after defining the Markov decision process model, the next step is to build a deep learning network that can effectively handle the above problems. The main task of the above network is to learn the mapping relationship between state and action, that is, to evaluate the value of different actions for a given state. In the above complex environment of satellite communications, the deep learning network can capture the nonlinear characteristics in the state space, so as to make more intelligent and forward-looking decisions.

[0112] In an optional implementation, you can choose to build a deep Q network, which has proven its effectiveness in dealing with reinforcement learning problems in high-dimensional state spaces. The core idea of ​​DQN is to use a deep neural network to approximate the Q function, i.e., the state-action value function. The input of the network is the state of the current time slot, and the output is the Q value estimate of each possible action.

[0113] The specific network structure design fully considers the characteristics of the satellite communication system. First, the status information of the current time slot is received through the input layer. This data can be represented as a multi-dimensional vector or matrix. Then, multiple convolutional layers are used to extract features from the input data. The convolution operation is particularly suitable for processing data with spatial correlation, such as the location information of the UE and the distribution of surrounding satellite beams. It can help the network capture the local correlation between spatial position and channel quality. After the convolution layer, the pooling layer is introduced to perform the maximum pooling operation, which can not only reduce the feature dimension and extract the most significant features, but also reduce the amount of calculation and improve the robustness of the model to small position changes.

[0114] The network then goes through a series of fully connected layers for further feature extraction and combination. These layers are able to learn higher-level abstract features and enhance the network's nonlinear expression capabilities, thereby capturing the complex interactions between factors such as communication quality and switching overhead. Finally, the number of neurons in the output layer is equal to the size of the action space, that is, the number of connected satellite beams, which directly corresponds to the Q-value estimate for each action.

[0115] Step 503: Based on the strategy exploration and action space, determine the action with the maximum action value as the optimal action.

[0116] Optionally, an ε-greedy strategy can be used. At each time slot, the current state is first input into the evaluation network to obtain a Q-value estimate for each possible action. Then, a random number can be generated. If the number is less than ε, an action is randomly selected from the action space; otherwise, the action with the largest Q value is selected. The above method ensures the use of the known optimal strategy while maintaining a certain degree of random exploration.

[0117] The ε value usually decays over time, from a higher initial value to a smaller fixed value. This allows the system to explore more in the early stages of learning and exploit more of the learned knowledge in the later stages. The initial value, minimum value, and decay rate of the ε value can be adjusted according to actual needs to strike a balance between exploration and exploitation.

[0118] After executing the selected action, the feedback from the environment is observed, including the new state and the reward obtained. This information is used to update the parameters of the Q network, thereby continuously improving the decision-making strategy. In this way, the system can learn which actions should be selected in different states to obtain the maximum long-term benefits.

[0119] In the specific implementation, the behavior subject is regarded as the intelligent agent, and the various peripheral elements that continuously interact with the intelligent agent are called the environment. The state of the intelligent agent at time slot t is , according to the random strategy Select Action After the environment responds to the action, the agent enters the next state , and get a reward in return The cumulative discount of the reward is , is the goal that the agent wants to maximize, where is the discount factor. The value function From the state Initially, the agent uses the strategy The expected return of making an action selection is: ;in, represents the discounted reward in time slot t, Indicates the state in which the agent chooses the current environment As the state of an agent , is the discount factor.

[0120] Using Temporal Difference (TD) learning to decompose the above estimate: ; in, represents the TD target, Indicates TD error.

[0121] Using state action value function Evaluate the state-action effect and provide guidance for action selection in a certain state.

[0122] ; in, Indicates time slot The following discount rewards, Indicates the state in which the agent chooses the current environment As the state of an agent , Indicates that the agent chooses to perform an action Actions as Agents , Indicates a strategy. is the discount factor.

[0123] When the agent is in state Follow a strategy Take Action Get the maximum reward, then the current behavior value function is called the optimal behavior value function ,Right now: ;in, Represents a discount reward, Indicates the status, Indicates action, Indicates strategy.

[0124] By executing actions through the greedy algorithm, we can continuously find a way to improve the Q value and achieve the effect of improving the strategy: ; The Q function is updated in a similar way to the value function, using the current state Update the Q function: ;in, represents the TD target, Indicates TD error.

[0125] Deep Q-learning uses an evaluation network and a target network to estimate Q values. The evaluation network is used to select actions based on the current state. The target network is responsible for calculating the TD target for Q value updates. The two networks have the same structure but different network parameters. The input is the state of the agent and the output is the Q value corresponding to each action. The network parameters of the evaluation network are , which is updated with each iteration to reduce the gap with the target value. The network parameters of the target network are , the network parameters will be evaluated every certain number of steps Copy to the target network to increase the stability of learning.

[0126] Network parameters The update formula is: .

[0127] Step 504: Use the candidate satellite beam corresponding to the optimal action as the second target satellite beam, and generate a switching strategy for switching to the second target satellite beam.

[0128] Specifically, after determining the second target satellite beam, the system will generate a corresponding switching strategy. The above strategy includes the target satellite beam information for switching, the expected switching time, and the resource reservation request that may be required. If the second target satellite beam is the same as the currently connected satellite beam, the strategy may be to keep the current connection unchanged; if they are different, the strategy will include specific switching instructions.

[0129] Step 104: Send the switching strategy to the DU end. The switching strategy is used by the UE end to switch the satellite beam.

[0130] Based on the above embodiment, as an optional embodiment, the present invention also provides a satellite beam pre-switching method, which aims to optimize the satellite beam switching process under the CU-DU separation architecture by introducing a new pre-switching mechanism. This mechanism is applicable to the case where the switching strategy involves a new CU end, which can significantly reduce the switching delay, improve the switching success rate, and minimize the service interruption time.

[0131] Specifically, before sending the switching strategy to the DU end, the following steps may also be included: Step 601: If the switching strategy involves a new CU end, a pre-switching request is sent to the new CU end. When a pre-switching confirmation instruction is received from the new CU end, a context update request is sent to the new DU end in the switching strategy, and a context modification instruction is sent to the DU end, so that the DU end completes the pre-switching configuration to the UE end.

[0132] In specific implementation, when the switching strategy generated by the CU end according to the optimization algorithm involves a new CU end, the original CU end first sends a pre-switching request to the new CU end. The above request contains key information of the UE end, such as the current service quality parameters, expected resource requirements, and switching time window. After receiving the request, the new CU end will evaluate its current resource status and network load to determine whether it can meet the service requirements of the UE end. If the new CU end can accommodate the UE end, it will return a pre-switching confirmation instruction to the original CU end. The above pre-coordination process allows the network to coordinate resource allocation before the actual switching occurs, effectively reducing the risk of resource competition and conflict that may occur during switching.

[0133] After receiving the pre-switching confirmation instruction from the new CU, the original CU immediately sends a context update request to the new DU specified in the switching strategy. The above request contains the complete context information of the UE, including but not limited to the current wireless configuration, QoS parameters, security context, etc. By transmitting this information in advance, the new DU can prepare the necessary resources for the upcoming UE in advance, such as allocating appropriate wireless channels, preparing data buffers, etc. The above advance preparation mechanism ensures that the new DU can quickly respond to the UE's access request during the actual switching, thereby minimizing the service interruption time during the switching process.

[0134] Next, the original CU sends a context modification instruction to the currently serving DU. The purpose of the above instruction is to instruct the current DU to configure pre-switching parameters for the UE. These parameters usually include new target cell information, switching trigger conditions, measurement configuration, etc. After receiving the above instruction, the DU generates an RRC reconfiguration message containing the pre-switching configuration and sends it to the UE. This allows the UE to prepare for possible switching while maintaining the current connection.

[0135] Step 602: If the switching strategy does not involve a new CU end, a context update request is sent to a new DU end in the switching strategy, and a context modification instruction is sent to the DU end, so that the DU end completes the pre-switching configuration to the UE end.

[0136] When the switching strategy generated by the CU side according to the optimization algorithm does not involve the new CU side, the entire pre-switching process will be simpler and more efficient. In the above case, the CU side first sends a context update request to the new DU side specified in the switching strategy. This request contains the complete context information of the UE side, including but not limited to the current wireless configuration, QoS parameters, security context and other key data. By transmitting this information in advance, the new DU side can prepare the necessary resources for the upcoming UE side in advance, such as allocating appropriate wireless channels, preparing data buffers, etc. The above-mentioned advance preparation mechanism ensures that during the actual switching, the new DU side can quickly respond to the UE side's access request, thereby minimizing the service interruption time during the switching process.

[0137] Next, the CU sends a context modification instruction to the currently serving DU. The purpose of this instruction is to instruct the current DU to configure pre-switching parameters for the UE. These parameters usually include new target cell information, switching trigger conditions, measurement configuration, etc. After receiving this instruction, the current DU generates an RRC reconfiguration message containing the pre-switching configuration and sends it to the UE. This step enables the UE to prepare for possible switching while maintaining the current connection.

[0138] After receiving the pre-switching configuration, the UE will complete the synchronization and measurement of the new cell in the background, but will not perform the actual switching operation. When the actual triggering switching conditions are met (for example, the current signal quality is lower than the preset threshold or the new cell signal is significantly better than the current cell), the UE can quickly perform the switching, greatly shortening the service interruption time during the switching process.

[0139] The above-mentioned pre-switching mechanism between different DUs in the same CU brings many benefits. First, since there is no coordination between CUs, the entire pre-switching process is simplified, further reducing the switching delay. Secondly, since the CU has a more comprehensive understanding of all DU resources under its jurisdiction, it can reserve and allocate resources more accurately, thereby improving the success rate of switching. Furthermore, the above-mentioned mechanism is particularly suitable for handling intra-beam switching or switching between adjacent beams in satellite communications, which is a very frequent operation in low-orbit satellite systems.

[0140] Step 603: If the switching strategy does not involve a new CU and a new DU, the CU provides a new uplink path endpoint identifier. The DU provides a new downlink path endpoint identifier and uses the previous uplink path to send an uplink packet data convergence protocol data unit to the CU until a new radio link control is established.

[0141] At this time, the DU uses the new UL GTP TEID to send messages. The CU continues to use the previous downlink endpoint identifier to send DL PDCP-DU to the downlink packet data convergence protocol data unit until PDCP reestablishment or PDCP data recovery is performed, and then starts to send using the new downlink endpoint identifier. The entire pre-switching process is the most concise and reduces the time delay of signaling transmission.

[0142] Based on the above examples, please refer to Figure 5 , Figure 5 A schematic diagram of a network architecture and switching classification scenario under a CU-DU separation architecture provided by the present invention. Figure 5 Based on Figure 6 and Figure 7 , Figure 6 A pre-switching flow chart of different DUs of the same CU under a CU-DU separation architecture provided by an embodiment of the present invention; Figure 7 A pre-switching flowchart of different CU ends and different DU ends under a CU-DU separation architecture provided in an embodiment of the present invention.

[0143] The following are combined Figure 5 , Figure 6 as well as Figure 7 , examples are given for the pre-switching command interaction of each end in the above three situations.

[0144] like Figure 6 As shown, the pre-switching process of different DU ends at the same CU end may specifically include the following steps: Step 701: The UE sends a measurement report message to the source DU.

[0145] Step 702: The source DU sends a UL RRC message transmission message to the CU to convey the received measurement report message.

[0146] Step 703: The CU initiates a pre-switching decision.

[0147] Step 704: The CU initiates a context update request to the target DU.

[0148] Step 705: The UE context modification process performs the UE context modification process between the CU and the source DU to modify the established UE context and instruct the source DU to stop data transmission at the UE.

[0149] Step 706: The source DU forwards the RRC reconfiguration message to the user.

[0150] Step 707: The UE responds to the CU with an RRC reconfiguration completion message.

[0151] Step 708: The UE implements pre-switching configuration.

[0152] Step 709: The UE meets the triggering condition for handover. The triggering condition follows the A3 event in 3GPP TS38.331, that is, the neighboring cell is higher than the primary serving cell.

[0153] Step 710: Perform a random access process on the candidate DU. If successful, the candidate DU becomes the target DU. The target DU sends a downlink data transmission status frame to notify the CU. The target DU also sends an access success message to notify the UE that it has successfully accessed the CU.

[0154] Step 711: The UE responds to the target DU with an RRC reconfiguration completion message.

[0155] Step 712: The target DU sends a UL RRC message transmission message to the CU to convey the received RRC reconfiguration complete message.

[0156] Step 713: The CU sends a UE context modification request message to the source DU and instructs the source DU to stop the user's data transmission. The source DU also sends a downlink data transmission status frame to notify the CU to send failed downlink data to the UE.

[0157] Step 714: The source DU responds to the CU with a UE context modification response message.

[0158] Step 715: After the target DU configures available resources for the user, the user disconnects from the source DU and starts synchronization with the target DU.

[0159] Step 716: The CU sends a context release message to the UE to inform the source DU that the link switching has been completed.

[0160] like Figure 7 As shown, the pre-switching process between different CU ends and different DU ends may specifically include the following steps: Step 801: The UE sends a measurement report message to the source DU.

[0161] Step 802: The source DU sends a UL RRC message transmission message to the source CU to convey the received measurement report message.

[0162] Step 803: The source CU initiates a pre-switching decision.

[0163] Step 804: The source CU decides to hand over to the UE according to the measurement report, ephemeris information, etc. After deciding to switch, the source CU sends a switching request message to the target CU to request preparation of resources for switching.

[0164] Step 805: The target CU receives the pre-switching request message, allows the switching and reserves resources for the switching.

[0165] Step 806: Perform a UE-side context setup process between the target CU and the target DU to establish a UE-side context.

[0166] Step 807: The target CU returns a pre-switching request confirmation message to the source CU.

[0167] Step 808: Modify the established UE context between the source DU and the source CU, and instruct the source DU to stop the UE's data transmission.

[0168] Step 809: The pre-switching configuration at the source CU is completed.

[0169] Step 810: The source DU forwards the RRC reconfiguration message to the user.

[0170] Step 811: The UE responds with an RRC reconfiguration completion message.

[0171] Step 812: The UE implements pre-switching configuration.

[0172] Step 813: The UE meets the triggering condition for handover. The triggering condition follows the A3 event in 3GPP TS38.331, that is, the neighboring cell is higher than the primary serving cell.

[0173] Step 814: Perform a random access process on the candidate DU end under the target CU end. The target DU end sends a downlink data transmission status frame to notify the target CU end. The target DU end also sends an access success message.

[0174] Step 815: The UE responds to the target DU with an RRC reconfiguration completion message.

[0175] Step 816: The target DU sends a UL RRC message transmission message to the target CU to convey the received RRC reconfiguration complete (RRCReconfigurationComplete) message.

[0176] Step 817: The target CU sends a user context modification request message to the source CU and instructs the UE to stop data transmission. The source CU also sends a downlink data transmission status frame to notify the target CU of the failed downlink data transmission to the UE.

[0177] Step 818: The target CU sends a user context release request message to the source CU to indicate that resources can be released.

[0178] Step 819: Execute the UE-side context release process between the source CU and the source DU to release the connection related to the user.

[0179] Step 820: The UE disconnects from the source DU and starts synchronizing with the target DU.

[0180] If the switching strategy does not involve a new CU end and a new DU end, a pre-switching process within the same CU end and the same DU end is performed, which may specifically include the following steps: Step 901: The CU provides a new uplink path endpoint identifier (UL GTP TEID) to the DU.

[0181] Step 902: The DU provides a new downlink path endpoint identifier (DL GTP TEID) and uses the previous UL GTP TEID to send an uplink packet data convergence protocol data unit (UL PDCP PDU) to the CU until a new radio link control (RLC) is established.

[0182] Step 903: The DU sends a message using the new UL GTP TEID.

[0183] Step 904: The CU continues to use the previous downlink DL GTP TEID to send a DL PDCP-DU to a downlink packet data convergence protocol data unit (DL PDCP PDU) until PDCP reestablishment or PDCP data recovery is performed, and then starts to send using the new DL GTP TEID.

[0184] Based on the above examples, please refer to Figure 8 , Figure 8 A flowchart of another multi-satellite beam switching method based on a CU-DU architecture provided in an embodiment of the present invention is provided. The method is applied to a DU end and may specifically include the following steps: Step 1001: Send a switching request to the CU.

[0185] Step 1002: Receive the switching strategy fed back by the CU end, where the switching strategy is obtained by solving the corresponding satellite beam switching problem under the CU-DU architecture using an optimization algorithm and prediction data that matches the current computing power of the CU end, wherein the prediction data is the prediction data of the UE end in the next time slot obtained by predicting the user data through the LSTM network.

[0186] Step 1003: Instruct the UE to switch satellite beams according to the switching strategy.

[0187] Specifically, when the DU end is another execution subject and implements the multi-satellite beam switching method based on the CU-DU architecture, its basic implementation principles and processes are generally consistent with those of the CU end, and reference may be made to the relevant embodiments in which the CU end is the execution subject in the above embodiments.

[0188] Fig. 9 An example of a physical structure diagram of an electronic device is shown in FIG. Fig. 9 As shown, the electronic device may include: a processor 1010, a communication interface 1020, a memory 1030 and a communication bus 1040, wherein the processor 1010, the communication interface 1020, and the memory 1030 communicate with each other through the communication bus 1040. The processor 1010 may call the logic instructions in the memory 1030 to execute a multi-satellite beam switching method based on the CU-DU architecture.

[0189] In addition, the logic instructions in the above-mentioned memory 1030 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0190] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the multi-satellite beam switching method based on the CU-DU architecture provided by the above methods.

[0191] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to execute the multi-satellite beam switching method based on the CU-DU architecture provided by the above methods.

[0192] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0193] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0194] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-satellite beam switching method based on CU-DU architecture, characterized in that: Applied to a CU end, the CU end establishes a communication connection with a DU end, the DU end establishes a communication connection with a UE end through a satellite beam, the UE end, the CU end, and the DU end constitute a CU-DU architecture, and the multi-satellite beam switching method based on the CU-DU architecture includes: Receiving a handover request sent by the DU, and obtaining user data of the UE in the handover request; Predicting the user data through the LSTM network to obtain predicted data of the UE in the next time slot; Determine an optimization algorithm that matches the current computing capability of the CU end, and solve the corresponding satellite beam switching problem under the CU-DU architecture according to the optimization algorithm and the prediction data to obtain a switching strategy; The switching strategy is sent to the DU end, and the switching strategy is used to instruct the UE end to switch the satellite beam.

2. The multi-satellite beam switching method based on CU-DU architecture according to claim 1, characterized in that: The user data includes the center position and channel quality of n historical time slots, and the prediction of the user data by the LSTM network to obtain the prediction data of the UE in the next time slot includes: Performing state aggregation on the center positions and channel qualities of the n historical time slots through an LSTM network to obtain the center position and channel quality of the UE end in the next time slot; The central position and channel quality of the UE end in the next time slot are used as prediction data of the UE end in the next time slot.

3. The multi-satellite beam switching method based on CU-DU architecture according to claim 1, characterized in that: When the current computing power is less than or equal to a threshold, the optimization algorithm is a greedy algorithm, and solving the optimization objective function corresponding to the satellite beam switching under the CU-DU architecture according to the optimization algorithm and the prediction data to obtain a switching strategy includes: Calculate the instantaneous benefit of the UE accessing each candidate satellite relative to the optimization objective function according to the prediction data; If there is a candidate satellite with the maximum instant benefit, determining the candidate satellite with the maximum instant benefit as the first target satellite beam; If there is no candidate satellite with the maximum instant benefit, randomly determining a candidate satellite as the first target satellite beam; A switching strategy for switching to the first target satellite beam is generated.

4. The multi-satellite beam switching method based on CU-DU architecture according to claim 1, characterized in that: When the current computing power is greater than a threshold, the optimization algorithm is a deep learning algorithm, and solving the optimization objective function corresponding to the satellite beam switching under the CU-DU architecture according to the optimization algorithm and the prediction data to obtain a switching strategy includes: The state space, action space and reward function of the UE are defined by a Markov decision process model, wherein the state space includes the center position and channel quality of the next time slot in the predicted data, the action space includes at least one candidate satellite beam to which the UE can connect, and the reward function includes a throughput reward, a switching delay penalty and a switching overhead penalty; Build a deep learning network, receive the state of the current time slot through the input layer of the deep learning network, extract state features through the convolution layer and pooling layer, and generate action value through the fully connected layer and output layer; Based on strategy exploration and action space, the action with the largest action value is determined as the optimal action; The candidate satellite beam corresponding to the optimal action is used as a second target satellite beam, and a switching strategy for switching to the second target satellite beam is generated.

5. The multi-satellite beam switching method based on CU-DU architecture according to claim 1, characterized in that: Before determining the optimization algorithm that matches the current computing power, the method further includes: Acquire the interaction information of the CU-DU architecture to build a system model and a channel model; The system model includes a user model and a dynamic model, wherein the user model is used to characterize the real-time performance of user data on the UE side, and the dynamic model is used to characterize the relative position relationship between the UE side and the satellite; The channel model includes a communication model and a switching cost model, wherein the communication model is used to characterize the signal instructions and transmission efficiency of the UE end, and the cost model is used to characterize the switching delay and signaling cost of switching satellite beams at the UE end; Defining the satellite beam switching problem and constraint conditions at the UE end according to the system model and the channel model; The satellite beam switching problem includes maximizing user throughput, minimizing switching delay, and minimizing signaling cost; The constraint conditions include that the UE has at least one connectable satellite beam in any time slot, and that the UE has a corresponding satellite beam connected in one time slot.

6. The multi-satellite beam switching method based on CU-DU architecture according to claim 1, characterized in that: Before sending the switching strategy to the DU end, the method further includes: If the switching strategy involves a new CU end, a pre-switching request is sent to the new CU end. When a pre-switching confirmation instruction is received from the new CU end, a context update request is sent to the new DU end in the switching strategy, and a context modification instruction is sent to the DU end, so that the DU end completes the pre-switching configuration to the UE end. If the switching strategy does not involve a new CU end, a context update request is sent to a new DU end in the switching strategy, and a context modification instruction is sent to the DU end, so that the DU end completes the pre-switching configuration to the UE end.

7. A multi-satellite beam switching method based on CU-DU architecture, characterized in that: Applied to a DU end, the DU end establishes a communication connection with a CU end, the DU end establishes a communication connection with a UE end through a satellite beam, the UE end, the CU end, and the DU end constitute a CU-DU architecture, and the multi-satellite beam switching method based on the CU-DU architecture includes: Sending a switching request to the CU end; Receive a switching strategy fed back by the CU end, where the switching strategy is obtained by solving the corresponding satellite beam switching problem under the CU-DU architecture using an optimization algorithm and prediction data that match the current computing power of the CU end, wherein the prediction data is prediction data of the UE end in the next time slot obtained by predicting the user data through an LSTM network; The UE is instructed to switch the satellite beam according to the switching strategy.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the multi-satellite beam switching method based on the CU-DU architecture as described in any one of claims 1 to 7 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the multi-satellite beam switching method based on the CU-DU architecture as claimed in any one of claims 1 to 7 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the multi-satellite beam switching method based on the CU-DU architecture as claimed in any one of claims 1 to 7 is implemented.

Images