A user association method for non-uniform service capacity coverage of resource cells

Abstract

The present invention proposes a user association method for non-uniform service capacity coverage of resource cells, and the implementation steps are: constructing an ultra-dense network system; initializing parameters; establishing an optimization problem of user association; calculating the Q table in the Q learning algorithm; and obtaining the user association result for non-uniform service capacity coverage. The present invention proposes a user association algorithm based on the Q learning algorithm, firstly solving the best user association result to maximize the network capacity while meeting the user's service quality requirements and the base station transmission power limit, and changing the network service structure while meeting the user's service quality requirements and the base station transmission power limit, so that the non-uniform user services in the network can be uniformly served, thereby reducing the interference caused by resource use conflicts to improve the network capacity.

Description

Technical Field

[0001] The invention belongs to the technical field of communication systems, and relates to a user association method for non-uniform service capacity coverage of resource cells, which can be applied to ultra-dense network downlinks with uneven user service distribution. Background Art

[0002] In recent years, with the rapid development and popularization of smart mobile terminals, mobile communication services have shown explosive growth, giving rise to a large number of mobile service demands. In order to improve network capacity, ultra-dense networking technology has been considered by the industry as the most effective way to improve network capacity for the next generation of mobile communication systems. Through the deployment of a large number of base stations, reducing the propagation distance of wireless signals, and increasing the degree of spatial multiplexing of spectrum resources, the network capacity has been greatly improved, that is, the traditional cellular network has gradually turned to miniaturization and densification, and finally formed an ultra-dense network. The deployed base stations have low power, small service range, and small number of service users. In the ultra-dense network scenario, the dense deployment of base stations leads to serious interference between cells, which is highly sudden when user services are unevenly distributed. Interference is caused by conflicts in the use of wireless resources;

[0003] Therefore, in ultra-dense network scenarios, we consider using resource cell generation technology to provide communication services to users. Since it can achieve flexible adaptation of coverage structure as business distribution changes, resource cells can be flexibly generated to convert interference signals into useful signals and eliminate capacity coverage holes. However, it is also necessary to solve the network interference between dense base stations and resource cells. Specifically, when base stations are densely deployed and user services are unevenly distributed, the conflicting use of resources will cause interference, making it difficult to increase network capacity. Users must also consider more factors when selecting base stations for association. A suitable user association must not only consider the transmission rate requirements of all users and improve network capacity, but also consider the interference problem in the network. This is a great challenge for the user association algorithm. A suitable user association algorithm is of great significance for reducing interference and improving network capacity.

[0004] Nowadays, user association algorithms are constantly being proposed to improve network performance in ultra-dense networks. For example, the patent application with application publication number CN112383932A, entitled "Joint Optimization Method for User Association and Resource Allocation Based on Clustering", discloses a joint optimization method for user association and resource allocation based on clustering, which constructs an ultra-dense network architecture based on the separation of control plane and user plane, formulates a network energy efficiency optimization scheme based on user association, sub-channel allocation and power coordination based on clustering, and proposes a joint optimization algorithm for user association, sub-channel allocation and power coordination based on clustering. The algorithm adopts an alternating optimization method, and formulates a user association scheme, a sub-channel allocation scheme, and a power coordination scheme in turn, which can eventually reduce the same-layer interference in ultra-dense networks, thereby improving network energy efficiency.

[0005] The disadvantage of the invention is that when calculating energy efficiency in its system, the interference caused by uneven user distribution and resource usage conflicts is not considered, nor is the network capacity. Based on this, the present invention will study the user association algorithm in the ultra-dense network, consider the interference caused by uneven user distribution and resource usage conflicts, thereby reducing interference and improving network capacity. Summary of the invention

[0006] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and propose a user association method for uneven service capacity coverage of resource cells, aiming to ensure that uneven user services in the network are served evenly, thereby reducing the interference caused by resource usage conflicts and improving network capacity.

[0007] To achieve the above object, the technical solution adopted by the present invention includes the following steps:

[0008] (1) Building an ultra-dense network system:

[0009] Construct a Y resource cell C={C 1 ,C 2 ,...,C y ,...,C Y} and N non-uniformly distributed users with service needs U = {U 1 ,U 2 ,...,U n ,...,U N} ultra-dense network system, each resource cell C y Contains K base stations, Each user U n Can only be associated with one base station, each resource cell C y Each base station in Can be associated with at least one user, where M≥1, N≥1, 1≤Y≤M, U n represents the nth user;

[0010] (2) Initialization parameters:

[0011] Initialize the nth user U n With C y The kth base station in The association between When U n and When associated, otherwise UserU n By associating with the nearest base station The obtained recent user association matrix with dimension N×M is X, and the association identification value of the nth row and mth column of X is UserU n The transmission rate is R n , R n The threshold value is R 0 , the coverage area of ​​the ultra-dense network system is a square with a side length of size, and the user U n To base station The channel gain is C y The kth base station in The power is

[0012] (3) Optimization problem of establishing user associations:

[0013] Establish the total network capacity function R of users with uneven distribution of business demands in ultra-dense network system total Maximize X R total As the goal, the optimization problem with C1, C2, C3, C4 and C5 as constraints is:

[0014]

[0015]

[0016]

[0017] C1:

[0018] C2:

[0019] C3:

[0020] C4:

[0021] C5:

[0022] in, For user U n Receive from associated base station The signal-to-interference-noise ratio, C j Remove C from resource cell Y y All resource cells, C j The power in The kth base station; C j All base stations to user U n The sum of the interference signals, σ represents Gaussian white noise;

[0023] (4) Calculate the Q table in the Q learning algorithm:

[0024] (4a) Initialize the number of iterations to e, the maximum number of iterations to E, E ≥ N × M, initialize the number of explorations to t, the maximum number of explorations to T, T ≥ M, and initialize the Q value to the value in the association matrix X. The same dimension is N×M Q table, all users U in the network system are initialized as the state set, M base stations under all resource cells in the network system are initialized as the action set A, the learning rate of Q learning is initialized as α∈(0,1), the discount factor γ∈(0,1), the action exploration probability is ε∈(0,1), the constant d is half of the size, and e=0, t=1;

[0025] (4b) Randomly select a user U from the state set U t As the initial state;

[0026] (4c) Select the current state U t The action base station in the action set A

[0027] (4d) Calculate in state U t Take action base station The reward value after And judge Is it established? If so, Set to 1 to keep the action base station Let user U in user association matrix X be t The corresponding row association identifiers are all 0, and the user U in the user association matrix X is updated t and base station The corresponding association flag is 1; the state-action pair is calculated using the Q function formula The value of Then through After updating the Q table, execute step (4e), otherwise, Set to -1 and use the Q function formula to calculate the state-action pair The value of and through After updating the Q table, set t = t + 1 and execute step (4b);

[0028] (4e) Determine whether t=T is true. If so, obtain the Q table after this iteration and execute step (4f). Otherwise, take the action at the current base station. A random action base station within the range d And with the action base station A state is randomly selected from the associated state set U' as the next state Ut+1 , let U t =U t+1 , and execute step (4c);

[0029] (4f) Determine whether e=E holds true. If so, obtain the optimized Q table. Otherwise, set e=e+1 and execute step (4b);

[0030] (5) Obtaining user association results for non-uniform service capacity coverage:

[0031] UserU n Select the base station with the largest Q value in each row in the optimized Q table The non-uniform service network capacity R in the optimization problem is obtained total The final user association matrix X that reaches the maximum value.

[0032] Compared with the prior art, the present invention has the following advantages:

[0033] In view of the uneven distribution of user services in ultra-dense network systems, the present invention designs a user association algorithm based on Q learning and optimizes three aspects. First, the Q value in the Q table associated with the initial user is the value in the association matrix of the most recent association. Second, the action selection strategy is designed according to the interference characteristics in the network system. Third, the selection strategy for the next state is related to the current action. Compared with the existing technology, it can change the network service structure while meeting the user service quality requirements and the base station transmission power limitations, so that uneven user services in the network can be served evenly, thereby reducing the interference caused by resource usage conflicts and improving network capacity. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a flow chart for realizing the present invention;

[0035] Figure 2 A graph showing the variation of network capacity with the number of iterations according to the present invention;

[0036] Figure 3 It is a simulation comparison diagram of network capacity performance between the present invention and the prior art. DETAILED DESCRIPTION

[0037] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0038] Reference Figure 1 , the present invention comprises the following steps:

[0039] Step 1) Build an ultra-dense network system:

[0040] Construct a Y resource cell C={C1 ,C 2 ,...,C y ,...,C Y} and N non-uniformly distributed users with service needs U = {U 1 ,U 2 ,...,U n ,...,U N} ultra-dense network system, each resource cell C y Contains K base stations, Each user U n Can only be associated with one base station, each resource cell C y Each base station in Can be associated with at least one user, where M≥1, N≥1, 1≤Y≤M, U n represents the nth user. In this embodiment, N=600, M=6, and Y is generated by the base station to adapt to the user distribution;

[0041] Step 2) Initialize parameters:

[0042] Initialize the nth user U n With C y The kth base station in The association between When U n and When associated, otherwise UserU n By associating with the nearest base station The obtained recent user association matrix with dimension N×M is X, and the association identification value of the nth row and mth column of X is UserU n The transmission rate is R n , R n The threshold value is R 0 , the coverage area of ​​the ultra-dense network system is a square with a side length of size, and the user U n To base station The channel gain is C y The kth base station in The power is

[0043] Step 3) Establish the optimization problem of user association:

[0044] Establish the total network capacity function R of users with uneven distribution of business demands in ultra-dense network system total Maximize X R totalAs the goal, the optimization problem with C1, C2, C3, C4 and C5 as constraints is:

[0045]

[0046]

[0047]

[0048] C1:

[0049] C2:

[0050] C3:

[0051] C4:

[0052] C5:

[0053] in, For user U n Receive from associated base station The signal-to-interference-noise ratio, C j Remove C from resource cell Y y All resource cells, C j The power in The kth base station; C j All base stations to user U n The sum of the interference signals, σ represents Gaussian white noise;

[0054] Step 4) Calculate the Q table in the Q learning algorithm:

[0055] Step 4a) Initialize the number of iterations to e, the maximum number of iterations to E, E ≥ N × M, initialize the number of explorations to t, the maximum number of explorations to T, T ≥ M, and initialize the Q value to the value in the association matrix X The same dimension is N×M Q table, all users U in the network system are initialized as the state set, M base stations under all resource cells in the network system are initialized as the action set A, the learning rate of Q learning is initialized as α∈(0,1), the discount factor γ∈(0,1), the action exploration probability is ε∈(0,1), the constant d is half of the size, and e=0, t=1;

[0056] Step 4b) Randomly select a user U from the state set U t As the initial state;

[0057] Step 4c) Select the current state U tThe action base station in the action set A

[0058] Step 4c1) Calculate user U t Associated to base station The signal-to-interference-noise ratio is UserU t Associated to M base stations The signal-to-interference-noise ratio is UserU t Select M action base stations The probability of selecting M actions for:

[0059]

[0060] in, Represents user U t The sum of signal to interference and noise ratios of all base stations in the system are respectively associated;

[0061] Step 4c2) Determine whether t<N×M is true. If so, select the probability of action according to the M base stations. Select the action base station with the highest probability from the action set A Otherwise, execute step 4c3);

[0062] Step 4c3) Randomly generate a number r between 0 and 1 and determine whether r < ε is true. If so, select the action probability according to the M base stations. Select the action base station with the highest probability from the action set A Otherwise, select Satisfy Action Base Station

[0063] Step 4d) Calculate in state U t Take action base station The reward value after And judge Is it established? If so, Set to 1 to keep the action base station Let user U in user association matrix X be t The corresponding row association identifiers are all 0, and the user U in the user association matrix X is updated t and base station The corresponding association flag is 1; the state-action pair is calculated using the Q function formula The value of Then through After updating the Q table, execute step (4e), otherwise, Set to -1 and use the Q function formula to calculate the state-action pair The value of and through After updating the Q table, set t = t + 1 and execute step (4b); the reward value value The calculation formulas are:

[0064]

[0065]

[0066] Among them, R total (X t-1 ), R total (X t ) represent the user in state U t Take action base station Total network capacity before and after;

[0067] Step 4e) Determine whether t=T is true. If so, obtain the Q table after this iteration and execute step (4f). Otherwise, the current action base station is taken. A random action base station within the range d And with the action base station A state is randomly selected from the associated state set U' as the next state U t+1 , let U t =U t+1 , and execute step 4c);

[0068] Step 4f) Determine whether e=E is true, if so, obtain the optimized Q table, otherwise, set e=e+1 and execute step 4b);

[0069] Step 5) Obtain the user association result of the non-uniform service capacity coverage:

[0070] UserU n Select the base station with the largest Q value in each row in the optimized Q table The non-uniform service network capacity R in the optimization problem is obtained total The final user association matrix X that reaches the maximum value.

[0071] The following is a further description of the technical effects of the present invention in conjunction with the simulation results:

[0072] 1. Experimental conditions and contents:

[0073] [Experimental running environment] The operating system is Microsoft Windows 10, and the programming simulation language is Python. The simulation area of ​​the experiment simulation is a 1000m×1000m network scenario, with a large number of ground users unevenly distributed in the ground area and each base station randomly distributed.

[0074] Experiment 1: The curve of network capacity changing with the number of iterations of the present invention is simulated, and the results are as follows: Figure 2 As shown;

[0075] Experiment 2: Comparison simulation of the network capacity performance of the present invention and the existing method is carried out. The results are as follows: Figure 3 shown.

[0076] 2. Experimental results analysis:

[0077] Reference Figure 2 , the horizontal axis represents the training time in seconds, and the vertical axis represents the network capacity in Mbps. Figure 2 It can be seen that the user association algorithm based on Q learning is feasible and can converge in the scenario set by the present invention;

[0078] Reference Figure 3 , the simulation comparison of the network capacity performance curve of the present invention and the prior art nearest association method under different base stations, the horizontal axis represents the base station, the unit is the number, the vertical axis represents the network capacity, the unit is Mbps. Figure 3 It can be seen that when the number of base stations is 6, 10, 14, 18, and 22, the present invention and the existing method have the same growth trend as the number of network base stations and users increases in proportion to 1:100, but the network capacity obtained by the present invention is higher than that of the existing method. When the number of base stations is 18 and the number of users is 1800, the network capacity of the present invention is increased by 28.1% at most compared with the existing method, effectively suppressing the interference caused by resource conflict use to obtain a larger network capacity gain.

Claims

1. A user association method for non-uniform service capacity coverage of resource cells, characterized in that The steps include: (1) Building an ultra-dense network system: Construct a Y resource cell C={C1,C2,...,C y ,...,C Y } and N non-uniformly distributed users with service needs U={U1,U2,...,U n ,...,U N } ultra-dense network system, each resource cell C y Contains K base stations, Each user U n Can only be associated with one base station, each resource cell C y Each base station in Can be associated with at least one user, where M≥1, N≥1, 1≤Y≤M, U n represents the nth user; (2) Initialization parameters: Initialize the nth user U n With C y The kth base station in The association between When U n and When associated, otherwise UserU n By associating with the nearest base station The obtained recent user association matrix with dimension N×M is X, and the association identification value of the nth row and mth column of X is UserU n The transmission rate is R n , R n The threshold value is R0, the coverage area of ​​the ultra-dense network system is a square with a side length of size, and the user U n To base station The channel gain is C y The kth base station in The power is (3) Optimization problem of establishing user associations: Establish the total network capacity function R of users with uneven distribution of business demands in ultra-dense network system total Maximize X R total As the goal, the optimization problem with C1, C2, C3, C4 and C5 as constraints is: C1: C3: C4: C5: in, For user U n Receive from associated base station The signal-to-interference-noise ratio, C j Remove C from resource cell Y y All resource cells, C j The power in The kth base station; C j All base stations to user U n The sum of the interference signals, σ represents Gaussian white noise; (4) Calculate the Q table in the Q learning algorithm: (4a) Initialize the number of iterations to e, the maximum number of iterations to E, E ≥ N × M, initialize the number of explorations to t, the maximum number of explorations to T, T ≥ M, and initialize the Q value to the value in the association matrix X. The same dimension is N×M Q table, all users U in the network system are initialized as the state set, M base stations under all resource cells in the network system are initialized as the action set A, the learning rate of Q learning is initialized as α∈(0,1), the discount factor γ∈(0,1), the action exploration probability is ε∈(0,1), the constant d is half of the size, and e=0, t=1; (4b) Randomly select a user U from the state set U t As the initial state; (4c) Select the current state U t The action base station in the action set A (4d) Calculate in state U t Take action base station The reward value after And judge Is it established? If so, Set to 1 to keep the action base station Let user U in user association matrix X be t The corresponding row association identifiers are all 0, and the user U in the user association matrix X is updated t and base station The corresponding association flag is 1; the state-action pair is calculated using the Q function formula The value of Then through After updating the Q table, execute step (4e), otherwise, Set to -1 and use the Q function formula to calculate the state-action pair The value of and through After updating the Q table, set t = t + 1 and execute step (4b); (4e) Determine whether t=T is true. If so, obtain the Q table after this iteration and execute step (4f). Otherwise, take the action at the current base station. A random action base station within the range d And with the action base station A state is randomly selected from the associated state set U' as the next state U t+1 , let U t =U t+1 , and execute step (4c); (4f) Determine whether e=E holds true. If so, obtain the optimized Q table. Otherwise, set e=e+1 and execute step (4b); (5) Obtaining user association results for non-uniform service capacity coverage: UserU n Select the base station with the largest Q value in each row of the optimized Q table The non-uniform service network capacity R in the optimization problem is obtained total The final user association matrix X that reaches the maximum value.

2. A user association method for non-uniform service capacity coverage of resource cells according to claim 1, characterized in that: The current state U selected in step (4c) t The action base station in the action set A The implementation steps are: (4c1) Calculate user U t Associated to base station The signal-to-interference-noise ratio is UserU t Associated to M base stations The signal-to-interference-noise ratio is UserU t Select M action base stations The probability of selecting M actions for: in, Represents user U t The sum of signal to interference and noise ratios of all base stations in the system are respectively associated; (4c2) Determine whether t<N×M is true. If so, select the probability of action based on the M base stations. Select the action base station with the highest probability from the action set A Otherwise, execute step (4c3); (4c3) Randomly generate a number r between 0 and 1 and determine whether r < ε. If so, select the probability of action according to the M base stations. Select the action base station with the highest probability from the action set A Otherwise, select Satisfy Action Base Station 3. A user association method for non-uniform service capacity coverage of resource cells according to claim 1, characterized in that: The reward value described in step (4d) value The calculation formulas are: Among them, R total (X t-1 ), R total (X t ) represent the user in state U t Take action base station Total network capacity before and after.

Images