A resource allocation method based on hybrid clustering in ultra-dense networks

Abstract

The invention belongs to the technical field of ultra-dense network resource allocation, and discloses a resource allocation method based on hybrid clustering in an ultra-dense network. The method comprises: obtaining data of original base stations and user locations, and preprocessing the data; clustering micro base stations by using a Canopy algorithm and a K-means algorithm; constructing an interference graph for users in each cluster after clustering, and grouping the users by using a coloring algorithm according to the interference graph; allocating sub-channels to users in each group by using a greedy algorithm; and allocating power to users in each group by using a water injection algorithm. The invention can accurately and quickly cluster micro base stations by using a clustering algorithm combining the two algorithms, and then group users, allocate sub-channels, and allocate power, so as to effectively reduce the same-layer interference between base stations and improve the throughput and performance of the system.

Description

Technical Field

[0001] The invention belongs to the technical field of ultra-dense network resource allocation, and discloses a resource allocation method based on hybrid clustering in an ultra-dense network. Background Art

[0002] Wireless networks are facing huge challenges due to the explosive growth of data traffic. In addition to meeting the requirements of high-speed data transmission, the fifth-generation mobile communication system (5G) also needs to meet the needs of different application scenarios such as large capacity, ultra-reliability, low latency and ubiquitous access. Ultra-dense networks (UDN) are a promising approach to meet these requirements.

[0003] In ultra-dense networks, the number of micro base stations is far greater than that of traditional networks. At the same time, densely distributed micro base stations will shorten the transmission distance between users and base stations, which can significantly improve the throughput of the entire system. However, the dense distribution of base stations and the reduction in distance will cause serious interference to users, resulting in a situation where losses outweigh benefits. Therefore, reducing interference and reasonably allocating resources to improve system performance are key issues in ultra-dense networks.

[0004] Research shows that clustering combined with resource allocation can reduce intra-layer and cross-layer interference. The problem of resource allocation in ultra-dense networks is an NP-hard problem, and it is very complicated to find the global optimal solution. Therefore, this optimization problem is decomposed into several sub-problems, and each of these sub-problems has a corresponding solution. For example, when clustering base stations, the k-means algorithm is used to divide the base stations with the greatest mutual interference into a cluster according to the distance between the base stations, so that the inter-cluster interference is minimized as much as possible, or even negligible. However, the k value and the initial cluster center of the traditional k-means algorithm are given in advance, which will lead to inaccurate final clustering results and fail to reflect the actual interference relationship, which will also lead to the inaccurate allocation of resources in the next step.

[0005] After the micro base stations are clustered, base stations in the same cluster may also cause interference to users connected to each other, making it difficult for users to obtain a rate that meets the service quality. Summary of the invention

[0006] In order to solve the problems existing in the above prior art, the present invention proposes a resource allocation method based on hybrid clustering in an ultra-dense network, the method comprising: constructing an ultra-dense network, the ultra-dense network being a two-layer heterogeneous cellular downlink ultra-dense network, that is, containing a central macro cell base station and multiple micro cell base stations in the network;

[0007] S1: Base stations and users are randomly distributed according to HPPP, and data in the network is obtained, which includes the location of the base station, the maximum transmission power of the base station, and the location of the user;

[0008] S2: Use Canopy algorithm and K-means algorithm to cluster micro base stations;

[0009] S3: construct an interference graph for the users in each cluster after clustering, and use the coloring algorithm to group the users according to the interference graph;

[0010] S4: Use a greedy algorithm to allocate subchannels to users in each group;

[0011] S5: Use the water injection algorithm to allocate power to users in each group.

[0012] Preferably, the process of clustering micro base stations using the Canopy algorithm and the K-means algorithm includes:

[0013] S21: Setting micro base station distance thresholds T1 and T2, where T1>T2;

[0014] S22: Collect from micro base station Select a point f from the , take f as the first class, and take this class as Canopy;

[0015] S23: From the collection Randomly select a point p from the , calculate the distance from p to all the generated Canopies, if the distance from the point to the Canopy is less than T1, add p to the Canopy, if the distance from the point to the Canopy is greater than T1, then take p as a new Canopy;

[0016] S24: Calculate the distance from point f to the new Canopy. If the distance is less than T2, remove point f from the set Otherwise, re-enter the collection Select points;

[0017] S25: Pair collection Repeat the above steps S23 and S24 until the set is empty, and we get the k value and the approximate k initial center sets C = {1, ..., c k , …, C K};

[0018] S26: Cluster the micro base stations using the K-means algorithm based on the k initial center sets.

[0019] Furthermore, the process of clustering micro base stations using the K-means algorithm includes:

[0020] S261: Determine the k value and the cluster center set C according to the clustering result of Canopy;

[0021] S262: Use Calculate the distances from all samples to the k centroids and assign them to the cluster with the closest distance; where f represents the points in the set and c k represents the k-th initial clustering center;

[0022] S263: Use (k = 1, 2,..., K) to calculate the new center of the cluster, and take the calculated mean as the new clustering center; n k represents the number of clusters, k represents the cluster number, and K represents the maximum cluster number;

[0023] S264: Repeat the above steps S262 and S263 until the clustering centers no longer change.

[0024] Preferably, the process of grouping users using a coloring algorithm according to the interference graph includes:

[0025] S31: In an FBS cluster, use γ i1 to represent the interference level of user i1, and calculate the interference level of each user: where FBS represents a femtocell;

[0026] where represents the sum of the interference signal powers received by user i1, represents the useful signal power received by user i1;

[0027] S32: Calculate the interference level between user and user

[0028] S33: Use the matrix E to describe the interference level weights between any two UEs in the FBS cluster according to the interference level, that is:

[0029]

[0030] where E th represents the interference weights between UEs within the same femtocell;

[0031] S34: Set the interference threshold T3. If E(i1, i2) ≥ T3, there is interference between user i1 and user i2; if E(i1, i2) < T3, there is no interference between user i1 and user i2;

[0032] S35: Put all users in a cluster as vertices into the set V, randomly select a vertex v from the set V, and put it into the initially empty set α, that is: α = {v};

[0033] S36: construct a candidate set β=V\α, that is, store the elements in the set V except the elements in the set α into the candidate set β;

[0034] S37: When the sum of the weights of the users in set α is less than E th When , select vertex v′ from the candidate set β, which can minimize the interference weight within the user cluster, and the candidate user cluster becomes α=α∪{v′}, and repeat S35-S36;

[0035] S38: When all vertices in α are removed from V, extract the vertices in α as a user cluster, and repeat S35 to S37 until V is an empty set and the user grouping is completed.

[0036] Preferably, the process of allocating subchannels using a greedy algorithm includes:

[0037] S41: sorting the user groups in descending order according to the number of sub-channels contained in the gateway units GU in the user groups;

[0038] S42: Calculate GU' l At the communication rate of each sub-channel, all sub-channels whose communication rates meet the quality of service are selected;

[0039] S43: Calculate GU' l The signal-to-noise ratio on the selected subchannel is used to find the best subchannel;

[0040] S44: Determine GU' in step S42 l Is the optimal subchannel occupied by other GUs? If not, assign the subchannel to GU' l , and remove the subchannel from the channel set L; if it is occupied, find a suboptimal subchannel in the remaining available subchannels and assign it to GU' l , and remove the subchannel from the set L;

[0041] S45: Repeat the above steps S42 to S44 until the sub-channel set L is empty.

[0042] Furthermore, calculate GU' l The formula for the signal-to-noise ratio on the selected subchannel is:

[0043]

[0044] in, represents the micro base station set, U f,i represents the user served by micro base station i, represents the power allocated by micro base station i to user through subchannel l, represents the channel gain of micro base station i on subchannel l to its serving user n, δ2 represents the variance of additive white Gaussian noise.

[0045] Preferably, the process of allocating power to users in each group using a water injection algorithm includes:

[0046] S51: Determine the power allocation objective function and constraint conditions for each user to allocate their own subchannel;

[0047] S52: According to the KKT theorem, the Lagrangian function is obtained; its expression is:

[0048]

[0049] Where p represents the power allocation vector, α represents the Lagrange constant multiplier, and β represents the Lagrange constant multiplier. represents the micro base station set, U f represents the micro base station user set, L represents the subchannel set, W represents the bandwidth of each subchannel, represents the signal-to-noise ratio of the signal received by user n from micro base station i in subchannel l, p f Indicates the maximum transmit power of FBS, represents the transmit power of micro base station i in subchannel l;

[0050] S53: Take the partial derivative of the power of the Lagrangian function to obtain the transmission power received by user n from micro base station i in subchannel l Its expression is:

[0051]

[0052]

[0053] in, represents the gain-to-noise ratio received by user n from micro base station i in subchannel l; [.] + It means comparing the number in the symbol with 0. If it is greater than 0, then output the number, otherwise select 0; represents the channel gain of micro base station i on subchannel l to its serving user n, U f,i represents the user served by micro base station i, represents the transmit power of micro base station j in subchannel l, δ 2 represents the variance of additive Gaussian white noise;

[0054] S54: As the water filling line of the water filling algorithm, the power allocated to each user is obtained according to the water filling line.

[0055] Furthermore, the power allocation objective function and constraints are:

[0056]

[0057]

[0058]

[0059] Among them, R total Indicates the total throughput of the network, C1 indicates that the allocated power cannot be a negative value, and C2 indicates that the total power allocated to all users cannot exceed the load of the current base station. represents the transmit power of micro base station i in subchannel l, p f Indicates the maximum transmit power of the FBS.

[0060] To achieve the above objectives, the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned resource allocation methods based on hybrid clustering in an ultra-dense network.

[0061] To achieve the above-mentioned purpose, the present invention also provides a resource allocation device based on hybrid clustering in an ultra-dense network, comprising a processor and a memory; the memory is used to store a computer program; the processor is connected to the memory, and is used to execute the computer program stored in the memory, so that the resource allocation device based on hybrid clustering in an ultra-dense network executes any of the above-mentioned resource allocation methods based on hybrid clustering in an ultra-dense network.

[0062] Beneficial effects of the present invention:

[0063] The present invention aims at solving the problems of spectrum shortage, serious interference and complex resource allocation in ultra-dense network scenarios, and proposes a resource allocation method based on hybrid clustering. First, the Canopy algorithm is used to perform "coarse" clustering of micro base stations to obtain the approximate number of clusters and the centers of clusters, and the K-means algorithm is used to perform "fine" clustering to evenly cluster the base stations according to distance. Secondly, an interference graph is constructed for the users in each cluster after clustering, and the users with small interference are grouped into one group using a graph coloring algorithm. Then, a greedy algorithm is used to allocate sub-channels to the users in each group. Finally, a water injection algorithm is used to allocate power to the users. In summary, the present invention can accurately and quickly cluster micro base stations through a clustering algorithm combining two algorithms, and then group users, allocate sub-channels and allocate power, which can effectively reduce the same-layer interference between base stations and improve the throughput and performance of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 It is the overall flow chart of the present invention;

[0065] Figure 2 This is a flow chart of clustering micro base stations based on Canopy and K-means algorithms in the present invention;

[0066] Figure 3 This is a flow chart of grouping users based on the graph coloring algorithm of the present invention;

[0067] Figure 4 A flow chart of allocating sub-channels to users according to the present invention;

[0068] Figure 5 An overall block diagram provided for an embodiment of the present invention;

[0069] Figure 6 This is a two-layer ultra-dense network model diagram of the present invention. DETAILED DESCRIPTION

[0070] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0071] The purpose of the embodiments of the present application is to provide a resource allocation method based on hybrid clustering in an ultra-dense network, so as to solve the problem of low overall network efficiency caused by interference in the ultra-dense network.

[0072] A specific implementation method of a resource allocation method based on hybrid clustering in an ultra-dense network, such as Figure 1 and Figure 5 As shown, the method includes:

[0073] S1: Base stations and users are randomly distributed according to HPPP, and data in the network is obtained, which includes the location of the base station, the maximum transmission power of the base station, and the location of the user;

[0074] S2: Use Canopy algorithm and K-means algorithm to cluster micro base stations;

[0075] S3: construct an interference graph for the users in each cluster after clustering, and use the coloring algorithm to group the users according to the interference graph;

[0076] S4: Use a greedy algorithm to allocate subchannels to users in each group;

[0077] S5: Use the water injection algorithm to allocate power to users in each group.

[0078] Among them, this paper considers a two-layer heterogeneous cellular downlink ultra-dense network, that is, the network contains a central macro base station (MBS) and multiple micro base stations (FBS, femto base station).

[0079] like Figure 2 As shown in the figure, when clustering micro base stations, the Canopy algorithm is first used to quickly obtain the number and location of the approximate centers of the clusters, and then the K-means algorithm is used to obtain stable and accurate clustering results. The process of clustering micro base stations by combining the Canopy algorithm and the K-means algorithm includes:

[0080] S21: Select two distance thresholds of the base station: T1 and T2, where T1>T2;

[0081] S22: Collect from micro base station Take out a point f and make f the first class, this class is called Canopy;

[0082] S23: Continue from the collection Take point p, calculate the distance from p to all generated Canopies, if the distance to a Canopy is less than T1, add p to the Canopy. If the distance from p to the center of all Canopies is greater than T1, take p as a new Canopy;

[0083] S24: Calculate the distance from point f to the new Canopy. If the distance from point f to the Canopy is less than T2, it means that it is close enough to the Canopy. In this case, remove f from the set. Delete to avoid f being added to other Canopies repeatedly;

[0084] S25: Pair collection Repeat the above steps S23 and S24 until the set is empty, the algorithm ends, and the “coarse” clustering is completed, and the k value and k initial center sets C = {1,…,c k , …, C K};

[0085] Among them, the size of T1 affects the number of clusters, and T1 is selected to be larger than T2.

[0086] The size of T2 affects the coverage radius of the cluster. The selection of T2 is:

[0087]

[0088] Where i and j both represent micro base stations, λ represents the proportional coefficient for adjusting the cluster size, and d i,j Represents the distance between micro base station i and micro base station j. The proportional coefficient λ for adjusting the cluster size changes dynamically with the base station density, and can be set as a coefficient function associated with the actual application scenario range and user density, and the coefficient is adjusted according to the actual scenario and experience.

[0089] Performing "fine clustering" on the femto base stations includes:

[0090] S221: Determine the value of k and the cluster center set C according to the clustering result of Canopy;

[0091] S222: Use to calculate the distances from all samples to the k centroids and assign them to the cluster with the nearest distance;

[0092] S223: Use (k = 1, 2,..., K) to calculate the new centers of the clusters and take the calculated mean as the new clustering centers; n k represents the number of clusters, k represents the serial number of the cluster, and K represents the maximum serial number of the cluster;

[0093] S224: Repeat the above steps S222 and S223 until the clustering centers no longer change.

[0094] Construct an interference graph for the users in each cluster after clustering, as Figure 3 shown. The process of grouping the users with less interference using the graph coloring algorithm includes:

[0095] S31: In an FBS cluster, use to represent the interference degree of user i1 and calculate the interference degree of each user:

[0096] where, represents the sum of the interference signal powers received by user i1, represents the useful signal power received by user i1.

[0097] S32: Calculate the interference degree between user and user :

[0098] S33: Use the matrix E to describe the interference degree weights between any two UEs in the FBS cluster. Among them

[0099]

[0100] where, E th represents the interference weights between UEs within the same femto base station. Since the interference between UEs within a femto base station is relatively large, a very high weight needs to be set. Let E th = ∞.

[0101] S34: Set the interference threshold T3. If E(i1, i2) ≥ T3, there is interference between user i1 and user i2; if E(i1, i2) < T3, there is no interference between user i1 and user i2;

[0102] S35: put all users in a cluster as vertices into set V;

[0103] S36: Randomly select a vertex v from the set V and put it into the initially empty set α, α = {v};

[0104] S37: construct a candidate set β=V\α; that is, store the elements in the set V except the elements in the set α into the candidate set β;

[0105] S38: When the sum of the weights of the users in set α is less than E th When , select vertex v′ from the candidate set β, which can minimize the interference weight within the user cluster, and the candidate user cluster becomes α=α∪{v′}, repeat S35~S37

[0106] S39: When all vertices in α are removed from V, extract the vertices in α as a user cluster, and repeat S35 to S38 until V is an empty set. The user grouping is completed and the number of user groups is represented by N.

[0107] like Figure 4 As shown, the process of allocating subchannels using a greedy algorithm for users in each group includes:

[0108] S41: sorting the user groups in descending order according to the number of gateway units GU contained in the user groups;

[0109] GU'={GU'1,GU'2,…,GU' N}

[0110] S42: Calculate GU' l At the communication rate of each sub-channel, all sub-channels whose communication rates meet the quality of service are selected;

[0111] S43: Calculate GU' l The signal-to-noise ratio on the selected subchannel is used to find the best subchannel;

[0112] S44: Determine GU' in step S42 l Is the optimal subchannel occupied by other GUs? If not, assign the subchannel to GU' l , and remove the subchannel from the set L; if it is occupied, find a suboptimal subchannel from the remaining available subchannels and assign it to GU' l , and remove the subchannel from the set L;

[0113] S45: Repeat the above steps S42 to S44 until the sub-channel set L is empty.

[0114] The subchannel set is represented by: L = {1, ..., l, ..., L}. The FBS set is represented by: The micro base station user set is represented as: U f ={1,…,u f ,…U f}, and n represents the user.

[0115] After completing FBS clustering and user grouping, the resource allocation level steps will be carried out. Figure 6 As shown, the ultra-dense network model of the present invention is a two-layer ultra-dense network model. For each macro base station user n in the cell, its communication service is directly provided by the macro base station and reuses a frequency band alone, which does not overlap with the micro base station frequency band. Therefore, cross-layer interference will not be considered in this paper. For micro base station user n, if FBSi transmits data to user n through subchannel l, the signal to interference plus noise ratio (γ) of user n on subchannel l under the service of micro base station i is expressed as:

[0116]

[0117] in, represents the power allocated by micro base station i to user through subchannel l, represents the channel gain of micro base station i on subchannel l to its serving user n. 2 Represents the variance of additive white Gaussian noise (AWGN).

[0118] The transmission rate of the micro base station user in subchannel l is obtained:

[0119]

[0120] Wherein, W represents the bandwidth of each subchannel.

[0121] The total throughput of micro base station users is:

[0122]

[0123] in, Indicates whether subchannel l is allocated to user n. This means that subchannel l is allocated to user n. This means that subchannel l is not allocated to user n.

[0124] Based on the FBS clustering and user grouping, in order to maximize the system throughput of UDN and consider the user's QoS guarantee, the following optimization problem can be established:

[0125]

[0126]

[0127]

[0128]

[0129]

[0130]

[0131] Among them, constraints C1 and C4 indicate that a subchannel can only be allocated to one user, and one user can only be allocated one subchannel; constraint C2 indicates that the allocated power cannot be a negative value; constraint C3 indicates that the total power allocated to all users cannot exceed the load of the current base station; constraint C5 indicates that the user's rate must reach the rate required by its service quality, R QOS Indicates the rate that guarantees the user's service quality.

[0132] When the clustering effect is ideal, the interference between clusters can be basically ignored, that is, it can be assumed that the interference between adjacent clusters has been eliminated.

[0133] The above formula is a mixed integer non-convex nonlinear optimization problem, which is still an NP-hard problem. Based on the clustering results in the ultra-dense network mentioned above, considering that each cluster has sufficient spatial isolation, the optimization problem in the above formula is solved.

[0134] Assume that each micro base station in the cluster has equal power allocation on each orthogonal subcarrier, that is, At this point, the optimization problem in the above formula can be converted into solving variables The subchannel optimization problem. In fact, is a binary discrete variable, which is difficult to solve. In order to solve the optimization problem, we can first convert the binary variable Put it in front The above formula still holds true. Then, the binary variable Relax to the interval [0,1], that is, the variable It can take any value continuously in the interval [0,1]. In this way, the original optimization problem is transformed into a convex optimization problem with inequality constraints. The objective function in the above formula can be transformed into the following optimization problem:

[0135]

[0136]

[0137]

[0138]

[0139]

[0140] The above optimization problem is solved by using the KKT condition in convex optimization theory, and the Lagrange multiplier method is used to obtain:

[0141]

[0142] Among them, χ, δ, ε are all Lagrange multipliers, and y represents the subchannel allocation vector.

[0143] according to have to:

[0144]

[0145] in, and That is to say, the subchannel should be allocated to the user who can obtain the maximum signal to interference noise ratio on the resource block. From this conclusion, a greedy resource allocation scheme can be adopted in ultra-dense networks.

[0146] The process of allocating power to users using the water injection algorithm includes:

[0147] S51: Determine the objective function and constraints of power allocation:

[0148]

[0149]

[0150]

[0151] S52: According to the KKT theorem, the Lagrangian function is obtained:

[0152]

[0153] S53: Take the partial derivative of the Lagrangian function with respect to power and obtain the power received by user n from micro base station i in the sub-signal

[0154] Transmitting power of channel l Its expression is:

[0155]

[0156]

[0157]

[0158] Where W represents the subchannel bandwidth, α represents the Lagrange multiplier, and β represents the Lagrange multiplier. represents the gain-to-noise ratio received by user n from micro base station i in subchannel l; [.] + It means comparing the number in the symbol with 0. If it is greater than 0, then output the number, otherwise select 0; represents the channel gain of micro base station i to its service user n on subchannel l, F represents the number of micro base station users, and U f,l represents the user served by micro base station i, represents the transmit power of micro base station j in subchannel l, δ 2 Represents the variance S54 of additive white Gaussian noise: As the water filling line of the water filling algorithm, the power allocated to each user is obtained according to the water filling line.

[0159] In one embodiment of the present invention, the present invention further includes a computer-readable storage medium having a computer program stored thereon, and when the program is executed by a processor, the program implements any of the above-mentioned resource allocation methods based on hybrid clustering in an ultra-dense network.

[0160] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to the computer program. The aforementioned computer program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk and other media that can store program codes.

[0161] A resource allocation device based on hybrid clustering in an ultra-dense network comprises a processor and a memory; the memory is used to store a computer program; the processor is connected to the memory and is used to execute the computer program stored in the memory, so that the resource allocation device based on hybrid clustering in an ultra-dense network executes any of the above-mentioned resource allocation methods based on hybrid clustering in an ultra-dense network.

[0162] Specifically, the memory includes: ROM, RAM, magnetic disk, USB flash drive, memory card or optical disk, etc., which can store program codes.

[0163] Preferably, the processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.

[0164] The above embodiments further illustrate the purpose, technical solutions and advantages of the present invention in detail. It should be understood that the above embodiments are only preferred implementation modes of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made to the present invention within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A resource allocation method based on hybrid clustering in an ultra-dense network, characterized in that: Including: Constructing an ultra-dense network, which is a two-layer heterogeneous cellular downlink ultra-dense network, that is, a central macro cellular base station and multiple micro cellular base stations are included in the network; S1: The base stations and users are randomly distributed according to HPPP, and the data in the network is obtained, including the locations of the base stations, the maximum transmission powers of the base stations, and the locations of the users; S2: Using the Canopy algorithm and the K-means algorithm to cluster the micro base stations; That is: S21: Setting the distance thresholds T1 and T2 for the micro base stations, where T1 > T2; S22: Collect from micro base station Select a point f from the , take f as the first class, and take this class as Canopy; S23: From the collection Randomly select a point p from the , calculate the distance from p to all the generated Canopies, if the distance from the point to the Canopy is less than T1, add p to the Canopy, if the distance from the point to the Canopy is greater than T1, then take p as a new Canopy; S24: Calculate the distance from point f to the new Canopy. If the distance is less than T2, remove point f from the set Otherwise, re-enter the collection Select points; S25: Pair collection Repeat the above steps S23 and S24 until the set is empty, and the k value and k initial center sets C = {1, ..., c k , …, C k }; S26: Using the K-means algorithm to cluster the micro base stations according to k initial center sets; specifically: S261: Determining the value of k and the cluster center set C according to the clustering results of Canopy; S262: Use Calculate the distances of all samples to k centroids and assign them to the cluster with the closest distance; where f represents the set The point in k represents the kth initial cluster center; S263: Use Calculate the new center of the cluster and use the calculated mean as the new cluster center; n k represents the number of clusters, k represents the sequence number of the cluster, and K represents the maximum sequence number of the cluster; S264: Repeating the above steps S262 and S263 until the cluster centers no longer change S3: Constructing an interference graph for the users in each cluster after clustering, and using a coloring algorithm to group the users according to the interference graph; specifically: S31: Used in an FBS cluster Indicates the interference level of user i1 and calculates the interference level of each user: Among them, FBS represents micro base station; in, represents the sum of the interference signal powers received by user i1, represents the useful signal power received by user i1; S32: Calculation User With users The degree of interference between S33: Describing the interference degree weights between any two UEs in the FBS cluster using the matrix E according to the interference degree, that is: Among them, E th Indicates the interference weight between UEs in the same micro base station; S34: Setting the interference threshold T3. If E(i1, i2) ≥ T3, there is interference between user i1 and user i2; if E(i1, i2) < T3, there is no interference between user i1 and user i2; S35: Putting all the users in a cluster as vertices into the set V, randomly selecting a vertex v from the set V, and putting it into the set α that is initially empty, that is: α = {v}; S36: Constructing the candidate set β = V\α, that is, storing the elements in the set V except the elements in the set α into the candidate set β; S37: When the sum of the weights of the users in set α is less than E th When , select vertex v′ from the candidate set β, which can minimize the interference weight within the user cluster, and the candidate user cluster becomes α=α∪{v′}, and repeat S35-S36; S38: When all the vertices in α are removed from V, extracting the vertices in α as a user cluster, and repeating S35 to S37 until V is an empty set, and the user grouping ends; S4: Using the greedy algorithm to allocate sub-channels to the users in each group; S5: Using the water-filling algorithm to perform power allocation for the users in each group.

2. The method for resource allocation based on hybrid clustering in an ultra-dense network according to claim 1, characterized in that: The process of allocating sub-channels using the greedy algorithm includes: S41: Sorting the user groups in descending order according to the number of sub-channels included in the gateway unit GU in the user group; S42: Calculate GU' l At the communication rate of each sub-channel, all sub-channels whose communication rates meet the quality of service are selected; S43: Calculate GU' l The signal-to-noise ratio on the selected subchannel is used to find the best subchannel; S44: Determine GU' in step S42 l Is the optimal subchannel occupied by other GUs? If not, assign the subchannel to GU' l , and remove the subchannel from the channel set L; if it is occupied, find a suboptimal subchannel in the remaining available subchannels and assign it to GU' l , and remove the subchannel from the set L; S45: Repeating the above steps S42 to S44 until the sub-channel set L is empty.

3. The method for resource allocation based on hybrid clustering in an ultra-dense network according to claim 2, characterized in that: Calculate GU' l The formula for the signal-to-noise ratio on the selected subchannel is: in, represents the micro base station set, U f,i represents the user served by micro base station i, represents the power allocated by micro base station i to user through subchannel l, represents the channel gain of micro base station i on subchannel l to its serving user n, δ 2 represents the variance of additive white Gaussian noise.

4. The method for resource allocation based on hybrid clustering in an ultra-dense network according to claim 1, characterized in that: The process of performing power allocation for the users in each group using the water-filling algorithm includes: S51: Determining the power allocation objective function and constraint conditions for each user to allocate its own sub-channel; S52: According to the KKT theorem, obtaining the Lagrangian function; its expression is: Where p represents the power allocation vector, α represents the Lagrange constant multiplier, and β represents the Lagrange constant multiplier. represents the micro base station set, U f represents the micro base station user set, L represents the subchannel set, W represents the bandwidth of each subchannel, represents the signal-to-noise ratio of the signal received by user n from micro base station i in subchannel l, p f Indicates the maximum transmit power of FBS, represents the transmit power of micro base station i in subchannel l; S53: Take the partial derivative of the power of the Lagrangian function to obtain the transmission power received by user n from micro base station i in subchannel l Its expression is: in, represents the gain-to-noise ratio received by user n from micro base station i in subchannel l; [.] + It means comparing the number in the symbol with 0. If it is greater than 0, then output the number, otherwise select 0; represents the channel gain of micro base station i on subchannel l to its serving user n, U f,i represents the user served by micro base station i, represents the transmit power of micro base station j in subchannel l, δ 2 represents the variance of additive Gaussian white noise; S54: As the water filling line of the water filling algorithm, the power allocated to each user is obtained according to the water filling line.

5. The method for resource allocation based on hybrid clustering in an ultra-dense network according to claim 4, characterized in that: The power allocation objective function and constraint conditions are: Among them, R total Indicates the total throughput of the network, C1 indicates that the allocated power cannot be a negative value, and C2 indicates that the total power allocated to all users cannot exceed the load of the current base station. represents the transmit power of micro base station i in subchannel l, p f Indicates the maximum transmit power of the FBS.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: The computer program is executed by the processor to implement the resource allocation method based on hybrid clustering in the ultra-dense network according to any one of claims 1 to 5.

7. A resource allocation device based on hybrid clustering in an ultra-dense network, characterized in that: Including a processor and a memory; the memory is used to store a computer program; the processor is connected to the memory and is used to execute the computer program stored in the memory, so that the resource allocation device based on hybrid clustering in an ultra-dense network executes the resource allocation method based on hybrid clustering in the ultra-dense network according to any one of claims 1 to 5.

Images