CN105188144A - Multi-user multiple input multiple output (MU-MIMO) based stream distribution proportional fairness scheduling method - Google Patents

Abstract

The invention discloses a multi-user multiple input multiple output (MU-MIMO) based stream distribution proportional fairness scheduling method, mainly solving the problem of proportional fairness scheduling ignoring user service quality in the prior art. The method comprises a first step of setting corresponding parameters according to a practical scheduling problem; a second step of computing a user weight in a utility function according to the fact that an emission probability meets mean values in normal distribution; a third step of generating sample values of the emission probability and computing a utility function value of a sample according to the user weight; a fourth step of screening out sample values with good performance so as to update the mean values in normal distribution; a fifth step of recording the change condition of the mean values and switching variances of different phases according to a corresponding rule; and a sixth step of ending an iterative process according to a termination condition, so as to obtain a scheduling result meeting proportional fairness. A satisfactory scheduling result can be obtained whether under the condition of sufficient available wireless resource or under the condition of insufficient available wireless resource, the computation complexity is low, the iterations are few, the scheduling efficiency of a base station is improved, and the method is applicable to a single base station system adopting an MIMO technology.

Description

Based on the flow assignment proportional fair dispatching method of MU-MIMO

Technical field

The invention belongs to wireless communication technology field, relate to a kind of flow assignment equitable proportion method based on MU-MIMO, can be used for the single base station system adopting MIMO technology.

Background technology

Along with current to the active demand of high data rate system, 802.11ac gets the attention as the new generation of wireless local area network technology standard after 802.11n.802.11ac is made up of several important technology, comprising being referred to as multi-user's multiple-input and multiple-output MU-MIMO technology.MU-MIMO refers to that multiple user communicates with base station in same running time-frequency resource, and base station end and user side are multiple antennas here.MU-MIMO not only achieves space division multiplexing in the communication of unique user, and achieves space division multiple access between multiple user, so MU-MIMO can not only increase the speed of unique user, what is more important improves throughput and the availability of frequency spectrum of whole system.

The relevant MU-MIMO resource allocation algorithm of great majority mainly concentrates on the quality of service requirement of the seldom concerned with user to the efficiency utilization of Radio Resource, and the utilance of simple pursuit Radio Resource can cause the unfairness of QoS of customer.When between user, channel gain differs greatly, the good user of channel condition can obtain excessive Radio Resource, and therefore the poor user of channel condition may be not being met by service request.

For above problem, existing distribution method (1:V.VallsandD.J.Leith, " ProportionalFairMU-MIMOin802.11WLANs, " IEEEWirelessCommunicationsLetters, vol.3, no.2, pp.221-224, 2014.2:A.CheccoandD.JLeiith, " ProportionalFairnessin802.11WirelessLANs, " IEEECommunicationLetters, vol.15, no.8, pp.807-809, 2011) convex optimization method is utilized to obtain proportional fair algorithm about the MU-MIMO flow assignment in 802.11ac WLAN (wireless local area network).Method (3:D.J.Leith, Q.Cao, V.G.Subramanian, " Max-MinFairnessin802.11MeshNetworks, " IEEE/ACMTransactiononNetwoking, vol.20, no.3, pp.756-769,2012.) by another kind of fairness policy,---minimax is fair---is applied in Mesh network, utilizes logarithmic convexity to realize the minimax justice of Mesh network.

Above method only have studied the situation of resource mean allocation, but have ignored the impact of quality of service requirement on Resourse Distribute of each user, especially when available resources are not enough, these mean allocation are not best allocative decision, this is because the user that the quality of service requirement when inadequate resource is high is assigned with relatively less resource, and the low user of quality of service requirement is assigned with relatively many resources, obviously unfair on the contrary.

Summary of the invention

The object of the invention is to the deficiency for said method, propose a kind of resource allocation methods based on MU-MIMO, under the prerequisite meeting QoS of customer, to obtain in system the good compromise that equitable proportion and throughput of system between user are maximum.

For achieving the above object, technical method of the present invention comprises the following steps:

(1) initiation parameter is set:

If Base Transmitter pattern count is F, the number of users needing service is M, and the mean vector of emission probability is g _t=(g _{1, t}..., g _k,t..., g _f,t), wherein g _k,trepresent the average of the sending probability Normal Distribution of the kth kind emission mode of base station, initial value the equal Normal Distribution of emission probability;

The standard deviation being located at the iteratively faster stage is σ ₁, be σ in the standard deviation of search stabilization sub stage ₂, and 0 < σ ₂< σ ₁< 0.05;

Practical business requirements set according to user establishes the service quality Q=[Q of user ₁..., Q _m..., Q _m], wherein Q _mrepresent the quality of service requirement of m user, m=1 ..., M;

Iteration ends parameter d=5 are set;

T represents iterations, initialization t=1.

(2) make t=t+1, calculate user's weights:

(2a), after calculating t-1 iteration, the spatial flow that each user gets is:

[S _1,t-1,...,S _m,t-1,...,S _M,t-1]＝[g _1,t-1,...,g _k,t-1,...,g _F,t-1]V,

Wherein S _{m, t-1}represent that the spatial flow that m user obtains according to the emission probability of emission mode each after the t-1 time iteration, V are Base Transmitter mode matrix;

(2b) according to the spatial flow that user after t-1 iteration gets, the service miss rate c of user in the t time iteration is calculated _m,t:

Work as S _{m, t-1}< Q _mtime, $c_{m,t}=1-\frac{S_{m,t-1}}{Q_m};$

Work as S _{m, t-1}>=Q _mtime, c _m,t=0;

(2c) according to the service miss rate of the user obtained, user's weights ω is calculated _m,t,

When $c_{m,t}>\stackrel{\&OverBar;}{c_t}$ Time, ${\mathrm{\&omega;}}_{m,t}=\mathrm{\&alpha;}{2}^{\mathrm{\&beta;}(c_{m,t}-\stackrel{\&OverBar;}{c_t})};$

When $c_{m,t}\&le;\stackrel{\&OverBar;}{c_t}$ Time, ω _m,t=1,

Wherein $\stackrel{\&OverBar;}{c_t}={\mathrm{\&Sigma;}}_{m=1}^M{c}_{m,t}/M$ Represent that each user serves the mean value of miss rate, α represents the importance index meeting QoS of customer, and β represents the importance index meeting equitable proportion, and α, β are positive integer;

(3) according to the normal distribution that sending probability is obeyed random generation N group sample value, calculates the sample value π ' after normalization _{n, k, t}, and according to the sample value π ' after normalization _{n, k, t}calculating sample utility function value H (π ' _n,t):

$H\left({\mathrm{\&pi;}}_{n,t}^{\&prime;}\right)=\underset{m}{\overset{M}{\&Sigma;}}\left\{({\mathrm{\&omega;}}_m-1)\right|Q_m-\underset{k=1}{\overset{F}{\&Sigma;}}{\mathrm{\&pi;}}_{n,k,t}^{\&prime;}{v}_{k,m}|I_{\{Q_m\&GreaterEqual;{\mathrm{\&Sigma;}}_{k=1}^F{\mathrm{\&pi;}}_{n,k,t}^{\&prime;}{v}_{k,m}\}}+Q_m-\underset{k=1}{\overset{F}{\&Sigma;}}{\mathrm{\&pi;}}_{n,k,t}^{\&prime;}{v}_{k,m}\},$

Wherein $I_{\{Q_m\&GreaterEqual;{\mathrm{\&Sigma;}}_{k=1}^F{\mathrm{\&pi;}}_{n,k,t}^{\&prime;}{v}_{k,m}\}}$ For indicative function, when $Q_m\&GreaterEqual;{\mathrm{\&Sigma;}}_{k=1}^F{\mathrm{\&pi;}}_{n,k,t}^{\&prime;}{v}_{k,m}$ Time, $I_{\{Q_m\&GreaterEqual;{\mathrm{\&Sigma;}}_{k=1}^F{\mathrm{\&pi;}}_{n,k,t}^{\&prime;}{v}_{k,m}\}}$ Value is 1, otherwise is 0; v _k,mrepresent that the kth kind emission mode of base station distributes to the spatial flow of m user;

(4) by the utility function value H of all samples (π ' _n,t) ascending arrangement, choose λ value as the t time utility function value γ of iteration at ρ quantile place _t=H _(λ), wherein represent and ρ N is rounded up, 0 < ρ≤0.1;

(5) by normal distribution $N(g_{k,t-1},{\mathrm{\&sigma;}}_1^2)$ Average g _{k, t-1}be updated to: $g_{k,t}=\frac{\underset{n=1}{\overset{N}{\&Sigma;}}{I}_{\{H\left({\mathrm{\&pi;}}_{n,t}^{\&prime;}\right)\&le;{\mathrm{\&gamma;}}_t\}}{\mathrm{\&pi;}}_{n,k,t}^{\&prime;}}{\underset{n=1}{\overset{N}{\&Sigma;}}{I}_{\{H\left({\mathrm{\&pi;}}_{n,t}^{\&prime;}\right)\&le;{\mathrm{\&gamma;}}_t\}}},$ Wherein for indicative function, when H (π ' _n,t)≤γ _ttime, indicative function value is 1, when H (π ' _n,t) > γ _ttime, indicative function value is 0, k=1 ..., F;

(6) according to the average g that (5) upgrade _k,tsituation of change in each iterative process, by normal distribution variance by the iteratively faster stage switch to search stabilization sub stage

(7) finishing iteration process is judged whether according to end condition:

If meet t>=d, and γ _t=γ _t-1=...=γ _t-dthen termination of iterations process, the sampled value that now in the t time iteration, minimum utility function value is corresponding, is the optimum emission probability of the Base Transmitter pattern meeting flow assignment equitable proportion, emission mode is selected according to this optimum emission probability in base station, realizes the equity dispatching of base station spatial flow;

Otherwise, go to step (2) and proceed next iteration, till meeting end condition.

The present invention has the following advantages:

1, the present invention owing to having taken into account QoS of customer and equitable proportion two factors in the flow assignment of base station, makes to be issued to equitable proportion and the maximized good compromise of throughput of system between user in the prerequisite required that guarantees QoS of customer;

2, because the present invention adopts user's weights to characterize the importance of allowable resource Different Optimization target in varied situations, therefore, it is possible to sufficient or deficiency obtains different scheduling result according to allowable resource:

When allowable resource is sufficient, the service miss rate of user is zero, the user's weights obtained by service miss rate are 1, thus make to realize the importance that the maximized importance of throughput of system is greater than equitable proportion between user, ensure that to be issued to throughput of system in the prerequisite of QoS of customer requirement maximum;

When allowable resource is not enough, there is nonzero value in the service miss rate of user, there is by user's weights of obtaining of service miss rate the numerical value being greater than 1, thus make to realize the importance of equitable proportion between user and be greater than the maximized importance of throughput of system, ensure that user obtains fair service;

3, computation complexity of the present invention is low, iterations is few, improves the efficiency of base station resource scheduling.

Accompanying drawing explanation

Fig. 1 is realization flow figure of the present invention;

Fig. 2 is the scheduling result comparison diagram to different service quality demand user in the not enough situation of usable radio resources with the present invention and conventional method;

Fig. 3 is the scheduling fairness comparison diagram to different service quality demand user in the not enough situation of usable radio resources with the present invention and conventional method;

Fig. 4 is the scheduling result comparison diagram to different service quality demand user in the sufficient situation of usable radio resources with the present invention and conventional method;

Fig. 5 is the scheduling fairness comparison diagram to different service quality demand user in the sufficient situation of usable radio resources with the present invention and conventional method.

Embodiment

With reference to Fig. 1, specific implementation step of the present invention is as follows:

Step 1, arranges initiation parameter.

If Base Transmitter pattern count is F, the number of users needing service is M, and the mean vector of emission probability is g _t=(g _{1, t}..., g _k,t..., g _f,t), wherein g _k,trepresent the average of the sending probability Normal Distribution of the kth kind emission mode of base station, initial value the equal Normal Distribution of emission probability;

Standard extent due to normal distribution determines the sample range of emission probability, and when standard deviation is larger, sample range is comparatively large, and iterations is few, but can cause last result inaccuracy; And when standard deviation is less, sample range is less, although can obtain separating comparatively accurately, but iterations can be caused too much, larger standard deviation can be adopted with near iteratively faster to optimum emission probability in the iteratively faster stage accordingly, and then change less standard deviation into, obtain comparatively accurate result.Therefore standard deviation is set to two, the standard deviation being namely located at the iteratively faster stage is σ ₁, be σ in the standard deviation of search stabilization sub stage ₂, and 0 < σ ₂< σ ₁< 0.05;

Practical business requirements set according to user establishes the service quality Q=[Q of user ₁..., Q _m..., Q _m], wherein Q _mrepresent the quality of service requirement of m user, m=1 ..., M;

Iteration ends parameter d=5 are set;

T represents iterations, initialization t=1.

Step 2, makes t=t+1, calculates user's weights.

The spatial flow that after the mean value computation the t-1 time iteration of the normal distribution upgraded by the t-1 time iteration, each user can get, to calculate the user's weights in current iteration with the spatial flow that can get according to the quality of service requirement of user;

(2a), after calculating t-1 iteration, the spatial flow that each user gets is:

[S _1,t-1,...,S _m,t-1,...,S _M,t-1]＝[g _1,t-1,...,g _k,t-1,...,g _F,t-1]V,

Wherein S _{m, t-1}represent that the spatial flow that m user obtains according to the emission probability of emission mode each after the t-1 time iteration, V are Base Transmitter mode matrix;

(2b) according to the spatial flow that user after t-1 iteration gets, the service miss rate c of user in the t time iteration is calculated _m,t:

Work as S _{m, t-1}< Q _mtime,

Work as S _{m, t-1}>=Q _mtime, c _m,t=0;

(2c) according to the service miss rate of the user obtained, user's weights ω is calculated _m,t,

When $c_{m,t}>\stackrel{\&OverBar;}{c_t}$ Time, ${\mathrm{\&omega;}}_{m,t}=\mathrm{\&alpha;}{2}^{\mathrm{\&beta;}(c_{m,t}-\stackrel{\&OverBar;}{c_t})};$

When time, ω _m,t=1,

Wherein represent that each user serves the mean value of miss rate, α represents the importance index meeting QoS of customer, and β represents the importance index meeting equitable proportion, and α, β are positive integer;

In above-mentioned (2a)-(2c) step, when allowable resource is not enough, all QoS of customer requirements can not be met, therefore serve miss rate and there is nonzero value, cause service miss rate different due to the difference of service routine, by the service miss rate of each user and comparing of average service miss rate, obtain corresponding user's weights: when the service miss rate of user is higher than average service miss rate, corresponding user's weights are larger, just can distribute more resource to this user when scheduling of resource; When the service miss rate of user is lower than average service miss rate, corresponding user's weights are less, just can distribute less resource to this user, thus reach the equitable proportion between user when scheduling of resource;

When usable radio resources is sufficient, due to the quality of service requirement of user can be met, so service miss rate is 0, weights are 1, now user's weights characterize user between equitable proportion become by-end, the main target of scheduling of resource becomes the throughput of maximization system, thus it is maximum to realize throughput of system.

Step 3, produces normalization random sample value.

According to the normal distribution that sending probability is obeyed random generation N group sample value, and calculate the sample value π ' after normalization by following formula _{n, k, t}:

${\mathrm{\&pi;}}_{n,k,t}^{\&prime;}=\frac{{\mathrm{\&pi;}}_{n,k,t}}{{\&Sigma;}_{i=1}^F{\mathrm{\&pi;}}_{n,i,t}},$

Wherein π _{n, k, t}represent the emission probability of kth kind emission mode in the t time iteration, n-th group of sample value, k=1 ..., F, n=1 ..., N, N>=100.

Step 4, according to the sample value π ' after normalization _{n, k, t}calculating sample utility function value H (π ' _n,t).

$H\left({\mathrm{\&pi;}}_{n,t}^{\&prime;}\right)=\underset{m}{\overset{M}{\&Sigma;}}\left\{({\mathrm{\&omega;}}_m-1)\right|Q_m-\underset{k=1}{\overset{F}{\&Sigma;}}{\mathrm{\&pi;}}_{n,k,t}^{\&prime;}{v}_{k,m}|I_{\{Q_m\&GreaterEqual;{\mathrm{\&Sigma;}}_{k=1}^F{\mathrm{\&pi;}}_{n,k,t}^{\&prime;}{v}_{k,m}\}}+Q_m-\underset{k=1}{\overset{F}{\&Sigma;}}{\mathrm{\&pi;}}_{n,k,t}^{\&prime;}{v}_{k,m}\},$

Wherein for indicative function, when time, value is 1, otherwise is 0; v _k,mfor the element in Base Transmitter matrix, represent that the kth kind emission mode of base station distributes to the spatial flow of m user.

Step 5, calculates the utility function value at ρ quantile place.

Its performance of sample value that utility function value is less is in the present invention better, and ρ quantile is that sampled value is divided into two-part separation by performance quality.The sample that utility function value is less than ρ quantile place utility function value is the sample that performance is good, and these screening samples is out used for the average that next step upgrades normal distribution; The sample that utility function value is greater than ρ quantile place utility function value is the sample of poor performance, is abandoned.

By the utility function value H of all samples (π ' _n,t) ascending arrangement, choose λ value as the t time utility function value γ of iteration at ρ quantile place _t=H _(λ), wherein represent and ρ N is rounded up, 0 < ρ≤0.1.

Step 6, upgrades the average of normal distribution.

Utility function in N number of sample is utilized to be less than γ _tsample value upgrade emission probability obey normal distribution average g _{k, t-1}, by normal distribution average g _{k, t-1}be updated to:

$g_{k,t}=\frac{\underset{n=1}{\overset{N}{\&Sigma;}}{I}_{\{H\left({\mathrm{\&pi;}}_{n,t}^{\&prime;}\right)\&le;{\mathrm{\&gamma;}}_t\}}{\mathrm{\&pi;}}_{n,k,t}^{\&prime;}}{\underset{n=1}{\overset{N}{\&Sigma;}}{I}_{\{H\left({\mathrm{\&pi;}}_{n,t}^{\&prime;}\right)\&le;{\mathrm{\&gamma;}}_t\}}},$

Wherein for indicative function, when H (π ' _n,t)≤γ _ttime, indicative function value is 1, when H (π ' _n,t) > γ _ttime, indicative function value is 0, k=1 ..., F;

Close to the base station meeting equitable proportion optimum emission probability search gradually according to this average update method.

Step 7, switches the variance yields of different phase normal distribution.

According to the average g upgraded in step 6 _k,tsituation of change in each iterative process, according to following rule by normal distribution variance by the iteratively faster stage switch to search stabilization sub stage

As t=2, with vectorial R=[R ₁..., R _k..., R _f] record mean vector g _tsituation of change, wherein R _kfor recording g _k,tthe mark value of situation of change: if g _k,trelative to its initial value diminish, then make R _k=-1; If g _k,trelative to its initial value become large, then make R _k=1; If g _k,trelative to its initial value do not change, then make R _k=0;

As t > 2, with vectorial P=[P ₁..., P _k..., P _f] record mean vector g _tsituation of change, wherein P _kfor recording g _k,tthe mark value of situation of change: if g _k,trelative to g _{k, t-1}diminish, then make P _k=-1; If g _k,trelative to g _{k, t-1}become large, then make P _k=1; If g _k,trelative to g _{k, t-1}do not change, then make P _k=0;

Calculate Y=P-R, if in the vector element of Y absolute value equal 2 number individual more than F/2, then by normal distribution variance by the iteratively faster stage switch to search stabilization sub stage otherwise, do not switch the variance of normal distribution.

Step 8, judges whether finishing iteration process according to end condition.

If meet t>=d, and γ _t=γ _t-1=...=γ _t-dthen termination of iterations process, the sampled value that now in the t time iteration, minimum utility function value is corresponding, is the optimum emission probability of the Base Transmitter pattern meeting flow assignment equitable proportion, emission mode is selected according to this optimum emission probability in base station, realizes the equity dispatching of base station spatial flow;

Otherwise, go to step 2 and proceed next iteration, till meeting end condition.

The impact of performance of the present invention further illustrates by following emulation:

A, simulated conditions

In simulations, for the down link of the access point AP of a support MU-MIMO in WLAN (wireless local area network), this AP need serve 4 users altogether.Suppose to have 4 kinds of emission modes in this AP, emission mode matrix is

$V=\left[\begin{array}{cccc}0& 4& 0& 4\\ 2& 0& 0& 1\\ 2& 2& 2& 0\\ 1& 2& 4& 2\end{array}\right],$

Simultaneously in order to more objective appraisal distribute to the fair degree of the spatial flow of user, here we adopt the fair index of the Jain reaching the ratio of service quality as the evaluation criterion of simulation result, and Jain is fair, and Index Definition is as follows:

$J(x_1,...,x_m,...,x_M)=\frac{{\left({\&Sigma;}_{m=1}^M{x}_m\right)}^2}{M{\&Sigma;}_{m=1}^M{x}_m^2},$

Wherein [x ₁..., x _m..., x _m] representing scheduling result, the value of Jain fairness index is more close to 1, and it is more fair to represent, equals 1 expression definitely fair.

The hits N=200 of each iteration, quantile ρ=0.1, calculates two positive integer constant α=8 of weights, β=4.

B, emulation content

Emulation 1: when usable radio resources deficiency, the standard deviation choosing the iteratively faster stage is σ ₁=0.01, the standard deviation of search stabilization sub stage is σ ₂=0.005, the spatial flow quality of service requirement number of each user is Q=[1.2,2.2,2.8,3.6], adopt dispatching method of the present invention and conventional method to carry out Resourse Distribute to different service quality demand user, as shown in Figure 2, the scheduling fairness contrast of two kinds of methods as shown in Figure 3 for allocation result.

Emulation 2: when usable radio resources abundance, the standard deviation choosing the iteratively faster stage is σ ₁=0.015, the standard deviation of search stabilization sub stage is σ ₂=0.008, the spatial flow quality of service requirement number of each user is Q=[0.5,1,1.5,2], dispatching method of the present invention and conventional method is adopted to carry out Resourse Distribute to different service quality demand user, as shown in Figure 4, the scheduling fairness contrast of two kinds of methods as shown in Figure 5 for allocation result.

C, analysis of simulation result

The sending probability adopting dispatching method of the present invention to obtain each sending mode of base station in Fig. 2 is π=[0.3653,0.1202,0.0471,0.4674], user 1 to user 4 points the quantity of spatial flow be S=[0.8020,1.5554,1.9638,2.516], the total space fluxion that base station sends is 6.8374; And conventional allocation algorithm is not owing to considering that the probability of each emission mode that service quality obtains is π=[1/3,0,1/3,1/3], the quantity of the spatial flow that each user assigns to is S=[1,2,2,2], and the total space fluxion that base station sends is 7.Visible by the scheduling result contrasting two kinds of methods, when inadequate resource, though the present invention has certain loss of throughput compared to the indiscriminate scheduling mode of conventional method, the user that can require for different service quality provides the service of same services miss rate.

The spatial flow that in Fig. 3, the present invention distributes reaches user 1, and to be respectively 66.83%, 70.7%, 70.14%, 69.89%, Jain justice index to the ratio of user 4 service quality be 0.9995; Conventional proportional fair allocat method reaches user 1, and to be respectively 83.33%, 90.91%, 71.43%, 55.56%, Jain fair index to the ratio of the service quality of user 4 be 0.9695.Visible by the fair index of Jain contrasting two kinds of methods, under out-of-resource condition, the present invention is compared to conventional scheduling method, and scheduling of resource result reaches higher fairness.

The sending probability adopting dispatching method of the present invention to obtain each sending mode of base station in Fig. 4 is π=[0.5774,0.0046,0.0783,0.3398], the quantity of the spatial flow that user 1 assigns to user 4 is S=[0.5056,2.4662,1.5158,2.9938], the total space fluxion that base station sends is 7.4814; And conventional allocation algorithm is not owing to considering that the probability of each emission mode that service quality obtains is π=[1/3,0,1/3,1/3], the quantity of the spatial flow that each user assigns to is S=[1,2,2,2], and the total space fluxion that base station sends is 7.Known when resource abundance by the scheduling result that contrasts two kinds of methods, compared to conventional scheduling method, the present invention can ensure the throughput that improve system under the prerequisite that each QoS of customer requires.

The spatial flow that in Fig. 5, the present invention distributes reaches user 1, and to be respectively 101.12%, 246.62%, 101.05%, 149.69%, Jain justice index to the ratio of the service quality of user 4 be 0.8638; Conventional proportional fair allocat method reaches user 1, and to be respectively 200%, 200%, 133.33%, 100%, Jain fair index to the ratio of the service quality of user 4 be 0.9304.Known by the fair index contrast of Jain contrasting two kinds of methods, when resource abundance, compared to conventional scheduling method, the present invention with less scheduling fairness for cost, can improve the throughput of system.

Can be found out by above simulation result and analysis, in the present invention, when Radio Resource is not enough, meeting equitable proportion is topmost target, and throughput is maximum, is by-end, with less throughput of system for cost obtains higher scheduling fairness; When Radio Resource is sufficient, to improve throughput of system for main target after meeting the quality of service requirement of each user, though the fairness of now dispatching declines to some extent, but increase the throughput of system.

Claims (3)

1., based on the flow assignment proportional fair dispatching method of MU-MIMO, comprise the steps:

(1) initiation parameter is set:

If Base Transmitter pattern count is F, the number of users needing service is M, and the mean vector of emission probability is g _t=(g _{1, t}..., g _k,t..., g _f,t), wherein g _k,trepresent the average of the sending probability Normal Distribution of the kth kind emission mode of base station, initial value the equal Normal Distribution of emission probability;

The standard deviation being located at the iteratively faster stage is σ ₁, be σ in the standard deviation of search stabilization sub stage ₂, and 0 < σ ₂< σ ₁< 0.05;

Practical business requirements set according to user establishes the service quality Q=[Q of user ₁..., Q _m..., Q _m], wherein Q _mrepresent the quality of service requirement of m user, m=1 ..., M;

Iteration ends parameter d=5 are set;

T represents iterations, initialization t=1.

(2) make t=t+1, calculate user's weights:

(2a), after calculating t-1 iteration, the spatial flow that each user gets is:

[S _1,t-1,...,S _m,t-1,...,S _M,t-1]＝[g _1,t-1,...,g _k,t-1,...,g _F,t-1]V,

Wherein S _{m, t-1}represent that the spatial flow that m user obtains according to the emission probability of emission mode each after the t-1 time iteration, V are Base Transmitter mode matrix;

(2b) according to the spatial flow that user after t-1 iteration gets, the service miss rate c of user in the t time iteration is calculated _m,t:

Work as S _{m, t-1}< Q _mtime, $c_{m,t}=1-\frac{S_{m,t-1}}{Q_m};$

Work as S _{m, t-1}>=Q _mtime, c _m,t=0;

(2c) according to the service miss rate of the user obtained, user's weights ω is calculated _m,t,

When time, ${\mathrm{\&omega;}}_{m,t}=\mathrm{\&alpha;}{2}^{\mathrm{\&beta;}(c_{m,t}-\stackrel{\&OverBar;}{c_t})};$

When time, ω _m,t=1,

Wherein represent that each user serves the mean value of miss rate, α represents the importance index meeting QoS of customer, and β represents the importance index meeting equitable proportion, and α, β are positive integer;

(3) according to the normal distribution that sending probability is obeyed random generation N group sample value, calculates the sample value π ' after normalization _{n, k, t}, and according to the sample value π ' after normalization _{n, k, t}calculating sample utility function value H (π ' _n,t):

$H\left({\mathrm{\&pi;}}_{n,t}^{\&prime;}\right)=\underset{m}{\overset{M}{\&Sigma;}}\left\{({\mathrm{\&omega;}}_m-1)\right|Q_m-\underset{k=1}{\overset{F}{\&Sigma;}}{\mathrm{\&pi;}}_{n,k,t}^{\&prime;}{v}_{k,m}|I_{\{Q_m\&GreaterEqual;{\mathrm{\&Sigma;}}_{k=1}^F{\mathrm{\&pi;}}_{n,k,t}^{\&prime;}{v}_{k,m}\}}+Q_m-\underset{k=1}{\overset{F}{\&Sigma;}}{\mathrm{\&pi;}}_{n,k,t}^{\&prime;}{v}_{k,m}\},$

Wherein for indicative function, when time, value is 1, otherwise is 0; v _k,mrepresent that the kth kind emission mode of base station distributes to the spatial flow of m user;

(4) by the utility function value H of all samples (π ' _n,t) ascending arrangement, choose λ value as the t time utility function value γ of iteration at ρ quantile place _t=H _(λ), wherein represent and ρ N is rounded up, 0 < ρ≤0.1;

(5) by normal distribution average g _{k, t-1}be updated to:

Wherein for indicative function, when H (π ' _n,t)≤γ _ttime, indicative function value is 1, when H (π ' _n,t) > γ _ttime, indicative function value is 0, k=1 ..., F;

(6) according to the average g that (5) upgrade _k,tsituation of change in each iterative process, by normal distribution variance by the iteratively faster stage switch to search stabilization sub stage

(7) finishing iteration process is judged whether according to end condition:

If meet t>=d, and γ _t=γ _t-1=...=γ _t-dthen termination of iterations process, the sampled value that now in the t time iteration, minimum utility function value is corresponding, is the optimum emission probability of the Base Transmitter pattern meeting flow assignment equitable proportion, emission mode is selected according to this optimum emission probability in base station, realizes the equity dispatching of base station spatial flow;

Otherwise, go to step (2) and proceed next iteration, till meeting end condition.

2. method according to claim 1, calculates the sample value π ' after normalization in wherein said step (3) _{n, k, t}, by following formulae discovery:

${\mathrm{\&pi;}}_{n,k,t}^{\&prime;}=\frac{{\mathrm{\&pi;}}_{n,k,t}}{{\mathrm{\&Sigma;}}_{i=1}^F{\mathrm{\&pi;}}_{n,i,t}},$

Wherein π _{n, k, t}represent the emission probability of kth kind emission mode in the t time iteration, n-th group of sample value, n=1 ..., N, k=1 ..., F, N>=100.

3. method according to claim 1, wherein said (6) carry out according to the following rules:

As t=2, with vectorial R=[R ₁..., R _k..., R _f] record mean vector g _tsituation of change, wherein R _kfor recording g _k,tthe mark value of situation of change: if g _k,trelative to its initial value diminish, then make R _k=-1; If g _k,trelative to its initial value become large, then make R _k=1; If g _k,trelative to its initial value do not change, then make R _k=0;

As t > 2, with vectorial P=[P ₁..., P _k..., P _f] record mean vector g _tsituation of change, wherein P _kfor recording g _k,tthe mark value of situation of change: if g _k,trelative to g _{k, t-1}diminish, then make P _k=-1; If g _k,trelative to g _{k, t-1}become large, then make P _k=1; If g _k,trelative to g _{k, t-1}do not change, then make P _k=0;

Calculate Y=P-R, if in the vector element of Y absolute value equal 2 number individual more than F/2, then by normal distribution variance by the iteratively faster stage switch to search stabilization sub stage otherwise, do not switch the variance of normal distribution.