A LEO satellite communication resource allocation method and system based on dynamic weighted graph partitioning

Abstract

This invention discloses a method and system for LEO satellite communication resource allocation based on dynamic weighted graph segmentation, relating to the field of satellite communication technology. Addressing the co-channel interference problem in large-scale MIMO satellite systems, this invention models the resource allocation optimization problem to maximize system sum rate as a partitioning optimization problem on a complete user graph. The weights of the edges in the graph are configured to represent the potential interference cost when two users reuse the same time-frequency resource. The optimization objective is to maximize the sum of edge weights between different partition groups. A dynamic weighted graph segmentation algorithm is employed, utilizing physical layer channel quality index feedback to introduce auxiliary weight variables. The edge weights of the graph are dynamically updated iteratively, and graph segmentation is performed until the algorithm converges. Finally, time-frequency resource allocation is completed based on the graph segmentation results. This invention can achieve near-optimal solution performance with low complexity and fast convergence speed, significantly improving the sum rate in highly dynamic satellite communication environments.

Description

Technical Field

[0001] This invention relates to user scheduling and time-frequency resource allocation in satellite communications, specifically to a method and system for LEO satellite communication resource allocation based on dynamic weighted graph segmentation, belonging to the field of wireless communication technology. Background Technology

[0002] Low Earth orbit (LEO) satellite constellations, with their low latency and wide-area coverage, are gradually becoming the cornerstone for future 6G networks to achieve global broadband connectivity. As a key enabling technology, massive MIMO (Multiple-Input Multiple-Output) technology, by utilizing the spatial degrees of freedom provided by large-scale antenna arrays, can significantly improve bandwidth and speed to meet ever-increasing traffic demands. However, the inherent wide beam coverage of satellite communications combined with the aggressive co-channel multiplexing strategy of massive MIMO inevitably leads to severe co-channel interference. To mitigate interference and maximize system bandwidth and speed, efficient resource allocation strategies are crucial. This is essentially a complex non-convex combinatorial optimization problem; achieving a balance between computational complexity and system performance, while adapting to the highly dynamic channel environment of satellite networks, is the core challenge currently facing this technology.

[0003] Existing technologies still have significant limitations in addressing the aforementioned problems. On the one hand, while the widely adopted greedy algorithms are computationally efficient, their short-sighted, one-by-one allocation strategy fails to capture the global interference structure generated by coupled allocation decisions, resulting in system performance often falling far short of the optimal solution and wasting spectrum resources. On the other hand, while integer linear programming (ILP) methods can theoretically obtain the global optimum, the computational complexity of ILP methods increases exponentially with the user scale due to the NP-hard nature of resource allocation problems, making them difficult to apply to satellite communication scenarios with large user scales or extremely high real-time requirements. Furthermore, some existing graph theory or deep learning methods often rely on static graph models or require huge model training overhead, making it difficult to flexibly adapt to the dynamic characteristics of LEO satellites. Summary of the Invention

[0004] Purpose of the invention: The purpose of this invention is to provide a method and system for allocating LEO satellite communication resources based on dynamic weighted graph segmentation, so as to solve the problems of severe co-channel interference, high computational complexity and difficulty in balancing performance of traditional resource allocation algorithms in existing large-scale MIMO satellite communication systems, thereby significantly improving the sum rate performance of the system while reducing computational overhead.

[0005] Technical solution: To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for allocating LEO satellite communication resources based on dynamic weighted graph segmentation includes the following steps:

[0007] Obtain the geographic coordinates of users to be scheduled within the satellite coverage area and map them to a relative coordinate system centered on the satellite to calculate the angle of arrival information;

[0008] A complete user graph model is constructed, where vertices correspond to users to be scheduled. The weight of the edge connecting any two vertices is configured to represent the potential interference cost when two users reuse the same time-frequency resource. The resource allocation problem of maximizing system and rate is transformed into a partitioning optimization problem of the complete user graph. The optimization objective is to maximize the sum of edge weights between different partition groups.

[0009] A dynamic weighted graph segmentation algorithm is used to iteratively solve the partitioning optimization problem. In each iteration, the algorithm calculates the channel quality index based on the previous graph segmentation result and the angle of arrival information. It then uses an introduced auxiliary weight variable and combines it with the channel gain to update the edge weights of the graph. The algorithm then applies the graph segmentation strategy to the graph structure after updating the edge weights until the convergence condition is met. The auxiliary weight variable is updated based on the channel quality index and is used to adjust the weight priority of the channel quality index in graph segmentation.

[0010] Based on the converged graph segmentation results, users within the same partition group are assigned to the same time-frequency resource block for spatial reuse, while users from different partition groups are assigned to mutually orthogonal time-frequency resource blocks.

[0011] Furthermore, the channel quality index is determined based on a constructed Gaussian approximation spatial correlation interference model. This Gaussian approximation spatial correlation interference model uses a Gaussian function to fit the spatial correlation coefficient of a large-scale uniform planar array. The spatial correlation coefficient is defined as an exponential function, and the exponent term of the exponential function is determined by the sum of the squares of the products of the antenna array size and the normalized angle difference between the user and the antenna array in the horizontal and vertical dimensions. The normalized angle difference in the horizontal / vertical dimensions is determined by the product of the normalized element spacing of the antenna array in the horizontal / vertical dimensions and the spatial angle difference between the two users in the horizontal / vertical dimensions. The spatial angle difference is calculated from the projection component of the angle of arrival information.

[0012] Furthermore, the partitioning optimization problem is achieved by constructing a maximum L-cut problem model, where L is the total number of resources. In the process of constructing the maximum L-cut problem model, the system and rate objective function are decoupled into a quadratic programming form using the Lagrange dual transformation. Based on the objective of minimizing interference between users in the same group, the decoupled problem is transformed into a maximum L-cut problem model, which maximizes the sum of the connection edge weights of users in different resource groups.

[0013] Furthermore, the dynamic weighted graph segmentation algorithm executes a closed-loop feedback control process involving interaction between the physical layer and the graph theory layer, including state awareness, weight correction, and topology reconstruction. The state awareness involves calculating the channel quality index of the physical layer based on the graph segmentation results of the previous round and the angle of arrival information. The weight correction involves adjusting the weights of edges in the graph model according to the channel quality index to requantify the interference relationships between users. The topology reconstruction involves re-partitioning the graph based on the adjusted weights, driving the resource allocation scheme to evolve towards system and rate improvement.

[0014] Furthermore, in the process of utilizing the introduced auxiliary weight variables and combining them with the edge weights of the channel gain update graph, the rule for edge weights is as follows: for any pair of user nodes, the edge weight is composed of the product of the first part and the second part; the first part is a dynamic symmetric interference term, which is determined according to the introduced auxiliary weight variables and the channel gain, and is used to quantify the mutual interference cost between users due to channel asymmetry in the iteration; the second part is a spatial correlation term, which is determined according to the angle of arrival information, and is used to characterize the physical spatial isolation between users.

[0015] Furthermore, the dynamic symmetric interference term is determined by the sum of the first sub-term and the second sub-term; the first sub-term is the product of the auxiliary weight variable of the first user and the channel gain ratio, and the second sub-term is the product of the auxiliary weight variable of the second user and the reciprocal of the channel gain ratio; wherein, the channel gain ratio refers to the ratio of the channel gain of the second user to the channel gain of the first user.

[0016] Furthermore, the update rule for the auxiliary weight variable is as follows: the auxiliary weight variable is initialized at the initial stage of the iteration; in each subsequent iteration, the signal-to-interference-plus-noise ratio (SIR) of each user under the current resource allocation is calculated, and the auxiliary weight variable is updated to the sum of a preset baseline constant and the current SIR; the update rule aims to adjust the weight priority of the user in graph segmentation according to the channel quality index represented by the current SIR, so that users with poor communication quality can obtain better resource isolation protection in the next iteration.

[0017] Furthermore, the dynamic weighted graph segmentation algorithm executes a multi-level graph segmentation strategy, including a coarsening stage, an initial partitioning stage, and a refinement stage. The coarsening stage iteratively merges node pairs in the graph based on the current edge weights to construct a series of coarsened graphs with progressively decreasing scales. The initial partitioning stage generates an initial resource partitioning scheme on the smallest coarsened graph. The refinement stage projects the partitioning scheme back to the original graph layer by layer, and swaps nodes at the partitioning boundaries through local search during each layer of projection to increase the sum of the cut edge weights at the current level.

[0018] The present invention also provides a computer system, including a memory, a processor, and a computer program / instructions stored in the memory and executable on the processor, wherein the computer program / instructions, when executed by the processor, implement the steps of the LEO satellite communication resource allocation method based on dynamic weighted graph segmentation.

[0019] The present invention also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the LEO satellite communication resource allocation method based on dynamic weighted graph segmentation.

[0020] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0021] 1. The satellite communication resource allocation method based on dynamic weighted graph segmentation proposed in this invention constructs a user complete graph model. The edge weights are configured to represent the potential interference cost when two users reuse the same time-frequency resource. The resource allocation problem of maximizing system and rate is transformed into a partitioning optimization problem of the user complete graph. The optimization objective is to maximize the sum of edge weights between different partition groups to achieve maximum interference isolation between users in different resource groups. The dynamic weighted graph segmentation algorithm is used to iteratively solve the partitioning optimization problem. A dynamic graph cut strategy driven by physical layer feedback is adopted to transform the complex non-convex resource allocation problem into a low-complexity solution of graph theory model. This eliminates the high computational resource requirements of traditional integer programming methods and the dependence of deep learning methods on large-scale training data. At the same time, it overcomes the limitation of traditional greedy algorithms being prone to getting trapped in local optima, and significantly reduces computational complexity and system overhead.

[0022] 2. In the satellite communication resource allocation method based on dynamic weighted graph segmentation proposed in this invention, a Gaussian approximation spatial correlation interference model is further constructed based on physical layer angle information. The Gaussian function is used to perform a second-order approximation on the original sine ratio form of the spatial correlation coefficient, thereby obtaining a concise exponential function expression. This approximation eliminates the complex trigonometric function calculations in the original formula and directly establishes an explicit relationship in which the interference cost monotonically decreases exponentially with the increase of the user angle interval. This enables the algorithm to quickly calculate the edge weights in the graph model based solely on the user's geographic coordinates, further significantly reducing the computational load of the system.

[0023] 3. Simulation results show that in small-scale scenarios, the system and rate of the resource allocation scheme obtained by the present invention can reach more than 98% of the optimal result of exhaustive search; in large-scale scenarios, compared with the traditional greedy algorithm, the present invention can achieve a rate performance gain of more than 20%, and the algorithm has a very fast convergence speed, usually only requiring 2 to 4 iterations to stabilize, and can effectively adapt to the highly dynamic satellite communication environment. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating the resource allocation method based on dynamic weighted graph segmentation provided in an embodiment of the present invention.

[0025] Figure 2 This is a schematic diagram of the dynamic weighted graph segmentation algorithm in an embodiment of the present invention;

[0026] Figure 3 This is a performance comparison chart of the system and rate as a function of signal-to-noise ratio in an embodiment of the present invention;

[0027] Figure 4 This is a schematic diagram illustrating the convergence performance of the dynamic weighted graph segmentation algorithm in an embodiment of the present invention. Detailed Implementation

[0028] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] This invention discloses a satellite communication resource allocation method based on dynamic weighted graph segmentation, which models the interference management and resource allocation problem in large-scale MIMO satellite communication as a maximum L-cut problem on a complete graph. The input consists of the geographical location coordinates and channel statistics of the users to be scheduled, and the output is the matching scheme between users and time-frequency resource blocks. This invention employs a physical layer feedback-driven strategy to construct a dynamic weighted graph segmentation (DWGP) algorithm, achieving rapid mapping from user spatial distribution to the optimal resource allocation scheme through iterative optimization.

[0030] like Figure 1 As shown in this embodiment, a satellite communication resource allocation method based on dynamic weighted graph segmentation mainly includes the following steps:

[0031] S1. Obtain the geographic coordinates of the users to be scheduled within the satellite coverage area and map them to a relative coordinate system centered on the satellite to calculate the angle of arrival information;

[0032] S2. Construct a complete user graph model, where vertices correspond to users to be scheduled. The weight of the edge connecting any two vertices is configured to represent the potential interference cost when two users reuse the same time-frequency resource. The resource allocation problem of maximizing system and rate is transformed into a partitioning optimization problem of the complete user graph. The optimization objective is to maximize the sum of edge weights between different partitioning groups.

[0033] S3. The partitioning optimization problem is solved iteratively using a dynamic weighted graph partitioning algorithm. In each iteration, the dynamic weighted graph partitioning algorithm calculates the channel quality index based on the previous graph partitioning result and the angle of arrival information. It then uses the introduced auxiliary weight variables and combines them with the channel gain to update the edge weights of the graph. The graph partitioning strategy is then applied to the graph structure after the edge weights are updated until the convergence condition is met. The auxiliary weight variables are updated based on the channel quality index and are used to adjust the weight priority of the channel quality index in graph partitioning.

[0034] S4. Based on the converged graph segmentation results, users within the same partition group are assigned to the same time-frequency resource block for spatial reuse, while users from different partition groups are assigned to mutually orthogonal time-frequency resource blocks.

[0035] In this embodiment, spatial correlation interference modeling can be performed based on the user's angle of arrival information relative to the satellite antenna array, quantifying potential interference between users as edge weights in a graph model. The DWGP algorithm does not employ a static graph partitioning strategy, but instead establishes a feedback loop between the physical layer and the graph theory layer: in each iteration, the physical layer channel quality index is calculated based on the resource allocation results of the previous round, and auxiliary variables are introduced using fractional programming theory to dynamically update the graph's edge weights, ensuring that the edge weights reflect the current interference level in real time. Subsequently, a multi-level graph partitioning strategy is used to solve the maximum L-cut problem of the weighted graph, gradually guiding the system to converge to the globally optimal solution with the maximum sum rate. This invention can effectively solve the non-convex resource allocation problem with low polynomial complexity.

[0036] In one possible implementation, the channel quality index is determined based on a constructed Gaussian approximation spatial correlation interference model to further reduce computational complexity. This Gaussian approximation model uses a Gaussian function to fit the spatial correlation coefficient of a large-scale uniform planar array. The spatial correlation coefficient is defined as an exponential function, and the exponent term of this exponential function is determined by the sum of the squares of the products of the antenna array size and the normalized angle difference between the user and the antenna array in the horizontal and vertical dimensions. The normalized angle difference in the horizontal / vertical dimensions is determined by the product of the normalized element spacing of the antenna array in the horizontal / vertical dimensions and the spatial angle difference between the two users in the horizontal / vertical dimensions, and this spatial angle difference is calculated from the projection component of the angle of arrival information.

[0037] The detailed implementation process of the method of the present invention will be illustrated below with reference to a specific system model.

[0038] In step S1, this embodiment considers the uplink transmission of a LEO satellite communication system, in which a satellite has an altitude of [missing information]. The satellite simultaneously serves a large number of single-antenna user terminals (UTs). The satellite has a network consisting of... A large-scale uniform planar array (UPA) consisting of three antennas, in which and These represent the number of antennas along the x-axis and y-axis, respectively. In LEO satellite systems, due to the high satellite altitude and sparse scattering, the channel is primarily dominated by the line-of-sight (LoS) path, making multipath effects negligible. It is assumed that significant frequency offset and propagation delay have been effectively pre-compensated at the receiver physical layer. This allows us to... The channel is simplified to a static vector. To focus on the resource allocation problem, this vector is modeled as complex channel gain. and array response vector The product of:

[0039] (1)

[0040] Among them, the array response vector of each user Angle of arrival (AoAs) The decision was made. These angles were derived from the known geographic coordinates of the satellite and the UT. The relative position vectors were obtained through a standard transformation to the satellite's central coordinate system. Based on this, AoAs can be calculated as follows:

[0041] (2)

[0042] in It is the arctangent function in the four quadrants. Based on the angle of arrival... The projection components define the spatial angle pairs. :

[0043] (3)

[0044] UPA's array response vector It is the Kronecker product of the guide vectors in each axis:

[0045] (4)

[0046] in Represents the Kronecker product operation; the guiding vector is... Given, applicable to having A uniform linear array of elements. Here, It is the carrier wavelength. This refers to the antenna spacing along the corresponding axis. (User) The average channel power gain is expressed as In the static LoS-dominated model, small-scale fading is averaged out, making... It becomes a deterministic value dependent on antenna gain and free-space path loss (FSPL). Expressed in decibels, its expression is: in and These are the antenna gains of the user terminal and the satellite, respectively. User Free space path loss.

[0047] In the large-scale MIMO LEO satellite system under consideration, orthogonal frequency division multiplexing (OFDM) is employed, derived from the ensemble. of Individual users One orthogonal time-frequency resource (indexed as) Data transmission. The scheduling strategy is based on a binary allocation matrix. This indicates that if the user Allocated to resources ,but Otherwise, it is 0. Therefore, satellites have resources The aggregated signal received above It is the sum of all scheduled user transmissions, expressed as:

[0048] (5)

[0049] in, It comes from users Normalized data symbols that satisfy , That is the transmit power of each UT, and It is the additive white Gaussian noise (AWGN) vector. When a linear receiver is used at the satellite end, UT The recovery signal is

[0050] (6)

[0051] in It's the receiver filter. We use a maximum ratio combining (MRC) receiver, assuming... Because it offers a good performance-complexity tradeoff. Specifically, it avoids the computationally intensive matrix inversion required by zero-forcing (ZF) and minimum mean square error (MMSE) receivers, whose complexity increases cubically with the number of users in the group due to matrix inversion. Based on this, and substituting the channel formula (1) and the average channel power gain, UT In resources The final SINR is:

[0052] (7)

[0053] in It is the transmit signal-to-noise ratio (SNR). This expression reveals the user's... The performance of this system critically depends on the spatial separation between users sharing resources, which is determined by spatially related terms. By substituting the formula for the guiding vector (4) and expanding the Kronecker product, we can obtain:

[0054] (8)

[0055] in The normalized angular difference in the horizontal dimension is determined by the normalized element spacing. The product of the spatial angular difference between the two users in the horizontal dimension is used to determine this. The normalized angular difference in the vertical dimension is determined by the normalized element spacing. The product of the spatial angle difference between the two users in the vertical dimension is used to determine the spatial correlation interference model. Formula (8) is the specific implementation of the Gaussian approximation spatial correlation interference model constructed in this embodiment. The exponential function part corresponds to the spatial correlation function of a large-scale uniform planar array that is approximated by a Gaussian function. The spatial correlation coefficient is defined as an exponential function, and the exponential term is determined by the size of the antenna array in the horizontal and vertical dimensions of the antenna array ( , ) and normalized angle difference ( , The sum of the squares of the products of ) is determined. It should be noted that in some possible implementations, the equation in formula (8) can also be used to calculate, that is, to directly calculate the product of the squares of the array factor modulus in the two dimensions; wherein, the array factor is expressed in the form of a sine ratio function, that is, it is composed of a sine function containing the antenna array size parameter divided by a sine function containing the angle difference parameter.

[0056] The above equation reveals the spatial correlation, a key driver of interference, determined by the angular separation between users and the geometry of the antenna array. An easily tractable approximation. This relationship is further clarified, showing that the correlation decays exponentially with increasing user-perspective separation. This insight reformulates the complex task of maximizing sum-rates into a more intuitive combinatorial problem: dividing users into groups with minimal intra-group spatial correlation. The formula... Easy-to-process approximation in Substituting these values, we obtain the final SINR expression used for optimization:

[0057] (9)

[0058] This expression indicates that performance is determined by the spatial interference factor between users in the group. Therefore, the global objective of maximizing system performance and rate is transformed into the problem of minimizing user partition combinations that minimize intra-group interference.

[0059] Based on the derived SINR expression, we now formally construct the optimization problem. (User) In resources The achievable rate on is: Therefore, the resource allocation problem can be expressed as:

[0060] (10)

[0061] The first constraint in formula (10) ensures that each user is assigned one and only one resource, and the second constraint guarantees that each resource is used. Problem P1 is a classic combinatorial optimization problem. Finding the optimal solution requires an exhaustive search of all... There are several possible allocation schemes, which are computationally infeasible for practical systems. Therefore, it is necessary to develop a low-complexity and efficient algorithm.

[0062] The computational complexity of problem P1 is driven by interference coupled between users, making direct solution infeasible. In step S2, this embodiment addresses this challenge by restating the problem within a graph-based framework. This method encapsulates the complexity of the physical layer by modeling users as vertices and their mutual interference as weighted edges, thereby transforming the intractable sum-rate maximization problem into a series of manageable graph partitioning tasks. Specifically, vertices in the graph correspond to users to be scheduled, and the weight of the edge connecting any two vertices is configured to represent the potential interference cost when two users reuse the same time-frequency resource. The resource allocation problem of maximizing system sum-rate is transformed into a partitioning optimization problem of the complete user graph. The optimization objective is to maximize the sum of edge weights between different partition groups to achieve maximum interference isolation between users in different resource groups.

[0063] In practice, the partitioning optimization problem is achieved by constructing a maximum L-cut problem model. In the process of constructing the maximum L-cut problem model, firstly, the Lagrange dual transformation in fractional programming theory is used to decouple the system and the rate objective function into a quadratic programming form that is easy to solve. Then, based on the objective of minimizing the interference between users in the same group, the decoupled problem is transformed into a maximum L-cut problem model. In this model, minimizing the sum of interference weights between users in the same resource group is mathematically equivalent to maximizing the sum of the connection edge weights between users in different resource groups.

[0064] The dynamic weighted graph segmentation algorithm executes a closed-loop feedback control process between the physical layer and the graph theory layer, including state awareness, weight correction, and topology reconstruction. State awareness involves calculating the channel quality index of the physical layer based on the graph segmentation results of the previous round and combining it with the angle of arrival information. Weight correction involves adjusting the weights of edges in the graph model according to the channel quality index to requantify the interference relationship between users. Topology reconstruction involves re-partitioning the graph based on the adjusted weights, driving the resource allocation scheme to evolve in the direction of system and rate improvement.

[0065] In the weight correction step, the edge weights are determined as follows: for any pair of user nodes, the edge weights consist of the product of the first and second parts. The first part is a dynamic symmetric interference term, determined based on the introduced auxiliary weight variable and channel gain, used to quantify the mutual interference cost between users due to channel asymmetry during iteration. The second part is a spatial correlation term, determined based on the angle of arrival information, used to characterize the physical spatial isolation between users. The dynamic symmetric interference term is determined by the sum of the first and second sub-terms. The first sub-term is the product of the auxiliary weight variable of the first user and the channel gain ratio, and the second sub-term is the product of the auxiliary weight variable of the second user and the reciprocal of the channel gain ratio. The channel gain ratio refers to the ratio of the channel gain of the second user to the channel gain of the first user. The update rule for the auxiliary weight variable is as follows: the auxiliary weight variable is initialized at the initial stage of the iteration; in each subsequent iteration, the signal-to-interference-plus-noise ratio (SIR) of each user under the current resource allocation is calculated, and the auxiliary weight variable is updated to the sum of the preset baseline constant and the current SIR.

[0066] In this embodiment, the framework utilizes two key techniques: fractional programming (FP) and block coordinate descent (BCD) to systematically transform the original problem P1 into tractable subproblems. First, to handle the non-convex logarithm and objective function, we apply the Lagrange dual transformation method. This is achieved by introducing auxiliary variables. , of which elements Indicates user In resource block The auxiliary variable of the signal-to-interference-plus-noise ratio is used to transform the original problem P1 into:

[0067] (11)

[0068] because and Due to the coupling between variables, this problem remains non-convex. Therefore, it is solved by employing the BCD method, alternately optimizing each variable block. When assigning matrices... When fixed, The optimal update has a closed-form solution: given a fixed allocation matrix Optimal Depend on Give. When When fixed, let Let P2 represent the objective function. about Since it is concave, the optimal solution can be obtained by setting its partial derivative to zero, that is, by letting... ,get .

[0069] In order to Optimization under fixed conditions We focus on the third term in the P2 objective function. By applying a high signal-to-noise ratio (SNR) approximation, This maximization problem is transformed into an equivalent weighted inverse SINR minimization problem, thus generating an easily handled subproblem: Substituting the SINR expression in equation (9), the subproblem can be transformed into the following quadratic distribution form:

[0070] (12)

[0071] in This is a constant weight in the subproblem, namely the auxiliary weight variable mentioned in step S3. The value "1" here refers to the preset benchmark constant. The subscript indicates the user in the current iteration step. The index of the allocated time-frequency resource block. This final form is the easily solvable subproblem solved in each iteration.

[0072] In problem P3, the second term of the objective function is relative to the assignment matrix. It is a constant. Therefore, optimization is equivalent to minimizing the first term, which represents the total normalized disturbance. The additive structure of this objective function is isomorphic to partitioning a complete graph. The problem involves vertices representing users. The weight of each edge... In each iteration Dynamic update, representing the cost of symmetric disturbance:

[0073] (13)

[0074] in Formula (13) corresponds to the weight update rule, where The term within parentheses represents the spatial correlation factor, and the term in parentheses is the dynamic symmetric interference term, used to quantify the cost of channel asymmetry. The entire iterative weight update process embodies the closed-loop feedback control process of the interaction between the physical layer and the graph theory layer.

[0075] Minimizing the objective in P3 is mathematically equivalent to minimizing the sum of edge weights within the partition. This, in turn, is equivalent to maximizing the sum of edge weights within the partition—the classic maximum L-cut problem. This leads to the following problem statement (i.e., the maximum L-cut problem model):

[0076] (14)

[0077] It should be noted that the maximization objective of formula (14) is mathematically equivalent to minimizing the interference weights among users in the same group, thus realizing the process of transforming resource allocation into a partitioning optimization problem.

[0078] Based on this formula, the Dynamic Weighted Graph Partitioning (DWGP) algorithm implemented in step S3 is an iterative framework, and the specific algorithm flow is as follows: Figure 2 As shown, the algorithm operates through a feedback mechanism: starting from an initial partition, each iteration first updates the edge weights of the graph based on the physical layer performance (SINR) of the current partition, and then solves the maximum L-cut problem P4 to obtain an improved partition. To address the NP-hard nature of the maximum L-cut subproblem, we employ an efficient multi-level partitioning heuristic method. Specifically, the dynamic weighted graph partitioning algorithm in this embodiment uses a multi-level graph partitioning strategy to solve the aforementioned maximum L-cut problem, which includes a coarsening stage, an initialization stage, and a refinement stage. In the coarsening phase, based on the current edge weights, node pairs in the graph are iteratively merged using a matching algorithm (such as a heavy edge matching algorithm or a random matching algorithm) to construct a series of coarsened graphs with progressively decreasing scales. In the initial partitioning phase, an initial resource partitioning scheme is generated on the smallest coarsened graph. In the refinement phase, the partitioning scheme is projected back to the original graph layer by layer, and during each layer of projection, a local search algorithm (such as the Kernighan-Lin algorithm or the Fiduccia-Mattheyses algorithm) is used to swap nodes at the partition boundaries to increase the sum of the cut edge weights at the current level. This iterative process of reweighting and repartitioning gradually guides user allocation towards a higher-quality solution.

[0079] In step S4, after obtaining the final resource allocation scheme, the system enters the execution phase. Based on the converged graph partitioning result, the satellite communication system schedules users belonging to the same partition group to the same time-frequency resource block for spatial multiplexing, and assigns different partition groups to mutually orthogonal time-frequency resource blocks. Subsequently, the system executes user data transmission based on the resource allocation scheme; that is, the user terminal transmits signals on the allocated time-frequency resource block, and the satellite uses a large-scale array antenna to receive and demodulate the user signals, thereby completing the physical layer communication transmission.

[0080] To verify the effectiveness of the resource allocation method based on dynamic weighted graph segmentation proposed in this invention, this embodiment underwent simulation testing in a typical low-Earth orbit satellite communication scenario: the satellite orbital altitude was set to 550km, and the configuration... A large-scale uniform planar array antenna with a carrier frequency of 2 GHz was used. To comprehensively evaluate the algorithm performance, this embodiment selected the Greedy algorithm, the Random algorithm, and the Exhaustive Search algorithm (implementable in small-scale scenarios) as benchmarks for comparison, and compared different user scales (small-scale). With large scale A comparative analysis was conducted under different signal-to-noise ratio conditions.

[0081] like Figure 3 As shown, the comparison results of the method of the present invention with various benchmark algorithms in terms of system and speed performance are illustrated. Experimental results show that, as Figure 3 As shown in (a), in a small-scale user scenario ( The performance curve of the method of this invention highly coincides with the optimal performance curve obtained by exhaustive search, achieving more than 98% of the theoretical optimal solution, thus proving the effectiveness of the algorithm in finding the global optimal solution; Figure 3 As shown in (b), in large-scale user scenarios ( Since the solution space explodes exponentially, exhaustive search becomes impractical. In this case, the method of this invention exhibits a significant performance advantage over the traditional greedy algorithm. Especially in high signal-to-noise ratio regions, because this invention effectively isolates strongly interfering users through graph segmentation, it achieves a sum rate performance gain of over 20% compared to the greedy algorithm, significantly improving the system's sum rate.

[0082] like Figure 4 The figure illustrates the convergence performance of the dynamic weighted graph segmentation algorithm in this embodiment of the invention. As can be seen from the figure, the summation rate of the system increases rapidly with the number of iterations and then quickly stabilizes. Specifically, regardless of whether the signal-to-noise ratio (SNR) is low or high, the algorithm of this invention typically requires only 2 to 4 iterations to reach convergence. This demonstrates the efficiency of the mechanism proposed in this invention for dynamically updating graph edge weights based on physical layer SINR feedback.

[0083] This invention also discloses a computer system, including a memory, a processor, and a computer program / instructions stored in the memory and executable on the processor. When the computer program / instructions are executed by the processor, they implement the steps of the LEO satellite communication resource allocation method based on dynamic weighted graph segmentation.

[0084] This invention also discloses a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the LEO satellite communication resource allocation method based on dynamic weighted graph segmentation.

[0085] The program code used to implement the method of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the steps of the method of the present invention to be performed. The program code can be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a standalone software package, or entirely on a remote machine or server. All aspects not detailed in this invention are well-known to those skilled in the art.

Claims

1. A method for allocating LEO satellite communication resources based on dynamic weighted graph segmentation, characterized in that, Includes the following steps: Obtain the geographic coordinates of users to be scheduled within the satellite coverage area and map them to a relative coordinate system centered on the satellite to calculate the angle of arrival information; A complete user graph model is constructed, where vertices correspond to users to be scheduled. The weight of the edge connecting any two vertices is configured to represent the potential interference cost when two users reuse the same time-frequency resource. The resource allocation problem of maximizing system and rate is transformed into a partitioning optimization problem of the complete user graph. The optimization objective is to maximize the sum of edge weights between different partition groups. A dynamic weighted graph segmentation algorithm is used to iteratively solve the partitioning optimization problem. In each iteration, the algorithm calculates the channel quality index based on the previous graph segmentation result and the angle of arrival information. It then uses an introduced auxiliary weight variable and combines it with the channel gain to update the edge weights of the graph. The algorithm then applies the graph segmentation strategy to the graph structure after updating the edge weights until the convergence condition is met. The auxiliary weight variable is updated based on the channel quality index and is used to adjust the weight priority of the channel quality index in graph segmentation. Based on the converged graph segmentation results, users within the same partition group are assigned to the same time-frequency resource block for spatial reuse, while users from different partition groups are assigned to mutually orthogonal time-frequency resource blocks.

2. The method for allocating LEO satellite communication resources based on dynamic weighted graph segmentation according to claim 1, characterized in that, The channel quality index is determined based on a constructed Gaussian approximation spatial correlation interference model. This Gaussian approximation spatial correlation interference model uses a Gaussian function to fit the spatial correlation coefficient of a large-scale uniform planar array. The spatial correlation coefficient is defined as an exponential function, and the exponent term of the exponential function is determined by the sum of the squares of the products of the antenna array size and the normalized angle difference between the user and the antenna array in the horizontal and vertical dimensions. The normalized angle difference in the horizontal / vertical dimensions is determined by the product of the normalized element spacing of the antenna array in the horizontal / vertical dimensions and the spatial angle difference between the two users in the horizontal / vertical dimensions. The spatial angle difference is calculated from the projection component of the angle of arrival information.

3. The method for allocating LEO satellite communication resources based on dynamic weighted graph segmentation according to claim 1, characterized in that, The partitioning optimization problem is achieved by constructing a maximum L-cut problem model, where L is the total number of resources. In the process of constructing the maximum L-cut problem model, the system and rate objective function are decoupled into a quadratic programming form using the Lagrange dual transformation. Based on the objective of minimizing interference between users within the same resource group, the decoupled problem is transformed into a maximum L-cut problem model, which maximizes the sum of the connection edge weights between users in different resource groups.

4. The method for allocating LEO satellite communication resources based on dynamic weighted graph segmentation according to claim 1, characterized in that, The dynamic weighted graph segmentation algorithm executes a closed-loop feedback control process involving interaction between the physical layer and the graph theory layer, including state awareness, weight correction, and topology reconstruction. State awareness involves calculating the physical layer's channel quality index based on the previous round's graph segmentation results and the angle-of-arrival information. Weight correction adjusts the weights of edges in the graph model according to the channel quality index to requantify interference relationships between users. Topology reconstruction re-partitions the graph based on the adjusted weights, driving the resource allocation scheme to evolve towards improved system performance and speed.

5. The method for allocating LEO satellite communication resources based on dynamic weighted graph segmentation according to claim 1, characterized in that, The rule for edge weights in the method of utilizing the introduced auxiliary weight variables and combining them with the edge weights of the channel gain update graph is as follows: for any pair of user nodes, the edge weight is composed of the product of the first part and the second part; the first part is a dynamic symmetric interference term, which is determined according to the introduced auxiliary weight variables and the channel gain, and is used to quantify the mutual interference cost between users due to channel asymmetry in the iteration; the second part is a spatial correlation term, which is determined according to the angle of arrival information, and is used to characterize the physical spatial isolation between users.

6. The method for allocating LEO satellite communication resources based on dynamic weighted graph segmentation according to claim 5, characterized in that, The dynamic symmetric interference term is determined by the sum of the first sub-term and the second sub-term; the first sub-term is the product of the auxiliary weight variable of the first user and the channel gain ratio, and the second sub-term is the product of the auxiliary weight variable of the second user and the reciprocal of the channel gain ratio; wherein, the channel gain ratio refers to the ratio of the channel gain of the second user to the channel gain of the first user.

7. The method for allocating LEO satellite communication resources based on dynamic weighted graph segmentation according to claim 5, characterized in that, The update rule for the auxiliary weight variable is as follows: the auxiliary weight variable is initialized at the initial stage of the iteration; in each subsequent iteration, the signal-to-interference-plus-noise ratio (SIR) of each user under the current resource allocation is calculated, and the auxiliary weight variable is updated to the sum of the preset baseline constant and the current SIR.

8. The method for allocating LEO satellite communication resources based on dynamic weighted graph segmentation according to claim 1, characterized in that, The dynamic weighted graph segmentation algorithm executes a multi-level graph segmentation strategy, including a coarsening stage, an initial partitioning stage, and a refinement stage. The coarsening stage iteratively merges node pairs in the graph based on the current edge weights, constructing a series of coarsened graphs with progressively decreasing scales. The initial partitioning stage generates an initial resource partitioning scheme on the smallest coarsened graph. The refinement stage projects the partitioning scheme back to the original graph layer by layer, and during each layer's projection, swaps nodes at the partition boundaries through local search to increase the sum of the cut edge weights at the current level.

9. A computer system comprising a memory, a processor, and computer programs / instructions stored in the memory and executable on the processor, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the LEO satellite communication resource allocation method based on dynamic weighted graph segmentation according to any one of claims 1-8.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the LEO satellite communication resource allocation method based on dynamic weighted graph segmentation according to any one of claims 1-8.

Images