CN105873127A - Heuristic user connection load balancing method based on random decision - Google Patents

Abstract

The invention discloses a discrete heuristic user connection load balancing method based on random decision in an ultra-dense heterogeneous network. A user has a connection probability vector for a selectable candidate base station, a service base station is selected randomly according to the connection probability vector on the basis of the current network environment, the connection feedback of the user is obtained from the selected base station, the connection probability vector of the user is updated till a certain member in the connection probability vector approaches to 1, and the service base station needing to be connected is determined. By adopting the method, newly-arrived users can select the optimal base station to be connected according to existing network load conditions. The connection is determined according to the local information, and signaling overheads and computation complexity of the overall network are reduced. Load balance and user rate fairness in the ultra-dense heterogeneous network can be improved, and meanwhile the resource utilization rate of the ultra-dense heterogeneous network can be improved.

Description

The load-balancing method connected based on the random heuristic user determined

Technical field

The present invention relates to the load connected in a kind of super-intensive heterogeneous network based on the random heuristic user determined Equalization methods, belongs to the networking technology area in radio communication.

Background technology

In the heterogeneous network of 5G (the 5th third-generation mobile communication technology) system, disposing highdensity small station becomes normal The strategy seen, thus form super-intensive heterogeneous network.

In super-intensive heterogeneous network, the base station of dissimilar (such as can make a distinction according to launching power) Constituting different layers, same type of base station is in same layer.The base station such as launching power little is referred to as small station, institute Having composition small station, small station layer, launch high-power base station and be referred to as macro station, all macro stations constitute macro station layer.Due to not Different with the transmitting power between the base station of layer, the reference signal from difference station causing user to receive receives merit Rate (RSRP, Reference Signal Received Power) has the biggest gap, if user uses tradition Maximum RSRP connection mechanism, greatly user can be caused to be linked into transmitting power relatively big, signal is preferable Base station on, thus cause the congested of these base stations.And some other transmitting power is less, but there is abundant money The base station in source is in idle condition.So so that offered load skewness weighs so that whole Internet resources are not Can adequately and reasonably be used, cause resource utilization ratio low, overall performance of network declines.

In wireless network, the load of the number of users reflection base station that can connect with base station, so seeking in network each The load balancing of base station can be regarded as and seek suitable user's method of attachment.In traditional cellular network, load is all Weighing apparatus mechanism is exactly by the resource reasonable distribution in heavy duty and light load area, and the user in heavy duty network is turned Move on to around load the lightest network, reach the state of Load Balanced distribution.So make whole system Resource utilization is effectively promoted.But in super-intensive heterogeneous network, due to the transmitting merit between big small station Rate differs greatly, and is unloaded to the user in small station by by the strong jamming from macro station from macro station, causes the letter of user Number relatively low with interference plus noise ratio (SINR, Signal to Interference plus Noise Ratio).Cause using The throughput degradation that family obtains from small station, it is impossible to make full use of small station resource.So in super-intensive heterogeneous network Load balancing through frequently with suitable interference management techniques, reduce from macro station be unloaded to that the user in small station is subject to dry Disturb, promote user to be unloaded to small station from the macro station of high capacity, further up to the load between macro station and small station Equilibrium, thus promote network performance further.So introduce in super-intensive network almost blank subframe (ABS, Almost Blank Subframe), namely macro station reserves according to certain ABS ratio in time domain Almost blank subframe, during these subframes, small station user will be declined by the interference of macro station, improve small station money The utilization rate in source.

Summary of the invention

Goal of the invention: for problems of the prior art with not enough, the present invention provides a kind of super-intensive isomery The load-balancing method connected based on user in network, it is proposed that a kind of discrete based on random determine heuristic User's method of attachment.Use the method in the present invention, it is possible to increase in super-intensive heterogeneous network load equilibrium and The fairness of user rate, and the utilization rate of Internet resources.

The present invention, for the purpose of the logarithm speed summation maximizing all users, connects optimization to user, thus real Existing load balancing.Owing to this problem complexity of direct solution is too high, it is to be appreciated that the channel gain information of the overall situation and The connection strategy of all users, this can cause the signal resource exorbitant expenditure of whole system actual to implement. Therefore, use a kind of discrete heuristic to allow each user dynamically and the base station of oneself optimum is selected independently. All of decision all completes in user side, and user has one for its candidate base station that can connect and connects probability Vector, according to current network environment, is randomly chosen serving BS, from choose according to connecting probability vector Base station obtains the connection feedback of oneself, updates the connection probability vector of oneself, until connecting certain in probability vector One member, close to 1, finally determines serving BS to be connected.

Technical scheme: a kind of load-balancing method connected based on the random heuristic user determined, to maximize In super-intensive heterogeneous network, the logarithm speed summation of all users is the load balancing mould that target sets up that user connects Type.This model promotes the fairness of network in the following areas, first, can promote with the speed promoting user for target Enable user to be connected to relative free to be assigned on the small station of more multiple resource, thus improve and load between big small station Harmony, secondly, logarithmic function is the function that a gain is successively decreased, and can improve the fairness of user rate. The user determined by this model is connected can also promote overall network throughput performance.

Assuming in super-intensive heterogeneous network, all of major station all uses ABS ratio 0≤α≤1 of synchronization, Each major station occupies the time resource of 1-α, during small station both can be operated in ABS, it is also possible to be operated in non- During ABS, so each small station to be regarded as two logical base station, an ABS small station, occupy the time of α Resource；One non-ABS small station, occupies the time resource of 1-α.User can be alternatively coupled to major station, ABS Small station or non-ABS small station.

All of major station set M represents, size is N_m, all of ABS small station is by gathering P^ABSRepresent, Size is N_p, all of non-ABS small station is by gathering P^nABSRepresenting, size is N_p, wherein N_pMore than N_m, According to the difference of network, N_pCan be N_mSeveral times, Ji Shibei, even hundreds of times.All of user by Set U represents, size is N_u。

The SINR value that user i receives from major station or non-ABS small station j:

${\mathrm{SINR}}_{ij}=\frac{P_j{h}_{ij}}{{\mathrm{\Σ}}_{n\∈(M\∪P^{nABS})/j}{P}_n{h}_{in}+{\mathrm{\σ}}^2},\∀i\∈U,j\∈M\∪P^{nABS}---\[1\]$

The SINR value that user i receives from ABS small station j:

${\mathrm{SINR}}_{ij}=\frac{P_j{k}_{ij}}{{\mathrm{\Σ}}_{n\∈P^{ABS}/j}{P}_n{h}_{in}+{\mathrm{\σ}}^2},\∀i\∈U,j\∈P^{ABS}---\[2\]$

Wherein, P_jIt is defined as being numbered the transmitting power of the base station of j, h_ijRepresent is from base station j to user i Channel gain, comprise path loss, shadow fading and antenna gain, σ²Represent noise power.

Channel capacity c that user i obtains from base station j_ijIt is calculated by following formula:

c_ij=Wlog (1+SINR_ij) [3]

Wherein W represents the bandwidth of base station.All of base station uses identical bandwidth.

The optimization problem that the present invention proposes is shown below.

$\underset{\left\{X\right\}}{max}\underset{i,j}{\Σ}{x}_{ij}{\mathrm{logR}}_{ij}---\[4a\]$

Constraints:

$\underset{j\∈B}{\Σ}{x}_{ij}=1,\∀i\∈U---\[4b\]$

$\underset{i}{\Σ}{x}_{ij}=k_j,\∀j\∈B---\[4c\]$

$\underset{j\∈B}{\Σ}{k}_j=N_u---\[4d\]$

$x_{ij}\∈\{0,1\},\∀i\∈U,\∀j\∈B---\[4e\]$

Wherein, x_ijIt is a binary variable, when user i is connected on the j of base station when, x_ij=1, be otherwise 0, k_jBeing defined as on jth base station the number of users connected, X represents all x_ijSet.All of major station, ABS small station, non-ABS small station composition collection of base stations, represent with B.Constraints [4b] is each in order to limit User is connected solely on a base station, and constraints [4d] is in order to guarantee that all of user is connected.R_ijTable Showing the speed that user i obtains from base station j, it is calculated by following formula:

$R_{ij}=\left\{\begin{array}{c}\frac{(1-\mathrm{\α})W}{k_j}log(1+{\mathrm{SINR}}_{ij}),\∀i\∈U,j\∈M\∪P^{nABS}\\ \frac{\mathrm{\α}W}{k_j}log(1+{\mathrm{SINR}}_{ij}),\∀i\∈U,j\∈P^{ABS}\end{array}\right.---\[5\]$

Owing to this problem complexity of direct solution is too high, therefore, a discrete heuristic is used to allow each User dynamically and is selected independently the base station of oneself optimum.All of decision all completes in user side, user for Its candidate base station that can connect has one and connects probability vector, according to current network environment, according to connection Probability vector is randomly chosen serving BS, obtains the connection feedback of oneself, update oneself from the base station chosen Connection probability vector, until connecting some member in probability vector, close to 1, finally to determine clothes to be connected Business base station.Detailed process is as follows:

1) base station selected strategy

At user side, each user is randomly chosen a base station according to randomized policy.Be fixedly selected one Base station is different, and user i is according to a probability vectorSelect base station with having probability, wherein q_ijRepresent that user i selects the probability of base station j.|a_i| represent the number of the candidate base station of user i.If base station Can provide better performance, then the connection probability of base station will be higher, and vice versa.In this way, often Individual user can each be connected automatically on the BS of optimum.User i selects the probability of base station j to be given by:

$q_{ij}=\frac{\exp \left\{{\mathrm{\γr}}_{ij}\right\}}{\underset{k\∈a_i}{\Σ}\exp \left\{{\mathrm{\γr}}_{ik}\right\}}---\[6\]$

WhereinAnd γ is a positive parameter, commonly referred to as temperature, the value of γ is close to 0 Time, can guarantee that the last solution of gained converges on optimal solution, but the value of γ is the biggest simultaneously, rate of convergence is the fastest, For ensureing convergence rate and convergence effect, value generally uses 0.1～0.01 such scope.r_ijRepresent is to use The connection feedback that family i obtains from base station j, defines according to utility function, and namely user obtains from base station instantly Logarithm handling capacity.Assume that user i selects base station j as serving BS, be connected to the number of users on the j of base station For k_j.Then the feedback function of user i is given by:

$\begin{array}{c}{r}_{ij}=R_{ij}(\mathrm{\α},k_j)\\ =\left\{\begin{array}{c}\log \left(\frac{(1-\mathrm{\α})W}{k_j}\log \right(1+{\mathrm{SINR}}_{ij}\left)\right),\∀i\∈U,j\∈M\∪P^{nABS}\\ \log \left(\frac{\mathrm{\α}W}{k_j}\log \right(1+{\mathrm{SINR}}_{ij}\left)\right),\∀i\∈U,j\∈P^{ABS}\end{array}\right.\\ =\left\{\begin{array}{c}\log \left(\right(1-\mathrm{\α}\left)W\log \right(1+{\mathrm{SINR}}_{ij}\left)\right)-\log \left(k_j\right),\∀i\∈U,j\∈M\∪P^{nABS}\\ \log \left(\mathrm{\α}W\log \right(1+{\mathrm{SINR}}_{ij}\left)\right)-\log \left(k_j\right),\∀i\∈U,j\∈P^{ABS}\end{array}\right.\end{array}---\[7\]$

2) connection feedback is calculated

User calculates the corresponding value of feedback in base station choosing connection according to [7] formula.All users in peak optimizating network Throughput performance, coordination each other is necessary.So feedback function has a base station Value Factors log(k_j)_,It reflects the loading condition of base station_。If k_jRise_,Illustrate to be connected to base station_jNumber of users Increase, then the feedback that user i obtains from base station j declines.Bear if it is desired to increase base station in feedback function further Carry consciousness, corresponding Value Factors p can be further added by_ij, it is proportional to k_j。p_ijBe equivalent to the total of base station j connection User is k_jThe Value Factors of Shi Tixian loading condition.Feedback function becomesWhereinRepresent into One step increases the feedback function after Value Factors.It is congested and certain that this Value Factors can reflect between user Degree alleviates it.

3) base station is estimated to update

In state t, each user only obtains the feedback of the base station instantly selected, and provides next round and connects probability more The value of feedback used in new:

$r_{ij}^{t+1}=\left\{\begin{array}{c}(1-\mathrm{\ϵ})r_{ij}^t+\mathrm{\ϵ}{\stackrel{\‾}{r}}_{ij}^t,a_i=j\\ r_{ij}^t,otherwise\end{array}\right.---\[8\]$

What ε represented is a weight factor, and meets 0≤ε≤1, and reflection is that user is for connecting the feedback of base station R in value renewal_ijWithShared weight ratio, and ε is the biggest,Shared weight is the biggest.Anti-according to newly obtain Feedback value updates the connection probability vector of user.

4) final serving BS is determined

Circulation performs first three step, until a certain member variable is close to 1 in the connection probability vector of user Time, then the base station that this variable is corresponding is the serving BS that user is final.Convergence rate to be increased, then can be even When connecing the scope that probability reaches more than 0.9, determine that this base station is serving BS, accuracy will be increased further, Then can choose the value of 0.95～more than 0.99 as determining border.

Beneficial effect: the present invention is with the logarithm speed summation of all users in maximization super-intensive heterogeneous network as mesh Mark sets up the Load Balancing Model that user connects.With discrete carrying out based on the random heuristic determined What user connected solves, and load effectively can be unloaded to the small station of relative free by the method from congested macro station On.All of decision all completes in user side, and user terminal has only to know that the information of local just can be according to certainly Oneself connection strategy is connected on instantly optimum base station, without know the SINR of other users and he Connection strategy.Model have employed ABS technology, allow the user being connected on small station during ABS avoid From the strong jamming of macro station, promote that user is unloaded on small station, ensures the performance of small station user simultaneously further. The user that the model be given according to the present invention calculates connects, and promotes network performance while reaching load balancing.

Load-balancing method based on the present invention has the advantage that

1. user can effectively be unloaded to the small station of relative free from congested major station, between each base station Load have higher fairness.

2. can effectively promote the speed of edge customer, the fairness of user performance improves

The most all of decision all completes in user side, and user terminal has only to know that the information of local just can root It is connected on instantly optimum base station according to the connection strategy of oneself.

Detailed description of the invention

Below in conjunction with specific embodiment, it is further elucidated with the present invention, it should be understood that these embodiments are merely to illustrate this Invention rather than restriction the scope of the present invention, after having read the present invention, those skilled in the art are to this The amendment of the bright various equivalent form of values all falls within the application claims limited range.

The load-balancing method connected based on the random heuristic user determined, to maximize super-intensive heterogeneous network In the logarithm speed summation of all users be the Load Balancing Model that target sets up that user connects.This model is following Aspect promotes the fairness of network, first, user can be promoted with the speed promoting user to be connected to relatively for target Free time can be assigned on the small station of more multiple resource, thus improves the harmony of load between big small station, secondly, Logarithmic function is the function that a gain is successively decreased, and can improve the fairness of user rate.Determined by this model User connects can also promote overall network throughput performance.

Assuming in super-intensive heterogeneous network, all of major station all uses ABS ratio 0≤α≤1 of synchronization, Each major station occupies the time resource of 1-α, during small station both can be operated in ABS, it is also possible to be operated in non- During ABS, so each small station to be regarded as two logical base station, an ABS small station, occupy the time of α Resource；One non-ABS small station, occupies the time resource of 1-α.User can be alternatively coupled to major station, ABS Small station or non-ABS small station.

All of major station set M represents, size is N_m, all of ABS small station is by gathering P^ABSRepresent, Size is N_p, all of non-ABS small station is by gathering P^nABSRepresenting, size is N_p, wherein N_pMore than N_m, According to the difference of network, N_pCan be N_mSeveral times, Ji Shibei, even hundreds of times.All of user by Set U represents, size is N_u。

The SINR value that user i receives from major station or non-ABS small station j:

${\mathrm{SINR}}_{ij}=\frac{P_j{h}_{ij}}{{\mathrm{\Σ}}_{n\∈(M\∪P^{nABS})/j}{P}_n{h}_{in}+{\mathrm{\σ}}^2},\∀i\∈U,j\∈M\∪P^{nABS}---\[1\]$

The SINR value that user i receives from ABS small station j:

${\mathrm{SINR}}_{ij}=\frac{P_j{k}_{ij}}{{\mathrm{\Σ}}_{n\∈P^{ABS}/j}{P}_n{h}_{in}+{\mathrm{\σ}}^2},\∀i\∈U,j\∈P^{ABS}---\[2\]$

Wherein, P_jIt is defined as being numbered the transmitting power of the base station of j, h_ijRepresent is from base station j to user i Channel gain, comprise path loss, shadow fading and antenna gain, σ²Represent noise power.

Channel capacity c that user i obtains from base station j_ijIt is calculated by following formula:

c_ij=Wlog (1+SINR_ij) [3]

Wherein W represents the bandwidth of base station.All of base station uses identical bandwidth.

The optimization problem that the present invention proposes is shown below.

$\underset{\left\{X\right\}}{max}\underset{i,j}{\Σ}{x}_{ij}{\mathrm{logR}}_{ij}---\[4a\]$

Constraints:

$\underset{j\∈B}{\Σ}{x}_{ij}=1,\∀i\∈U---\[4b\]$

$\underset{i}{\Σ}{x}_{ij}=k_j,\∀j\∈B---\[4c\]$

$\underset{j\∈B}{\Σ}{k}_j=N_u---\[4d\]$

$x_{ij}\∈\{0,1\},\∀i\∈U,\∀j\∈B---\[4e\]$

Wherein, x_ijIt is a binary variable, when user i is connected on the j of base station when, x_ij=1, be otherwise 0, k_jBeing defined as on jth base station the number of users connected, X represents all x_ijSet.All of major station, ABS small station, non-ABS small station composition collection of base stations, represent with B.Constraints [4b] is each in order to limit User is connected solely on a base station, and constraints [4d] is in order to guarantee that all of user is connected.R_ijTable Showing the speed that user i obtains from base station j, it is calculated by following formula:

$R_{ij}=\left\{\begin{array}{c}\frac{(1-\mathrm{\α})W}{k_j}log(1+{\mathrm{SINR}}_{ij}),\∀i\∈U,j\∈M\∪P^{nABS}\\ \frac{\mathrm{\α}W}{k_j}log(1+{\mathrm{SINR}}_{ij}),\∀i\∈U,j\∈P^{ABS}\end{array}\right.---\[5\]$

Owing to this problem complexity of direct solution is too high, therefore, a discrete heuristic is used to allow each User dynamically and is selected independently the base station of oneself optimum.All of decision all completes in user side, user for Its candidate base station that can connect has one and connects probability vector, according to current network environment, according to connection Probability vector is randomly chosen serving BS, obtains the connection feedback of oneself, update oneself from the base station chosen Connection probability vector, until connecting some member in probability vector, close to 1, finally to determine clothes to be connected Business base station.Detailed process is as follows:

1) base station selected strategy

At user side, each user is randomly chosen a base station according to randomized policy.Be fixedly selected one Base station is different, and user i is according to a probability vectorSelect base station with having probability, wherein q_ijRepresent that user i selects the probability of base station j.|a_i| represent the number of the candidate base station of user i.If base station Can provide better performance, then the connection probability of base station will be higher, and vice versa.In this way, often Individual user can each be connected automatically on the BS of optimum.User i selects the probability of base station j to be given by:

$q_{ij}=\frac{\exp \left\{{\mathrm{\γr}}_{ij}\right\}}{\underset{k\∈a_i}{\Σ}\exp \left\{{\mathrm{\γr}}_{ik}\right\}}---\[6\]$

WhereinAnd γ is a positive parameter, commonly referred to as temperature, the value of γ is close to 0 Time, can guarantee that the last solution of gained converges on optimal solution, but the value of γ is the biggest simultaneously, rate of convergence is the fastest, For ensureing convergence rate and convergence effect, value generally uses 0.1～0.01 such scope.r_ijRepresent is to use The connection feedback that family i obtains from base station j, defines according to utility function, and namely user obtains from base station instantly Logarithm handling capacity.Assume that user i selects base station j as serving BS, be connected to the number of users on the j of base station For k_j.Then the feedback function of user i is given by:

$\begin{array}{c}{r}_{ij}=R_{ij}(\mathrm{\α},k_j)\\ =\left\{\begin{array}{c}\log \left(\frac{(1-\mathrm{\α})W}{k_j}\log \right(1+{\mathrm{SINR}}_{ij}\left)\right),\∀i\∈U,j\∈M\∪P^{nABS}\\ \log \left(\frac{\mathrm{\α}W}{k_j}\log \right(1+{\mathrm{SINR}}_{ij}\left)\right),\∀i\∈U,j\∈P^{ABS}\end{array}\right.\\ =\left\{\begin{array}{c}\log \left(\right(1-\mathrm{\α}\left)W\log \right(1+{\mathrm{SINR}}_{ij}\left)\right)-\log \left(k_j\right),\∀i\∈U,j\∈M\∪P^{nABS}\\ \log \left(\mathrm{\α}W\log \right(1+{\mathrm{SINR}}_{ij}\left)\right)-\log \left(k_j\right),\∀i\∈U,j\∈P^{ABS}\end{array}\right.\end{array}---\[7\]$

2) connection feedback is calculated

User calculates the corresponding value of feedback in base station choosing connection according to [7] formula.All users in peak optimizating network Throughput performance, coordination each other is necessary.So feedback function has a base station Value Factors log(k_j)_,It reflects the loading condition of base station_。If k_jRise_,Illustrate to be connected to the number of users of base station j Increase, then the feedback that user i obtains from base station j declines.Bear if it is desired to increase base station in feedback function further Carry consciousness, corresponding Value Factors p can be further added by_ij, it is proportional to k_j。p_ijBe equivalent to the total of base station j connection User is k_jThe Value Factors of Shi Tixian loading condition.Feedback function becomesWhereinRepresent into One step increases the feedback function after Value Factors.It is congested and certain that this Value Factors can reflect between user Degree alleviates it.

3) base station is estimated to update

In state t, each user only obtains the feedback of the base station instantly selected, and provides next round and connects probability more The value of feedback used in new:

$r_{ij}^{t+1}=\left\{\begin{array}{c}(1-\mathrm{\ϵ})r_{ij}^t+\mathrm{\ϵ}{\stackrel{\‾}{r}}_{ij}^t,a_i=j\\ r_{ij}^t,otherwise\end{array}\right.---\[8\]$

What ε represented is a weight factor, and meets 0≤ε≤1, and reflection is that user is for connecting the feedback of base station R in value renewal_ijWithShared weight ratio, and ε is the biggest,Shared weight is the biggest.Anti-according to newly obtain Feedback value updates the connection probability vector of user.

4) final serving BS is determined

Circulation performs first three step, until a certain member variable is close to 1 in the connection probability vector of user Time, then the base station that this variable is corresponding is the serving BS that user is final.Convergence rate to be increased, then can be even When connecing the scope that probability reaches more than 0.9, determine that this base station is serving BS, accuracy will be increased further, Then can choose the value of 0.95～more than 0.99 as determining border.

Embodiment

1. initialize each user connection probability for its candidate base station, if the candidate base station number of user i is a_i, then can connect probability vector and be initialized as

2. each user connects, according to it, the selection serving BS that probability vector is random in candidate base station.

3. the value of feedback that each user is given according to the serving BS selected by formula [7] calculating, calculates further according to formula [8] Next round connects value of feedback to be used during probability vector updates.

4. each user updates connection probability vector according to formula [6] value of feedback being newly calculated in previous step.

5. repeated execution of steps 2,3,4, until certain element is close to 1 in the connection probability vector of user, than If value is more than 0.95, then the base station of its correspondence is defined as the serving BS that this user is final.

The connection probability vector of the most all users has close to 1 element time, the most all of user connects really Fixed.

Claims (3)

1. the load-balancing method connected based on the random heuristic user determined, it is characterised in that: use Family has one for its candidate base station that can connect and connects probability vector, according to current network environment, root It is randomly chosen serving BS according to connecting probability vector, from the base station chosen, obtains the connection feedback of oneself, more The newly connection probability vector of oneself, close to 1 until connecting some member in probability vector, finally determines and to connect The serving BS connect.

2. the load-balancing method connected based on the random heuristic user determined as claimed in claim 1, It is characterized in that:

Assuming in super-intensive heterogeneous network, all of major station all uses ABS ratio 0≤α≤1 of synchronization, Each major station occupies the time resource of 1-α, during small station both can be operated in ABS, it is also possible to be operated in non- During ABS, so each small station to be regarded as two logical base station, an ABS small station, occupy the time of α Resource；One non-ABS small station, occupies the time resource of 1-α；User can be alternatively coupled to major station, ABS Small station or non-ABS small station；

All of major station set M represents, size is N_m, all of ABS small station is by gathering P^ABSRepresent, Size is N_p, all of non-ABS small station is by gathering P^nABSRepresenting, size is N_p, wherein N_pMore than N_m, All of user is represented by gathering U, and size is N_u；

The SINR value that user i receives from major station or non-ABS small station j:

${\mathrm{SINR}}_{ij}=\frac{P_j{h}_{ij}}{{\mathrm{\Σ}}_{n\∈(M\∪P^{nABS})/j}{P}_n{h}_{in}+{\mathrm{\σ}}^2},\∀i\∈U,j\∈M\∪P^{nABS}---\[1\]$

The SINR value that user i receives from ABS small station j:

${\mathrm{SINR}}_{ij}=\frac{P_j{h}_{ij}}{{\mathrm{\Σ}}_{n\∈P^{ABS}/j}{P}_n{h}_{in}+{\mathrm{\σ}}^2},\∀i\∈U,j\∈P^{ABS}---\[2\]$

Wherein, P_jIt is defined as being numbered the transmitting power of the base station of j, h_ijRepresent is from base station j to user i Channel gain, comprise path loss, shadow fading and antenna gain, σ²Represent noise power；

Channel capacity c that user i obtains from base station j_ijIt is calculated by following formula:

c_ij=Wlog (1+SINR_ij) [3]

Wherein W represents the bandwidth of base station；All of base station uses identical bandwidth；

The optimization problem proposed is shown below:

$\underset{\left\{X\right\}}{max}\underset{i,j}{\Σ}{x}_{ij}{\mathrm{logR}}_{ij}---\[4a\]$

Constraints:

$\underset{j\∈B}{\Σ}{x}_{ij}=1,\∀i\∈U---\[4b\]$

$\underset{i}{\Σ}{x}_{ij}=k_j,\∀j\∈B---\[4c\]$

$\underset{j\∈B}{\Σ}{k}_j=N_u---\[4d\]$

$x_{ij}\∈\{0,1\},\∀i\∈U,\∀j\∈B---\[4e\]$

Wherein, x_ijIt is a binary variable, when user i is connected on the j of base station when, x_ij=1, be otherwise 0, k_jBeing defined as on jth base station the number of users connected, X represents all x_ijSet；All of major station, ABS small station, non-ABS small station composition collection of base stations, represent with B；Constraints [4b] is each in order to limit User is connected solely on a base station, and constraints [4d] is in order to guarantee that all of user is connected；R_ijTable Showing the speed that user i obtains from base station j, it is calculated by following formula:

$R_{ij}=\left\{\begin{array}{c}\frac{(1-\mathrm{\α})W}{k_j}log(1+{\mathrm{SINR}}_{ij}),\∀i\∈U,j\∈M\∪P^{nABS}\\ \frac{\mathrm{\α}W}{k_j}log(1+{\mathrm{SINR}}_{ij}),\∀i\∈U,j\∈P^{ABS}\end{array}\right.---\[5\]$

Use a discrete heuristic to allow each user dynamically and the base station of oneself optimum is selected independently. All of decision all completes in user side, and user has one for its candidate base station that can connect and connects probability Vector, according to current network environment, is randomly chosen serving BS, from choose according to connecting probability vector Base station obtains the connection feedback of oneself, updates the connection probability vector of oneself, until connecting certain in probability vector One member, close to 1, finally determines serving BS to be connected.

3. the load-balancing method connected based on the random heuristic user determined as claimed in claim 1, It is characterized in that, determine that the detailed process that user connects is as follows:

1) base station selected strategy

At user side, each user is randomly chosen a base station according to randomized policy.Be fixedly selected one Base station is different, and user i is according to a probability vectorSelect base station with having probability, wherein q_ijRepresent that user i selects the probability of base station j.|a_i| represent the number of the candidate base station of user i；If base station Can provide better performance, then the connection probability of base station will be higher, and vice versa.In this way, often Individual user can each be connected automatically on the BS of optimum.User i selects the probability of base station j to be given by:

$q_{ij}=\frac{\exp \left\{{\mathrm{\γr}}_{ij}\right\}}{\underset{k\∈a_i}{\Σ}\exp \left\{{\mathrm{\γr}}_{ik}\right\}}---\[6\]$

WhereinAnd γ is a positive parameter, commonly referred to as temperature, the value of γ is close to 0 Time, can guarantee that the last solution of gained converges on optimal solution, but the value of γ is the biggest simultaneously, rate of convergence is the fastest, r_ijRepresent is that the connection that user i obtains from base station j is fed back, and defines according to utility function, and namely user works as Under the logarithm handling capacity that obtains from base station；Assume that user i selects base station j as serving BS, be connected to base station Number of users on j is k_j；Then the feedback function of user i is given by:

$\begin{array}{c}{r}_{ij}=R_{ij}(\mathrm{\α},k_j)\\ =\left\{\begin{array}{c}\log \left(\frac{(1-\mathrm{\α})W}{k_j}\log \right(1+{\mathrm{SINR}}_{ij}\left)\right),\∀i\∈U,j\∈M\∪P^{nABS}\\ \log \left(\frac{\mathrm{\α}W}{k_j}\log \right(1+{\mathrm{SINR}}_{ij}\left)\right),\∀i\∈U,j\∈P^{ABS}\end{array}\right.\\ =\left\{\begin{array}{c}\log \left(\right(1-\mathrm{\α}\left)W\log \right(1+{\mathrm{SINR}}_{ij}\left)\right)-\log \left(k_j\right),\∀i\∈U,j\∈M\∪P^{nABS}\\ \log \left(\mathrm{\α}W\log \right(1+{\mathrm{SINR}}_{ij}\left)\right)-\log \left(k_j\right),\∀i\∈U,j\∈P^{ABS}\end{array}\right.\end{array}---\[7\]$

2) connection feedback is calculated

User calculates the corresponding value of feedback in base station choosing connection according to [7] formula.All users in peak optimizating network Throughput performance, coordination each other is necessary；So feedback function has a base station Value Factors log(k_j), it reflects the loading condition of base station；If k_jRise, illustrate to be connected to the number of users of base station j Increase, then the feedback that user i obtains from base station j declines；Bear if it is desired to increase base station in feedback function further Carry consciousness, corresponding Value Factors p can be further added by_ij, it is proportional to k_j；p_ijBe equivalent to the total of base station j connection User is k_jThe Value Factors of Shi Tixian loading condition；Feedback function becomesWhereinRepresent into One step increases the feedback function after Value Factors；It is congested and certain that this Value Factors can reflect between user Degree alleviates it；

3) base station is estimated to update

In state t, each user only obtains the feedback of the base station instantly selected, and provides next round and connects probability more The value of feedback used in new:

$r_{ij}^{t+1}=\left\{\begin{array}{c}(1-\mathrm{\ϵ})r_{ij}^t+\mathrm{\ϵ}{\tilde{r}}_{ij}^t,a_i=j\\ r_{ij}^t,otherwise\end{array}\right.---\[8\]$

What ε represented is a weight factor, and meets 0≤ε≤1, and reflection is that user is for connecting the feedback of base station R in value renewal_ijWithShared weight ratio, and ε is the biggest,Shared weight is the biggest；Anti-according to newly obtain Feedback value updates the connection probability vector of user；

4) final serving BS is determined

Circulation performs first three step, until a certain member variable is close to 1 in the connection probability vector of user Time, then the base station that this variable is corresponding is the serving BS that user is final.