Multi-cell massive MIMO user grouping method and system based on graph clustering

Abstract

The present invention discloses a multi-cell large-scale MIMO user grouping method and system based on graph clustering, which realizes the pairing of users and base stations, as well as the allocation of time-frequency resource blocks. The present invention performs user grouping with the goal of maximizing the system sum rate, and the same group of users uses the same time-frequency resource blocks to communicate with the same base station. First, some users are selected as nodes, and the edges and weights are determined according to the statistical channel covariance matrix to construct a multi-cell user relationship graph; then, the above graph is converted into a density-connected tree using local density; then, the tree is cut according to the average edge weight to obtain the initial user grouping result; finally, the group to which the remaining users belong is determined according to the user grouping criterion. The present invention can achieve high system sum rate with low complexity, and effectively solve the user scheduling and time-frequency resource allocation problems of the multi-cell large-scale MIMO system.

Description

Technical Field

[0001] The present invention relates to a multi-cell large-scale MIMO system resource allocation technology based on graph clustering, in particular to user scheduling and time-frequency resource allocation, and belongs to the field of wireless communication technology. Background Art

[0002] Massive Multiple-Input Multiple-Output (MIMO) technology is one of the key technologies for meeting the demand for ultra-high transmission rates. By equipping base stations with a large number of antenna arrays, massive MIMO systems can serve more users using the same resource blocks (RBs). In multi-cell environments, users sharing the same RBs can generate co-channel interference (CCI), significantly limiting the performance of massive MIMO systems. An effective solution is to group users across multiple cells, simultaneously pairing users with base stations and allocating RBs to reduce CCI and improve system performance. Common user grouping methods include greedy or heuristic search, reinforcement learning, and unsupervised learning. Greedy or heuristic search methods typically find local optima but may not reach the global optimum. Reinforcement learning methods, in dense user scenarios, incur high computational complexity due to the significant increase in the action space. Unsupervised learning methods, which exploit implicit data features to group users, offer the advantage of low complexity and are particularly well-suited for multi-cell user grouping. Graph clustering, as an unsupervised learning method, demonstrates significant computational advantages when processing large-scale data. By modeling data as a graph and utilizing the graph's topological features for clustering, graph clustering can effectively identify group structures in the data, reducing computational complexity while improving the quality and efficiency of user grouping.

[0003] Furthermore, most user grouping methods rely on instantaneous channel state information, which changes rapidly over time, potentially leading to frequent updates to user grouping results or suboptimal grouping results within a certain period. Statistical channel state information changes more slowly and can be accurately acquired, and research has demonstrated its significant utility in user scheduling. In response to the characteristics of multi-cell communications, this paper applies graph clustering technology based on statistical channel information to the multi-cell user grouping problem, achieving the dual goals of low computational complexity and high system summation rate. Summary of the Invention

[0004] Purpose of the invention: In the multi-cell transmission scenario of massive MIMO, to address the problems of strong multi-cell interference and high complexity of resource allocation in dense user scenarios, the present invention discloses a multi-cell massive MIMO user grouping method and system based on graph clustering, which simultaneously realizes the pairing of users and base stations and the allocation of time-frequency resources. Compared with existing similar technical means, this method can reduce computational complexity and improve system and rate.

[0005] Technical solution: In order to achieve the above-mentioned purpose, the present invention adopts the following technical solution: a multi-cell large-scale MIMO user grouping method based on graph clustering, wherein the method takes maximizing the system sum rate as the goal, groups users according to the statistical channel covariance matrix, and the same group of users uses the same time-frequency resource block to communicate with the same base station. The specific steps are as follows: Step 1, establish a graph, select some users as nodes based on the channel gain between users and each base station to establish a multi-cell user relationship graph, and determine the edges and weights between user nodes according to the statistical channel covariance matrix; Step 2, construct a tree, convert the above-mentioned multi-cell user relationship graph into a density-connected tree according to the principle of local density maximization; Step 3, cut the tree, cut the tree according to the average edge weight, preferentially remove the edge with the largest minimum average edge weight of the two connected node groups, obtain the same number of node groups as the number of time-frequency resource blocks to be allocated, and obtain the initial user grouping result in combination with the base station paired with the user; Step 4, supplement the grouping, determine the user group to which the remaining users belong according to the user grouping criterion of giving priority to no interference between users and secondly considering the minimum system sum rate loss.

[0006] Furthermore, users are grouped according to the statistical covariance matrix of the beam domain channel between the user and each base station. The beam domain channel is the channel obtained by angular sampling of the spatial channel by the sampling matrix, and its statistical covariance matrix is ​​a diagonal matrix, in which each diagonal element represents the channel energy in the corresponding direction; the direction corresponding to the diagonal element exceeding the threshold is the main channel direction between the user and the base station, and the direction corresponding to the maximum diagonal element in the main channel direction is the direction of the maximum channel gain between the user and the base station.

[0007] Furthermore, in step 1, some users are selected as nodes, and each node is paired with the base station with the largest channel gain. The selected node satisfies that the ratio of the channel gain between it and the paired base station to the sum of the channel gains between it and all base stations exceeds a threshold; the direction assigned to each node is the direction of the maximum channel gain between it and the paired base station. If two nodes do not interfere with each other in their respective assigned directions, there is an edge between them, and the weight of the edge is equal to the minimum value of the maximum channel gain of the two nodes multiplied by the ratio of the number of their common main channel directions to the total number of main channel directions.

[0008] Assume that users k1 and k2 are nodes in the graph, paired with base stations n1 and n2, respectively, and assigned directions l1 and l2, respectively. The two users do not interfere with each other in their respective assigned directions if the primary channel direction between user k1 and base station n2 does not include l2, and the primary channel direction between user k2 and base station n1 does not include l1.

[0009] Furthermore, in step 2, the constructed multi-cell user relationship graph is converted into a density-connected tree based on the principle of local density maximization, ensuring that all nodes are connected to each other without forming closed loops. The local density of two nodes is equal to the edge weight between them multiplied by the ratio of the number of common adjacent nodes to the total number of adjacent nodes. The specific steps for constructing the density-connected tree are as follows: step 2.1, randomly select a node as the starting point of the density-connected tree; step 2.2, select the adjacent node with the highest local density with the node in the tree and connect it to the density-connected tree; step 2.3, repeat step 2.2 until all nodes are added to the density-connected tree.

[0010] Furthermore, in step 3, the initial user grouping result is obtained by cutting the tree, and the specific steps are as follows: step 3.1, calculate the D value of each edge, that is, the minimum value of the average edge weight of the two node groups connected by the edge; step 3.2, remove the edge with the largest D value; step 3.3, repeat step 3.2 until L node groups are generated, where L represents the number of time-frequency resource blocks to be allocated; step 3.4, according to the base station corresponding to the node, the above grouping is further divided into N×L node groups, where N represents the number of cells.

[0011] Furthermore, in step 4, the direction in which the users to be grouped will be assigned when entering each user group is first determined, that is, the direction in which the users to be grouped are least interfered with when communicating with the corresponding base station on the time-frequency resource blocks corresponding to each user group, and then the group to which the users belong is determined based on the user grouping criteria.

[0012] Furthermore, the user grouping criteria are as follows: priority is given to ensuring that there is no interference between the user to be grouped and the users in the group to be joined; if the user to be grouped and all users in a user group do not interfere with each other in their respective assigned directions, the user joins the group; if there are multiple groups that do not interfere with each other, the group in which the main channel directions of all users in the group have the highest overlap with the main channel directions of the user to be grouped is preferentially selected; secondly, considering the system sum rate loss, the user group with the lowest clustering cost is selected for the user to be grouped, where the clustering cost of a user joining a user group is the difference between the original system sum rate and the system sum rate after the user joins the group.

[0013] The present invention also provides a multi-cell massive MIMO user grouping system based on graph clustering, including a central scheduler for grouping users based on a statistical channel covariance matrix with the goal of maximizing system sum and rate, so that users in the same group communicate with the same base station using the same time-frequency resource blocks. The central scheduler includes:

[0014] A graph building module is used to select some users as nodes based on the channel gains between users and each base station to establish a multi-cell user relationship graph, and determine the edges and weights between user nodes based on the statistical channel covariance matrix;

[0015] A tree construction module, configured to convert the multi-cell user relationship graph into a density connection tree according to a local density maximization principle;

[0016] A tree cutting module is used to cut the tree according to the average edge weight, preferentially removing the edge with the largest minimum average edge weight between two node groups, to obtain the same number of node groups as the number of time-frequency resource blocks to be allocated, and to obtain the initial user grouping result based on the base stations paired with the users;

[0017] The supplementary grouping module is used to determine the user groups to which the remaining users belong based on the user grouping criteria of first considering no interference between users and secondly considering minimum system and rate loss.

[0018] The present invention also provides a computer program product, comprising a computer program / instruction, which, when executed by a processor, implements the steps of the multi-cell massive MIMO user grouping method based on graph clustering.

[0019] Beneficial Effects: The proposed multi-cell massive MIMO user grouping method based on graph clustering simultaneously achieves user and base station pairing and time-frequency resource block allocation, effectively reducing co-channel interference and improving system summation and rate. Compared with the baseline algorithm, it significantly improves system summation and rate by more than 20%. Furthermore, this method effectively reduces computational complexity, running in half the time of hierarchical clustering methods in dense user scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a flow chart of a multi-cell user grouping method based on graph clustering according to an embodiment of the present invention.

[0021] Figure 2 Schematic diagram of a density connection tree in an embodiment of the present invention.

[0022] Figure 3 This is a comparison chart of the performance of the embodiment of the present invention and the baseline algorithm as the signal-to-noise ratio changes.

[0023] Figure 4 This is a comparison chart of the performance of the embodiment of the present invention and the baseline algorithm as the number of users changes. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and effect of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the present invention.

[0025] like Figure 1 As shown, an embodiment of the present invention discloses a multi-cell large-scale MIMO user grouping method based on graph clustering, with the goal of maximizing the system sum rate. Users are grouped according to the statistical channel covariance matrix, and the same group of users uses the same time-frequency resource block to communicate with the same base station. In this embodiment, users are grouped according to the statistical covariance matrix of the beam domain channel between the user and each base station, where the beam domain channel is the channel after the sampling matrix performs angular sampling on the spatial channel, and its statistical covariance matrix is ​​a diagonal matrix, and each diagonal element represents the channel energy in the corresponding direction; the direction corresponding to the diagonal element exceeding the threshold can be set as the main channel direction between the user and the base station, and the direction corresponding to the maximum diagonal element in the main channel direction is the direction of maximum channel gain between the user and the base station.

[0026] Specifically, in this embodiment, the multi-cell massive MIMO user grouping method based on graph clustering specifically includes:

[0027] Step 1: Based on the channel gains between users and base stations, some users are selected as nodes to establish a multi-cell user relationship graph, and the edges and weights between user nodes are determined according to the statistical channel covariance matrix.

[0028] In this embodiment, some users are selected as nodes, and each node is paired with the base station with the largest channel gain. The selected nodes meet the requirement that the ratio of the channel gain between the node and the paired base station to the sum of the channel gains between the user and all base stations exceeds a threshold. The direction assigned to each node is the direction of the maximum channel gain between the node and the paired base station. If two nodes do not interfere with each other in their assigned directions, an edge exists between them, and the weight of the edge is equal to the minimum value of the maximum channel gain of the two nodes multiplied by the ratio of the number of their shared primary channel directions to the total number of primary channel directions.

[0029] For example, assume that users k1 and k2 are nodes in the graph, paired with base stations n1 and n2 respectively, and assigned directions l1 and l2 respectively. The two users do not interfere with each other in their respective assigned directions, which means that the main channel direction between user k1 and base station n2 does not include l2, and the main channel direction between user k2 and base station n1 does not include l1.

[0030] Step 2: Convert the multi-cell user relationship graph into a density connection tree based on the principle of local density maximization.

[0031] In this embodiment, the local density of two nodes is equal to the edge weight between them multiplied by the ratio of the number of common adjacent nodes to the total number of adjacent nodes; the steps of constructing the density connection tree include:

[0032] Step 2.1, randomly select a node as the starting point of the density connection tree;

[0033] Step 2.2, select the adjacent node with the largest local density to the node in the tree and connect it to the density-connected tree;

[0034] Step 2.3, repeat step 2.2 until all nodes are added to the density-connected tree.

[0035] Step 3: Cut the tree according to the average edge weight, and preferentially remove the edge with the largest minimum average edge weight between the two node groups connected, so as to obtain the same number of node groups as the number of time-frequency resource blocks to be allocated, and combine the base stations with which the users are paired to obtain the initial user grouping results.

[0036] In this embodiment, the step of obtaining the initial user grouping result by cutting the tree includes:

[0037] Step 3.1, calculate the D value of each edge, that is, the minimum value of the average edge weight of the two node groups connected by the edge;

[0038] Step 3.2, remove the edge with the largest D value;

[0039] Step 3.3, repeat step 3.2 until L node groups are generated, where L represents the number of time-frequency resource blocks to be allocated;

[0040] Step 3.4: further divide the above group into N×L node groups according to the base stations corresponding to the nodes, where N represents the number of cells.

[0041] Step 4: Determine the user groups to which the remaining users belong based on the user grouping criterion of giving priority to no interference between users and secondly considering minimum system and rate loss.

[0042] In an embodiment, the direction in which the users to be grouped will be assigned when entering each user group can be determined first, that is, the main channel direction in which the users to be grouped are least interfered with when communicating with the corresponding base station on the time-frequency resource blocks corresponding to each user group, and then the group to which the users belong can be determined based on the user grouping criteria.

[0043] Exemplarily, the user grouping criteria can be described as:

[0044] Prioritize non-interference between the user to be grouped and the users in the group to be added: If the user to be grouped and all users in a certain user group do not interfere with each other in their respective assigned directions, the user will be added to the group; if there are multiple groups that do not interfere with each other, the group with the highest overlap between the main channel directions of all users in the group and the main channel direction of the user to be grouped is selected;

[0045] Secondly, considering the system sum rate loss, the user group with the smallest clustering cost is selected for the users to be grouped. The clustering cost of a user joining a user group is the difference between the original system sum rate and the system sum rate after the user joins the group.

[0046] The following further illustrates the user grouping method based on graph clustering provided by the embodiment of the present invention in conjunction with a specific multi-cell massive MIMO system scenario.

[0047] Consider a multi-cell massive MIMO system consisting of N base stations and K users. Each base station is equipped with a uniform linear array of M antennas. The antenna array spacing is d = 0.5λ, where λ is the carrier wavelength. Each user is equipped with a single antenna. After OFDM modulation, the system has L time-frequency resource blocks to be allocated. It is assumed that the channel coefficients in each time-frequency resource block remain constant. (The following analysis focuses on a single time slot and does not involve resource allocation in the time dimension.)

[0048] For a uniform linear array with half-wavelength spacing, the steering vector of the antenna array can be expressed as

[0049]

[0050] Where Ω = cos(θ) is the direction cosine and θ is the angle of arrival. On the time-frequency resource block l, the channel between base station n and user k is It can be expressed as

[0051]

[0052] Among them, Q n,k The number of paths between base station n and user k, a n,k,q is the channel gain of the qth path, f c is the carrier frequency, Δf is the frequency interval between adjacent time-frequency resource blocks, and are the time delay and direction cosine of the qth path respectively. and are the set of time delays and direction cosines of the multipath between base station n and user k, respectively.

[0053] Define the matrix V as in are direction cosines uniformly sampled in the range [-1,1), and X = sM is the number of direction cosines sampled, where Then, V is a semi-unitary matrix, that is, VV H =I, I is the unit matrix, H represents the conjugate transpose. n,k,l It can be approximated as

[0054]

[0055] in, The xth element of

[0056]

[0057] Among them, the collection and In this embodiment, V is called the spatial sampling matrix. It is called beam domain channel.

[0058] For the beam domain channel, the statistical channel covariance matrix Its (x, y)th element is calculated as

[0059]

[0060] Where * represents conjugate, and δ(·) is the Kronecker delta function. From equation (4), we can see that different elements of the beam domain channel represent the channel gains of different paths. When using large antennas, the base station can separate signals at different angles. Based on the uncorrelated scattering assumption, If each element of is an independent complex random variable with zero mean and different variance, then equation (a) holds, Γ n,k is a diagonal matrix. In addition, Γ n,k The xth diagonal element [Γ n,k ] x represents the channel energy of user k in the xth sampling direction of base station n. Since the number of channel multipaths is much smaller than the number of antennas, Γ n,k is a sparse diagonal matrix.

[0061] Consider the uplink transmission of a multi-cell massive MIMO system. Indicates the user grouping scheme, where user group Contains users communicating with base station n through time-frequency resource block l. The user set using time-frequency resource block l is expressed as The set of users served by base station n is expressed as The signal received by base station n on time-frequency resource block l can be expressed as

[0062]

[0063] in, is additive Gaussian noise, σ2 is the noise variance, is the data signal of user k, p k is the transmission power. Assume that each user sends the signal with equal power, that is, p k =p (p is a constant). Under linear detection, about the user The received signal is expressed as

[0064]

[0065] in, is the receiving vector of base station n on time-frequency resource block l for user k, and satisfies ||w n,K,l ||2=1, Contains interference and noise. Then, the user The signal-to-interference-noise ratio can be calculated as

[0066]

[0067] in, Since V is a semi-unitary matrix, substituting (3) into the expression of signal-to-interference-noise ratio yields equation (b). The uplink ergodic achievable rate is given by

[0068]

[0069] In order to reduce the computational complexity of the Monte Carlo average in the ergodic rate, this embodiment gives the following approximate rate

[0070]

[0071] in, It is γ n,k,l The numerator and denominator of are respectively the expected results. According to the properties of generalized Rayleigh entropy, The maximum value of

[0072]

[0073] in, Indicates that base station n is allocated to user direction, equal Then, the uplink sum rate can be expressed as

[0074]

[0075] With the goal of maximizing the system sum rate, the multi-cell user grouping problem can be expressed as

[0076]

[0077] The first and second conditions ensure that each user belongs to a user group and uses a time-frequency resource block and a base station for communication. The third condition ensures that all users are scheduled.

[0078] Since Γ n,k Is a sparse matrix, first define the set Contains the main channel direction between base station n and user k, where the constant Constant a> 0. Condition [Γ n,k ] x ≥max[θ·tr(Γ n,k ),a] represents [Γ n,k ] x Exceeds the threshold a and is much larger than the smaller element in the sparse matrix. is the direction of maximum channel gain between base station b and user k. According to formula (11), user Interference is received from other users assigned to the same time-frequency resource block and direction. Therefore, when grouping users, the channel direction of the users in the same group must be carefully considered. To maximize the system sum rate, this embodiment proposes the following user grouping criteria.

[0079] Rule 1: If and Users k1 and k2 are allocated the same time-frequency resource blocks, where n1 and n2 are the base station indexes of the pairing of users k1 and k2 (n1 may be equal to n2).

[0080] Criterion 2: If users k1, k2, and k3 all meet Criterion 1, and User k1 is preferentially allocated to the same time-frequency resource block as k2.

[0081] Rule 3: If and User k1 is assigned to user group in

[0082] Criteria 4: If user k1 and user group Meets criterion 3, and User k1 is assigned to the user group first

[0083] Criteria 1 and 2 are the pairing criteria between two users. Criteria 1 states that if the two users do not interfere with each other in the direction of maximum channel gain with their paired base stations, they can use the same time-frequency resource block. Criteria 2 states that if there are multiple non-interfering users, the user to be decided and the user with the highest overlap in the main channel direction use the same time-frequency resource block. Criteria 3 and 4 are the user grouping criteria when the initial user grouping is known. Criteria 3 states that if the user to be grouped and all users in a certain user group do not interfere with each other in their respective assigned directions, the user can be assigned to that group. Criteria 4 states that if there are multiple non-interfering groups, the user to be grouped joins the group with the highest overlap in the main channel direction with all users in the group.

[0084] Based on the multi-cell user grouping problem and user grouping criteria described above, this embodiment uses a weighted graph To describe the multi-cell user relationship, ε and Represents the set of nodes, edges and weights respectively. The central scheduler regards some users as nodes, and the node set is defined as

[0085]

[0086] Among them, the constant is the threshold. If the user User k and base station n k The channel gain between user k and base station n is significantly greater than the channel gain between user k and other base stations. k Pairing. Only when two users meet the grouping criterion 1 (i.e., and )hour, There is an edge between two nodes in The weight is determined by the minimum of the maximum channel gain of the two nodes and the ratio of the number of main channel directions they share to the total number of main channel directions, which is given by the following formula

[0087]

[0088] in, Given by the following formula

[0089]

[0090] According to the above graph, the local density of any two nodes can be calculated, which is determined by the edge weight between them and the ratio of the number of common adjacent nodes to the total number of adjacent nodes. The formula is expressed as:

[0091]

[0092] Among them, the adjacent node set

[0093] After calculating the local density, the multi-cell user relationship graph is converted into a density-connected tree. In this tree, each node in the original graph is connected to its most similar node based on local density. Specifically, a node is randomly selected as the starting point of the density-connected tree. Then, adjacent nodes with the highest local density to the node in the tree are gradually connected to the density-connected tree until all nodes are successfully added to the tree.

[0094] like Figure 2 As shown, the density-connected tree connects all nodes and does not form closed cycles. Cutting an edge in the density-connected tree can split the graph into two parts. Definition To evaluate the quality of graph clustering,

[0095]

[0096] Delete the edge with the largest D value to cut the density connection tree until L groups are obtained. According to the base station corresponding to each user, the L groups can be further divided into N×L groups This is the initial user grouping result.

[0097] For collections For users outside the group, especially those at the cell edge, the central scheduler groups them according to user grouping criteria 3 and 4. If there is no user group that meets grouping criteria 3 with the users to be grouped, then according to the clustering cost function Determine user groups. The clustering cost function is defined as

[0098]

[0099] in, is the sum rate of all users using time-frequency resource block l, After adding user k, we get the set The user group assigned to user k is

[0100] To help those skilled in the art better understand the present invention, the following demonstrates the performance and computational complexity of the multi-cell user grouping method based on graph clustering in this embodiment, using a specific system configuration. The simulation environment was generated using the QuaDRiGa platform, with a base station equipped with M = 128 antennas and each user equipped with a single antenna. Over 500 samples were simulated and analyzed, and the proposed algorithm was compared with benchmark algorithms: SC-MS, MADOC, and AUC.

[0101] Figure 3The user grouping performance of different algorithms was compared for a system configuration of K = 100, L = 6, and N = 3. At low signal-to-noise ratios, the differences in system sum rates between the algorithms were minimal, but the proposed method demonstrated superior performance. At high signal-to-noise ratios, the proposed method achieved 11%, 14%, and 19% higher system sum rates than the SC-MS, MADOC, and AUC algorithms, respectively.

[0102] Figure 4 The algorithm performance was compared as the number of users varied under a high signal-to-noise ratio (SNR) system configuration (L = 12, N = 3). As the number of users increased, the system sum rate gradually improved. However, when the number of users exceeded 700, the system sum rates using the SC-MS, MADOC, and AUC algorithms began to decline. Due to the effectiveness of the proposed user grouping criterion, the proposed method demonstrated superior performance in improving the system sum rate.

[0103] Table 1 Operation schedule

[0104]

[0105] *(N, L, K) represent the number of base stations, the number of time-frequency resource blocks, and the number of users, respectively

[0106] Table 1 provides the runtime of each algorithm under different system configurations. The MADOC algorithm, which uses the K-means clustering principle, has the shortest runtime, followed closely by the method of the present invention. The runtimes of the SC-MS and AUC algorithms increase significantly with the increase in the number of users.

[0107] Based on the same inventive concept, an embodiment of the present invention also discloses a multi-cell large-scale MIMO user grouping system based on graph clustering, including a central scheduler for grouping users according to a statistical channel covariance matrix with the goal of maximizing the system and rate. The same group of users uses the same time-frequency resource block to communicate with the same base station. The central scheduler includes: a graph establishment module for selecting some users as nodes based on the channel gain between the users and each base station to establish a multi-cell user relationship graph, and determining the edges and weights between the user nodes according to the statistical channel covariance matrix; a tree construction module for converting the above-mentioned multi-cell user relationship graph into a density-connected tree according to the principle of local density maximization; a tree cutting module for cutting the tree according to the average edge weight, preferentially removing the edges with the largest minimum average edge weight of the two connected node groups, obtaining the same number of node groups as the number of time-frequency resource blocks to be allocated, and combining the base stations paired with the users to obtain the initial user grouping results; a supplementary grouping module for determining the user groups to which the remaining users belong based on the user grouping criterion of giving priority to no interference between users and secondly considering the minimum system and rate loss.

[0108] The detailed working process of each module described above can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here. The division of the modules is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules can be combined or integrated into another system.

[0109] The present invention also provides a computer program product comprising a computer program / instructions that, when executed by a processor, implement the steps of the graph clustering-based multi-cell massive MIMO user grouping method. The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the steps of the method of the present invention are implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

Claims

1. A multi-cell massive MIMO user grouping method based on graph clustering, characterized in that: The method aims to maximize the system sum rate, groups users according to the beam domain channel statistical covariance matrix, and the same group of users communicates with the same base station using the same time-frequency resource block, including the following steps: Step 1: Based on the channel gains between users and each base station, some users are selected as nodes to establish a multi-cell user relationship graph, and the edges and weights between user nodes are determined according to the statistical channel covariance matrix; Step 2: Convert the multi-cell user relationship graph into a density connection tree based on the principle of local density maximization; Step 3: Cut the tree based on the average edge weight, preferentially removing the edge with the largest minimum average edge weight between the two connected node groups, to obtain the same number of node groups as the number of time-frequency resource blocks to be allocated, and combine the base stations with which the users are paired to obtain the initial user grouping result; Step 4: Determine the user groups to which the remaining users belong based on the user grouping criterion of giving priority to no interference between users and secondly considering minimum system and rate loss.

2. The multi-cell massive MIMO user grouping method based on graph clustering according to claim 1, characterized in that: Users are grouped according to the statistical covariance matrix of the beam domain channel between the user and each base station. The beam domain channel is the channel obtained by angular sampling of the spatial channel by the sampling matrix. Its statistical covariance matrix is ​​a diagonal matrix, and each diagonal element represents the channel energy in the corresponding direction. The direction corresponding to the diagonal element exceeding the threshold is the main channel direction between the user and the base station, and the direction corresponding to the maximum diagonal element in the main channel direction is the direction of the maximum channel gain between the user and the base station.

3. The multi-cell massive MIMO user grouping method based on graph clustering according to claim 1, characterized in that: In step 1, some users are selected as nodes, and each node is paired with the base station with the largest channel gain. The selected nodes satisfy the requirement that the ratio of the channel gain between the node and the paired base station to the sum of the channel gains between the user and all base stations exceeds a threshold; the direction assigned to each node is the direction of the maximum channel gain between the node and the paired base station. If the two nodes do not interfere with each other in their respective assigned directions, an edge exists between them, and the weight of the edge is equal to the minimum value of the maximum channel gain of the two nodes multiplied by the ratio of the number of their common main channel directions to the total number of main channel directions; the main channel direction between the user and the base station is the direction corresponding to the diagonal element in the statistical channel covariance matrix whose channel energy exceeds a set threshold.

4. The multi-cell massive MIMO user grouping method based on graph clustering according to claim 3, characterized in that: Assume that users k1 and k2 are nodes in the figure, paired with base stations n1 and n2 respectively, and assigned directions l1 and l2 respectively. The two users do not interfere with each other in their respective assigned directions if the main channel direction between user k1 and base station m2 does not include l2, and the main channel direction between user k2 and base station n1 does not include l1.

5. The multi-cell massive MIMO user grouping method based on graph clustering according to claim 1, characterized in that: In step 2, the local density of two nodes is equal to the edge weight between them multiplied by the ratio of the number of common adjacent nodes to the total number of adjacent nodes; The steps to construct a density-linking tree include: Step 2.1, randomly select a node as the starting point of the density connection tree; Step 2.2, select the adjacent node with the largest local density to the node in the tree and connect it to the density-connected tree; Step 2.3, repeat step 2.2 until all nodes are added to the density-connected tree.

6. The multi-cell massive MIMO user grouping method based on graph clustering according to claim 1, characterized in that: In step 3, the step of obtaining the initial user grouping result by cutting the tree includes: Step 3.1, calculate the D value of each edge, that is, the minimum value of the average edge weight of the two node groups connected by the edge; Step 3.2, remove the edge with the largest D value; Step 3.3, repeat step 3.2 until L node groups are generated, where L represents the number of time-frequency resource blocks to be allocated; Step 3.4: further divide the above group into N×L node groups according to the base stations corresponding to the nodes, where N represents the number of cells.

7. The multi-cell massive MIMO user grouping method based on graph clustering according to claim 1, characterized in that: In step 4, the direction in which the users to be grouped will be assigned when entering each user group is first determined, that is, the main channel direction in which the users to be grouped receive the least interference when communicating with the corresponding base station on the time-frequency resource blocks corresponding to each user group, and then the group to which the users belong is determined based on the user grouping criteria.

8. The multi-cell massive MIMO user grouping method based on graph clustering according to claim 7, characterized in that: The user grouping criteria are: Prioritize non-interference between the user to be grouped and the users in the group to be added: If the user to be grouped and all users in a certain user group do not interfere with each other in their respective assigned directions, the user will be added to the group; if there are multiple groups that do not interfere with each other, the group with the highest overlap between the main channel directions of all users in the group and the main channel direction of the user to be grouped is selected; Secondly, considering the system sum rate loss, the user group with the smallest clustering cost is selected for the users to be grouped. The clustering cost of a user joining a user group is the difference between the original system sum rate and the system sum rate after the user joins the group.

9. A multi-cell massive MIMO user grouping system based on graph clustering, characterized in that: The system includes a central scheduler configured to group users according to a beam-domain channel statistical covariance matrix with the goal of maximizing system summation rate, so that users in the same group communicate with the same base station using the same time-frequency resource blocks. The central scheduler includes: A graph building module is used to select some users as nodes based on the channel gains between users and each base station to establish a multi-cell user relationship graph, and determine the edges and weights between user nodes based on the statistical channel covariance matrix; A tree construction module, configured to convert the multi-cell user relationship graph into a density connection tree according to a local density maximization principle; A tree cutting module is used to cut the tree according to the average edge weight, preferentially removing the edge with the largest minimum average edge weight between the two connected node groups, to obtain the same number of node groups as the number of time-frequency resource blocks to be allocated, and to obtain the initial user grouping result based on the base stations paired with the users; The supplementary grouping module is used to determine the user groups to which the remaining users belong based on the user grouping criteria of first considering no interference between users and secondly considering minimum system and rate loss.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the multi-cell massive MIMO user grouping method based on graph clustering are implemented according to any one of claims 1 to 8.

Images