A ground-non-ground fusion network resource allocation method and system based on an upper confidence bound algorithm

Abstract

This invention discloses a method and system for resource allocation in a terrestrial-non-terrestrial integrated network based on the upper confidence bound algorithm, belonging to the field of wireless communication technology. The method includes: first, dynamically evaluating and selecting access to a terrestrial or non-terrestrial network based on the location information and link quality indicators of user equipment; second, constructing a composite reward function that integrates throughput, latency, collision probability, and coverage reachability; further, using the upper confidence bound algorithm to calculate the confidence score of each satellite link to balance "utilization" and "exploration," intelligently selecting the optimal access satellite; and third, effectively suppressing multi-user channel conflicts through channel load broadcasting, backoff and reselection mechanisms, and access admission control. This invention can adapt to dynamic network changes, improve the resource utilization efficiency, system throughput, and link stability of the terrestrial-ground integrated network, and is applicable to scenarios such as 5G / 6G mobile communication and emergency communication.

Description

Technical Field

[0001] This invention belongs to the field of cellular wireless communication and satellite network fusion scheduling. It designs a multi-user resource allocation method and system for a fusion communication system of terrestrial network (TN) and non-terrestrial network (NTN) based on the Upper Confidence Bound (UCB) algorithm. Background Technology

[0002] With the large-scale deployment of fifth-generation mobile communication technology (5G), global wireless communication networks are gradually evolving towards higher bandwidth, lower latency, and wider coverage. However, traditional terrestrial cellular communication systems are limited by infrastructure construction capabilities and geographical factors, making it difficult to achieve comprehensive coverage in vast rural areas, remote mountainous regions, marine areas, and air and rail transportation environments. To address these coverage blind spots, NTN, especially Low Earth Orbit (LEO) satellite networks, is widely considered an important supplementary means to 5G and future sixth-generation communication systems (6G). Compared to Medium Earth Orbit (MEO) and Geostationary Earth Orbit (GEO) satellite systems, LEO satellites have a significantly lower orbital altitude, resulting in shorter transmission latency (20-45 ms) and higher link quality. Furthermore, due to the dynamic trajectory characteristics of LEO satellites, multiple satellites can form a large-scale constellation system, achieving high-frequency coverage of the Earth's surface at all times and in all areas, effectively compensating for the communication capabilities of areas that terrestrial base stations cannot reach. This TN-NTN converged network has become an important research direction of concern for international telecommunications standardization organizations such as the 3rd Generation Partnership Project (3GPP) and the International Telecommunication Union (ITU).

[0003] TN-NTN convergence is not only valuable in conventional scenarios but also demonstrates unique advantages in special scenarios such as emergency communications and disaster relief. For example, when natural disasters such as earthquakes, floods, and typhoons cause ground base stations to fail or backbone networks to be interrupted, satellite networks can serve as backup communication paths, ensuring command communications and emergency response. Currently, against the backdrop of the advancement of the Integrated Communication, Navigation, and Sensing (ICNS) architecture, satellite communications are gradually shifting from a "supplementary role" to a "mainstay of convergence."

[0004] However, the TN-NTN converged system also brings new architectural complexities and scheduling challenges. First, user equipment (UE) in the converged system may access both terrestrial base stations (Next Generation Node B, gNB) and LEO satellites, forming a heterogeneous architecture with multiple paths and access points. Dynamically determining the optimal access path under different spatiotemporal conditions and achieving load balancing and coordination among multiple networks is crucial for the scheduling of the converged system. Second, because LEO satellites continuously orbit the Earth, their visibility window with the UE is highly time-varying. Each satellite has a limited service time for a single location, and link quality is affected by multiple factors such as orbital position, obstruction, interference, and path loss. In many typical scenarios, such as mobile terminals on trains, vehicles, and ships, the UE itself is also in a high-speed movement state, further complicating its reachability relative to the satellite and making the stability and availability of the communication link more unpredictable, posing a significant challenge to resource scheduling. If the scheduling strategy cannot dynamically assess changes in network topology and link status in real time, it can easily lead to resource allocation failures, handover failures, or throughput collapses. Furthermore, the spatial distribution of ground users and their service requests exhibit strong spatiotemporal randomness and non-uniformity, leading to resource overload in some areas and idle resources in others, thus exacerbating the difficulty of link resource scheduling. When multiple UEs compete for limited satellite channel resources, if they simultaneously select the same channel on the same satellite, channel conflicts or interference are highly likely to occur, severely reducing system throughput and reliability.

[0005] To address the aforementioned issues, this invention proposes a resource allocation method for TN-NTN converged networks based on the UCB algorithm. This method comprehensively considers multiple objectives such as throughput, latency, fairness, and coverage reachability, dynamically optimizes the scheduling strategy, and avoids multi-user access conflicts, thereby supporting the robust operation of the converged network in complex scenarios. Summary of the Invention

[0006] Based on the shortcomings of the existing technologies, this invention proposes a multi-user resource intelligent scheduling method for TN-NTN converged networks based on the UCB algorithm. It aims to solve the problems of reduced throughput, increased latency, and uneven service caused by difficulties in multi-path access selection, high link dynamic uncertainty, and frequent channel resource conflicts in existing converged communication networks.

[0007] The technical solution of this invention is: a method for allocating resources in a ground-non-ground fusion network based on the upper confidence bound algorithm, comprising the following steps:

[0008] Step 1, Network Access Status Assessment: Based on the location information and link quality indicators of the user equipment (UE), calculate the terrestrial path score and satellite path score, and determine whether the UE accesses the terrestrial network (TN) or the non-terrestrial network (NTN).

[0009] Step 2, Reward Function Construction: Construct link reward values ​​based on link throughput, latency, collision level, and coverage reachability;

[0010] Step 3, UCB Decision Strategy: Calculate the UCB for each satellite link based on the historical average reward and confidence upper bound. ij The system scores satellite links and selects the one with the highest score for access, among which UCB... ij This indicates that in the UE i Access the j Link score for each satellite;

[0011] Step 4, Multi-UE Channel Conflict Control: Reduce multi-user access conflicts through channel load broadcasting, backoff and polling reselection mechanisms, and access admission threshold control;

[0012] Step 5: Online learning and scheduling optimization: Real-time collection of link performance data, updating of reward function and confidence model to achieve dynamic optimization of scheduling strategy.

[0013] Furthermore, Step 1 also includes constructing an accessible network evaluation function. and , respectively representing the first i Ground path score and satellite path score for each UE:

[0014] ;

[0015] in,( x i , y i , z i ) is User Equipment (UE) i Geographical location, Ω gNB This indicates the coverage area of ​​the ground base station gNB. Ⅱ[•] is an indicator function used to determine whether the location of the UE falls within the coverage area of ​​the ground base station gNB. Norm (•) indicates normalization; and They are UE i Reference Signal Received Power (RSRP) and Signal to Interference plus Noise Ratio (SINR) between the gNB and the gNB.w 1, w 2 and w 3 represents adjustable weights;

[0016] ;

[0017] in, v i It is UE i A collection of visible satellites; and Representing UE i With satellite j The signal-to-interference-plus-noise ratio (SINR) and time delay between the two signals; or 1 and or 2 is an adjustable weight used to balance throughput and latency;

[0018] Constructing binary network access state variables l i ,

[0019] ;

[0020] In other words, if the ground path score is higher, l i =0, then the UE accesses the gNB; otherwise, l i =1, access to LEO satellite.

[0021] Furthermore, in Step 2, when l i =1, calculate UE i With satellite j Link reward value r ij :

[0022] ;

[0023] in, T max This represents the theoretically achievable maximum link throughput under constraints such as system size, bandwidth, and modulation / coding scheme. D max This represents the maximum tolerable communication latency threshold of the system under a specific business scenario. T ij , D ij , C ij and R ij These represent link throughput, communication latency, collision level, and coverage reachability metrics, respectively. α , β, d and This is an adjustable weight. C ij The estimation can be based on historical access behavior statistics or satellite broadcast load information. The "historical access behavior statistics" mentioned in this invention include indicators such as the frequency of each UE's selection of different satellite links, access success rate, number of collisions and their proportion within a statistical period; the "satellite broadcast load information" includes fields such as the number of active connected users on the current link, the historical average congestion rate of each channel, the proportion of remaining available resources, and the remaining service time; the above information can be collected and reported by the UE itself, or periodically broadcast by the satellite node in the control channel as input to the collision assessment function; R ij Used to delineate satellites j UE under current spatiotemporal conditions i Service sustainability and link feasibility; this indicator comprehensively characterizes the link's "whether it can be covered" and "coverage quality," specifically including: 1) Spatial coverage factors: the geometric relationship between the satellite and the UE (such as elevation angle, visibility window). If the elevation angle is too low or about to disappear from the field of view, then... R ij 1) Reduce; 2) Channel reachability factors: whether the link has line-of-sight (LOS) conditions, and the probability of obstruction; 3) Service continuity factors: the remaining service time predicted by the satellite trajectory, and the access success rate of the current link; 4) Resource availability factors (optional): the proportion of remaining satellite broadcast resources, the number of active connections, etc.; These sub-indicators can be normalized and weighted to obtain a unified coverage reachability score. R ij ∈[0,1], the higher the score, the easier it is for the link to achieve stable coverage and continuous communication under the current spatiotemporal conditions; for example, it can be in the following form:

[0024] ;

[0025] in, It is a satellite j For UE i The estimated remaining visibility time, t max This is the maximum serviceable time for reference. i ij It is a satellite j With UE i The current included angle, i max This is the maximum elevation angle, usually taken as 90°. It is the link LOS availability probability. , and The parameters are used to adjust the weights.

[0026] Furthermore, in Step 3, UCB ij The calculation is as follows:

[0027] ;

[0028] Among them, UCB ij This indicates that in the UE i Access the j When there are 1 satellite, the link score is obtained based on the upper confidence bound algorithm. Indicates UE i Access the j The historical average reward value of a satellite t i Indicates UE i Total number of NTN access attempts n ij satellite j The number of times selected, This is the upper bound adjustment term for the confidence level.

[0029] Furthermore, in Step 4, the multi-UE channel conflict control mechanism includes:

[0030] Satellite periodic broadcast channel load information;

[0031] The UE triggers an exponential backoff or polling reselection mechanism when a conflict occurs;

[0032] Set a maximum access capacity threshold for each channel; if the limit is exceeded, access requests will be rejected or redirected.

[0033] Furthermore, in Step 5, the online learning mechanism includes:

[0034] Real-time collection of link performance data;

[0035] Dynamically update the reward function value and confidence model;

[0036] The system periodically evaluates its overall performance and adaptively adjusts the evaluation function and weight parameters.

[0037] The present invention provides a terrestrial-non-terrestrial converged network resource allocation system, comprising:

[0038] User equipment (UE) is used to report location information and link quality.

[0039] Ground-based gNBs provide terrestrial network access services;

[0040] Low Earth Orbit (LEO) satellites provide non-terrestrial network access services;

[0041] The central dispatch and control center is used to execute the resource allocation method and coordinate the allocation of TN and NTN resources.

[0042] The system described in this invention supports TN and NTN collaborative operation: it determines the network type that the UE can currently access based on UE location information and link status indicators. When the accessibility score function of the TN link is greater than that of the NTN link, the TN link is used to access the communication network first; otherwise, the connectivity between the UE and multiple LEO satellites is dynamically evaluated, and data services are completed through the NTN link.

[0043] This invention innovatively introduces the UCB algorithm, jointly modeling multi-dimensional communication performance indicators such as link throughput, transmission latency, and collision probability into a reward function. During user equipment access, the scheduling system dynamically calculates the confidence score of each optional path by learning historical link feedback information in real time, and achieves an adaptive trade-off between maximizing the current average reward and exploring potential high-quality links, thereby continuously optimizing the connection strategy between the UE and the satellite, and realizing the optimal global performance configuration of the converged network in resource-constrained environments.

[0044] Furthermore, this invention addresses the resource conflict caused by multiple UEs simultaneously selecting the same satellite channel by designing an access control mechanism and a channel load information broadcasting strategy. Without introducing centralized global scheduling, it effectively reduces the probability of interference and enhances the multi-user collaborative communication capability.

[0045] The beneficial effects of this invention are:

[0046] This invention proposes a multi-link scheduling mechanism that integrates the UCB algorithm and a composite reward function, targeting the collaborative environment of 5G terrestrial networks and low-Earth orbit satellite non-terrestrial networks, and has the following beneficial effects:

[0047] 1) Improve resource utilization efficiency: By balancing the "exploration-utilization" relationship in multi-path selection through the UCB algorithm, intelligent scheduling of UE and satellite links is achieved, effectively improving the overall system throughput performance.

[0048] 2) Optimize latency experience: Incorporate link latency into the reward function metric, and construct a composite reward standard together with throughput and conflict probability. In scheduling decisions, it dynamically balances the needs of efficient transmission and low latency, which is suitable for application scenarios with high real-time requirements.

[0049] 3) Enhance network adaptability: The proposed mechanism learns the connection performance between different satellites and UEs based on historical observations, and can continuously adapt to dynamic changes in network topology caused by orbital motion, obstruction changes and environmental disturbances, thereby improving system robustness.

[0050] 4) Mitigating channel conflict risks: By designing access control mechanisms and broadcasting satellite load information, UEs are guided to avoid high-load channels, reducing interference caused by multiple users selecting the same satellite and the same channel at the same time, thereby improving link stability and system fairness. Attached Figure Description

[0051] Figure 1 Schematic diagram of TN-NTN converged communication system;

[0052] Figure 2 Flowchart of resource allocation in TN-NTN converged network. Detailed Implementation

[0053] like Figure 1 As shown, the TN-NTN converged communication system of this invention includes the following core components:

[0054] Ground base station (gNB): Provides conventional 5G access services to UEs within the coverage area. It can be configured with 100 MHz bandwidth (FR1) or 400 MHz bandwidth (FR2), with a coverage radius of 0.5–5km and a corresponding downlink peak rate of up to 10Gbps.

[0055] Multiple low Earth orbit (LEO) satellites: can be deployed in orbits at altitudes of 500–1200 km, with a single satellite covering a radius of approximately 1000 km. Typical link latency is 20–45 ms, and downlink bandwidth is in the range of 20–100 MHz, providing ground communication capabilities (especially in areas not covered by gNB signals).

[0056] Satellite Gateway Node: Receives data relayed by satellite and transmits it to the core network or central dispatch and control center. In one optional embodiment, the satellite gateway node can be deployed at a fixed ground site or edge cloud data center, establishes a link with LEO satellite through the Ka / Ku band, and the forwarding capacity of a single node can reach 1–10 Gbps. It is also compatible with the 5G-NR NTN standard interface to achieve interoperability with the core network.

[0057] Central Dispatch and Control Center: Collects UE feedback information and coordinates resource allocation between gNB and LEO satellites. In one optional embodiment, the central dispatch and control center can be deployed in the core network or edge cloud environment, with GPU / FPGA acceleration capabilities to support real-time scheduling decisions of <50 ms. It can periodically collect link performance data, dynamically update reward function weights and access control parameters, and interface with 3GPP management and control functions through standardized interfaces to achieve collaborative resource allocation between gNB and LEO satellites.

[0058] User Equipment (UE): The receiving sensitivity can be approximately -100 dBm, and it supports multi-network handover with a latency of less than 200 ms. It can dynamically select to connect to a terrestrial gNB or satellite based on network coverage and communication quality.

[0059] The present invention will be further described below with reference to the accompanying drawings. The specific steps are as follows (e.g.) Figure 2 ):

[0060] Step 1 Network Access Status Assessment

[0061] Step 1, Network Access Status Assessment: Based on the location information and link quality indicators of the user equipment (UE), calculate the terrestrial path score and satellite path score, and determine whether the UE accesses the TN or NTN accordingly;

[0062] The UE obtains its current geographical location information in real time through an integrated GNSS module (such as GPS) or inertial navigation system and uploads it to the central dispatch and control center or the local gNB processing module. The system maintains a gNB coverage boundary database (e.g., modeled in polygonal or circular form) and determines whether the UE is within the coverage area of ​​any gNB based on its location information. The UE simultaneously listens to the terrestrial gNB's Synchronization Signal Block (SSB) and measures channel quality metrics such as Reference Signal Received Power (RSRP) and Signal-to-Interference Plus Noise Ratio (SINR). The number of visible satellites and satellite channel quality at the UE are recorded. An accessibility network evaluation function is constructed. and , respectively representing the first i Ground path score and satellite path score for each UE:

[0063] ;

[0064] Among them, Ω gNB Indicates the gNB coverage area, ( x i , y i , z i ) is UE i The geographical location, Ⅱ[•] is an indicator function used to determine whether the location falls within the gNB coverage area; Norm (•) indicates normalization; and They are UE i RSRP and SINR between gNB;w 1, w 2 and w 3 represents adjustable weights.

[0065] ;

[0066] in, v i It is UE i A collection of visible satellites; and Representing UE i With satellite j SINR and latency between; or 1 and or 2 is an adjustable weight used to balance throughput and latency.

[0067] Constructing binary network access state variables l i ,

[0068] ;

[0069] In other words, if the ground path score is higher, l i =0, then the UE accesses the gNB; otherwise, l i =1, UE accesses LEO satellite.

[0070] Step 2: Reward Function Construction

[0071] Reward function construction: Construct link reward values ​​based on link throughput, latency, collision level, and coverage reachability;

[0072] Every UE i Connecting the first j After a satellite is launched, the system comprehensively considers link throughput, communication latency, channel conflict risk, and coverage reachability to calculate its link reward value. r ij :

[0073] ;

[0074] in, T max This represents the theoretically achievable maximum link throughput under constraints such as system size, bandwidth, and modulation / coding scheme. D max This represents the maximum tolerable communication latency threshold of the system under a specific business scenario. T ij , D ij , Cij and R ij These represent link throughput, communication latency, collision level, and coverage reachability metrics, respectively. α , β , d and This is an adjustable weight. C ij The estimation can be based on historical access behavior statistics or satellite broadcast load information. The "historical access behavior statistics" mentioned in this application include, but are not limited to, indicators such as the frequency of each UE's selection of different satellite links, access success rate, number of collisions, and their proportion within a statistical period. The "satellite broadcast load information" may include fields such as the number of active connected users on the current link, the historical average congestion rate of each channel, the proportion of remaining available resources, and the remaining service time. The above information can be collected and reported autonomously by the UE, or periodically broadcast by satellite nodes in the control channel as input to the collision assessment function. R ij Used to delineate satellites j UE under current spatiotemporal conditions i Service sustainability and link feasibility. This indicator comprehensively characterizes the link's "whether it can be covered" and "coverage quality," specifically including: 1) Spatial coverage factors: the geometric relationship between the satellite and the UE (such as elevation angle and visibility window). If the elevation angle is too low or about to disappear from the field of view, then... R ij The factors include: 1) Reduction; 2) Channel reachability factors: whether the link has line-of-sight (LOS) conditions and the probability of obstruction; 3) Service continuity factors: the remaining service time predicted by the satellite trajectory and the access success rate of the current link; 4) Resource availability factors (optional): the proportion of remaining satellite broadcast resources and the number of active connections. These sub-indicators can be normalized and weighted to obtain a unified coverage reachability score. R ij ∈[0,1], the higher the score, the easier it is for the link to achieve stable coverage and continuous communication under the current spatiotemporal conditions. For example, it can be expressed in the following form:

[0075] ;

[0076] in, It is a satellite j For UE i The estimated remaining visibility time, t max This is the maximum serviceable time for reference. i ij It is a satellite j With UE i The current included angle, imax This is the maximum elevation angle, usually taken as 90°. It is the link LOS availability probability. , and The parameters are used to adjust the weights.

[0077] Step 3 UCB Decision-Making Strategy

[0078] Step 3, UCB decision strategy: Calculate the UCB score of each satellite link based on the historical average reward and confidence upper bound, and select the satellite link with the highest score for access;

[0079] For each UE, a UCB (Unified Credit Buffer) strategy dynamically determines which satellite link it should access. The UCB score is determined by the historical average performance metric (i.e., average reward function value) of each available satellite and a confidence upper bound adjustment term, the latter compensating for uncertainties caused by fewer observations of some links. Through this mechanism, the scheduling strategy achieves a dynamic balance between "prioritizing verified high-performance links" and "exploring potentially high-quality links." In each round of decision-making, the system selects the satellite link with the highest score for access, and after transmission is completed, updates the reward value of that link based on feedback link performance. This mechanism can balance "utilizing the best-performing link" and "exploring other links" in a dynamic environment, effectively avoiding getting trapped in local optima and improving the stability and resource utilization efficiency of the fusion system in complex environments. For UEs... i Its relationship with satellites j UCB between ij The calculation is as follows:

[0080] ;

[0081] Among them, UCB ij This indicates that in the UE i Access the j When there are 1 satellite, the link score is obtained based on the upper confidence bound algorithm. Indicates UE i Access the j The historical average reward value of a satellite t i Indicates UE i Total number of NTN access attempts n ij satellite j The number of times selected, As an adjustment term for the upper bound of confidence, it can be seen that if a certain satellite's n ij Smaller, meaning it is selected very infrequently, even if its average performance is low. Even if the confidence level is not high, it may still result in a large confidence adjustment term, thus achieving a higher UCB in the current round. ij The value is selected by the scheduling system, which ensures that the system remains willing to explore potentially high-quality but insufficiently tested links; as a satellite is repeatedly selected, n ij As the score increases, the confidence level gradually decreases, and the score gradually approaches the average of the true reward. This demonstrates the strategy of prioritizing the use of verified high-quality links.

[0082] Step 4: Multi-UE Channel Collision Control

[0083] Step 4, Multi-UE Channel Conflict Control: Reduce multi-user access conflicts through channel load broadcasting, backoff and polling reselection mechanisms, and access admission threshold control;

[0084] In a converged TN-NTN network architecture, multiple UEs may choose to connect to the same satellite and occupy the same frequency band or channel in the same time slot. Without coordination, this can easily lead to channel conflicts and interference, severely affecting system throughput and link stability. Therefore, this invention proposes the following conflict control strategy:

[0085] (1) Channel load broadcasting mechanism: Each satellite node periodically broadcasts its current channel load status information on the control channel, including but not limited to: the number of current access UEs on each available channel, the historical average congestion rate of each channel, the remaining proportion of available resources, etc. This broadcast information is decoded by all candidate UEs before access selection, and is used as one of the input features of the access strategy, thereby realizing the early avoidance of high-load channels and reducing the probability of collision.

[0086] (2) Backoff and polling reselection mechanism: If the UE detects a collision when attempting to access the target channel (e.g., no satellite acknowledgment response is received or the channel energy is too high), the backoff mechanism is triggered:

[0087] Index retreat: Delay 2 k ×Slot duration, where k This represents the number of consecutive conflicts.

[0088] Polling reselection: Reselect an uncongested channel from the available channel set and try it, and prioritize access to the channel with the best performance according to the dynamic priority strategy.

[0089] (3) Access Admission Threshold Control: The system sets a maximum access capacity threshold for each channel of each satellite. N max Once the current number of connected users, N current ≥ N maxNew access requests will be rejected or redirected to other available channels or satellite nodes. This mechanism prevents local hotspots from causing link congestion and improves overall system fairness and reachability.

[0090] Step 5 Online Learning and Scheduling Optimization

[0091] Step 5: Online learning and scheduling optimization: Real-time collection of link performance data, updating of reward function and confidence model to achieve dynamic optimization of scheduling strategy.

[0092] To achieve adaptive optimization of long-term scheduling strategies, this invention introduces an online learning mechanism based on the aforementioned UCB decision-making. After each round of resource allocation, the system collects real-time communication performance metrics of the UE and its accessed links, including actual throughput, average latency, Reference Signal Received Power (RSRP), Signal-to-Interference plus Noise Ratio (SINR), link packet loss rate, and access success rate, and updates the corresponding reward function estimate and link confidence model accordingly. The average reward value for all links is calculated. Number of scheduling selections n ij All adjustments are made dynamically during this process to reflect the latest changes in the network environment, user distribution, and link status. By continuously iterating the above evaluation-decision-feedback process, the system achieves online optimization of scheduling strategies for converged networks under dynamic conditions of multiple users, multiple links, and multiple interferences.

[0093] In addition, the scheduling system can periodically evaluate overall scheduling performance (such as average system throughput, task completion delay, channel collision rate, etc.) and adjust the network evaluation function accordingly. and The system adaptively reconfigures parameters such as the confidence term, reward function weight factor, and dynamic priority adjustment strategy in the UCB algorithm to further improve the system's adaptability to complex environments and resource utilization efficiency.

Claims

1. A method for resource allocation in a terrestrial-non-terrestrial fusion network based on the upper confidence bound algorithm, characterized in that, Includes the following steps: Step 1: Network Access Status Assessment: Based on the location information and link quality indicators of the user equipment (UE), calculate the terrestrial path score and satellite path score, and determine whether the UE accesses the terrestrial network (TN) or the non-terrestrial network (NTN). Step 2, Reward Function Construction: Construct link reward values ​​based on link throughput, latency, collision level, and coverage reachability; Step 3, UCB Decision Strategy: Calculate the UCB for each satellite link based on the historical average reward and confidence upper bound. ij The system scores satellite links and selects the one with the highest score for access, among which UCB... ij This represents the link score when UEi accesses the j-th satellite; Step 4, Multi-UE Channel Conflict Control: Reduce multi-user access conflicts through channel load broadcasting, backoff and polling reselection mechanisms, and access admission threshold control; Step 5, Online Learning and Scheduling Optimization: Real-time collection of link performance data, updating of reward function and confidence model, to achieve dynamic optimization of scheduling strategy; Step 1 also includes constructing an accessible network evaluation function. and Let represent the ground path score and satellite path score of the i-th UE, respectively. ； Among them, (x i , y i , z i ) represents the geographical location of the user equipment (UEi), Ω gNB This indicates the coverage area of ​​the ground base station gNB. Ⅱ[•] is an indicator function used to determine whether the location of the UE falls within the coverage area of ​​the ground base station gNB. Norm(•) indicates normalization processing. and These are the reference signal received power and the signal-to-interference-plus-noise ratio between UEi and gNB, respectively; w1, w2 and w3 are adjustable weights. ； Among them, v i It is the set of visible satellites for UEi; and η1 and η2 represent the signal-to-interference-plus-noise ratio (SINR) and time delay between UEi and satellite j, respectively; η1 and η2 are adjustable weights used to balance throughput and time delay. Constructing binary network access state variables λ i , ； That is, if the ground path score is higher, λ i =0, then the UE accesses the gNB; otherwise, λ i =1, access to LEO satellite; In Step 2, when λ i =1, calculate the link reward value r between UEi and satellite j. ij : ; Among them, T max D represents the theoretically achievable maximum link throughput under constraints such as system size, bandwidth, and modulation / coding scheme. max T represents the maximum tolerable communication latency threshold of the system under a specific business scenario. ij D ij C ij and R ij α, β, δ represent link throughput, communication latency, collision level, and coverage reachability metrics, respectively. This is an adjustable weight, C ij A unified coverage reachability score R can be estimated based on historical access behavior statistics or satellite broadcast load information. ij ∈[0,1], the higher the score, the easier it is for the link to achieve stable coverage and continuous communication under the current spatiotemporal conditions; it takes the following form: ; in, It is the estimated remaining visibility time of satellite j for UEi, t max This is the maximum serviceable time for reference, θ ij θ is the current angle between satellite j and UEi. max This is the maximum elevation angle, usually taken as 90°. It is the link LOS availability probability. , and The parameters are used to adjust the weights.

2. The method according to claim 1, characterized in that, In Step 3, UCB ij The calculation is as follows: ； Among them, UCB ij This represents the link score obtained based on the upper confidence bound algorithm when UEi accesses the j-th satellite. t represents the historical average reward value for UEi accessing the j-th satellite. i n represents the total number of NTN access attempts for UEi. ij The number of times satellite j is selected. This is the upper bound adjustment term for the confidence level.

3. The method according to claim 1, characterized in that, In Step 4, the multi-UE channel conflict control mechanism includes: Satellite periodic broadcast channel load information; The UE triggers an exponential backoff or polling reselection mechanism when a conflict occurs; Set a maximum access capacity threshold for each channel; if the limit is exceeded, access requests will be rejected or redirected.

4. The method according to claim 1, characterized in that, In Step 5, the online learning mechanism includes: Real-time collection of link performance data; Dynamically update the reward function value and confidence model; The system periodically evaluates its overall performance and adaptively adjusts the evaluation function and weight parameters.

5. A terrestrial-non-terrestrial fusion network resource allocation system based on the upper confidence bound algorithm, used to implement the method described in any one of claims 1 to 4, characterized in that, include: User equipment (UE) is used to report location information and link quality. Ground-based gNBs provide terrestrial network access services; Low Earth Orbit (LEO) satellites provide non-terrestrial network access services; The central dispatch and control center is used to execute the resource allocation method and coordinate the allocation of TN and NTN resources.

Images